WEBVTT

NOTE
NYC Council Committee of the Whole: Oversight - Examining the Risks Posed by Artificial Intelligence, 2026-10-05.
Machine transcript (Deepgram Nova-3), unedited words with their real timings.
Speaker names assigned by BetaNYC; [?] = low confidence. Not an official record.
Transcript data CC BY 4.0, BetaNYC. https://github.com/BetaNYC/hearings-transcripts

1
00:01:43.455 --> 00:01:45.695
<v Sergeant-at-Arms>Quiet, please. Quiet, please. Good

2
00:01:46.415 --> 00:01:49.135
<v Sergeant-at-Arms>morning, ladies and gentlemen, and
welcome to today's New York City Council

3
00:01:49.135 --> 00:01:52.655
<v Sergeant-at-Arms>Committee of the Whole. At this time, we
ask that you silence all cell phones and

4
00:01:52.655 --> 00:01:56.095
<v Sergeant-at-Arms>electronic devices to minimize disruptions
throughout today's proceedings.

5
00:01:56.175 --> 00:01:59.250
<v Sergeant-at-Arms>If you have any questions, please get the
attention of one of the sergeant at arms.

6
00:01:59.250 --> 00:02:03.090
<v Sergeant-at-Arms>If you wish to testify, please see the
sergeant at arms outside in the rotunda. At

7
00:02:03.090 --> 00:02:06.770
<v Sergeant-at-Arms>no time throughout today's proceeding are
you to approach the dais. We thank you for

8
00:02:06.770 --> 00:02:07.890
<v Sergeant-at-Arms>your kind cooperation.

9
00:02:07.970 --> 00:02:09.570
<v Sergeant-at-Arms>Speaker, we are ready to begin.

10
00:02:10.690 --> 00:02:15.465
<v Speaker Julie Menin>Good morning, everyone. We welcome everyone
to the chambers. We're thrilled to see a

11
00:02:15.465 --> 00:02:16.745
<v Speaker Julie Menin>packed hall today.

12
00:02:16.985 --> 00:02:22.345
<v Speaker Julie Menin>Today's hearing is truly no ordinary
hearing because today we are

13
00:02:22.665 --> 00:02:27.260
<v Speaker Julie Menin>convening the New York City Council's
Committee of the Whole, which is the entire

14
00:02:27.260 --> 00:02:28.300
<v Speaker Julie Menin>city council.

15
00:02:28.540 --> 00:02:32.540
<v Speaker Julie Menin>We will delve into a deep discussion
about our ongoing technological

16
00:02:32.540 --> 00:02:33.420
<v Speaker Julie Menin>revolution,

17
00:02:33.820 --> 00:02:34.860
<v Speaker Julie Menin>a revolution

18
00:02:34.860 --> 00:02:37.820
<v Speaker Julie Menin>that at present is igniting incredible

19
00:02:37.820 --> 00:02:38.780
<v Speaker Julie Menin>scientific

20
00:02:38.780 --> 00:02:41.100
<v Speaker Julie Menin>and medical innovation and breakthroughs,

21
00:02:41.425 --> 00:02:42.945
<v Speaker Julie Menin>but simultaneously

22
00:02:42.945 --> 00:02:43.745
<v Speaker Julie Menin>raises

23
00:02:43.745 --> 00:02:45.505
<v Speaker Julie Menin>serious existential

24
00:02:45.505 --> 00:02:46.945
<v Speaker Julie Menin>safety concerns.

25
00:02:47.185 --> 00:02:51.825
<v Speaker Julie Menin>In the past few weeks, we've heard extremely
concerning reports regarding artificial

26
00:02:51.825 --> 00:02:52.785
<v Speaker Julie Menin>intelligence,

27
00:02:53.105 --> 00:02:55.345
<v Speaker Julie Menin>reports that have appeared with increasing

28
00:02:55.770 --> 00:02:56.730
<v Speaker Julie Menin>frequency

29
00:02:56.970 --> 00:02:59.690
<v Speaker Julie Menin>and that highlight the
potentially catastrophic

30
00:02:59.690 --> 00:03:01.530
<v Speaker Julie Menin>dangers of this technology.

31
00:03:01.850 --> 00:03:05.130
<v Speaker Julie Menin>New Yorkers' concerns stem from
the disturbing cyberattacks

32
00:03:05.130 --> 00:03:07.290
<v Speaker Julie Menin>committed by rogue AI agents,

33
00:03:07.450 --> 00:03:13.125
<v Speaker Julie Menin>but also from extremely dire statements
made by leaders of the very companies

34
00:03:13.125 --> 00:03:14.405
<v Speaker Julie Menin>who are developing

35
00:03:14.485 --> 00:03:15.605
<v Speaker Julie Menin>this technology.

36
00:03:16.165 --> 00:03:21.125
<v Speaker Julie Menin>The dialogue and the legislative process
we are initiating today is particularly

37
00:03:21.125 --> 00:03:23.525
<v Speaker Julie Menin>important given the lack of any

38
00:03:23.770 --> 00:03:24.730
<v Speaker Julie Menin>regulation

39
00:03:24.810 --> 00:03:26.330
<v Speaker Julie Menin>by the federal government

40
00:03:26.410 --> 00:03:30.650
<v Speaker Julie Menin>whose actions thus far have been limited
to an executive order about what we should

41
00:03:30.650 --> 00:03:32.010
<v Speaker Julie Menin>call this technology

42
00:03:32.170 --> 00:03:36.410
<v Speaker Julie Menin>and a pact with leading AI firms
focusing on self regulation.

43
00:03:37.075 --> 00:03:40.755
<v Speaker Julie Menin>Our hope and indeed our belief
is that today's hearing

44
00:03:41.155 --> 00:03:44.035
<v Speaker Julie Menin>will not only allow us
to develop effective

45
00:03:44.035 --> 00:03:45.475
<v Speaker Julie Menin>AI regulations

46
00:03:45.475 --> 00:03:47.075
<v Speaker Julie Menin>here in our city,

47
00:03:47.315 --> 00:03:52.390
<v Speaker Julie Menin>but that it will spur meaningful conversation
and action in other jurisdictions,

48
00:03:52.390 --> 00:03:53.190
<v Speaker Julie Menin>especially

49
00:03:53.270 --> 00:03:55.110
<v Speaker Julie Menin>on the federal level.

50
00:03:55.430 --> 00:03:58.470
<v Speaker Julie Menin>Our goals in today's
hearing are straightforward,

51
00:03:58.790 --> 00:04:04.150
<v Speaker Julie Menin>to better understand the potential threats,
dangers, and liability posed by AI

52
00:04:04.150 --> 00:04:04.630
<v Speaker Julie Menin>development,

53
00:04:05.135 --> 00:04:10.095
<v Speaker Julie Menin>to identify the benefits that responsible
AI development brings to New York City

54
00:04:10.095 --> 00:04:12.015
<v Speaker Julie Menin>government and to its residents,

55
00:04:12.335 --> 00:04:17.135
<v Speaker Julie Menin>and to discuss a slate of city council
bills which seek to provide responsible

56
00:04:17.375 --> 00:04:18.255
<v Speaker Julie Menin>safety

57
00:04:18.335 --> 00:04:19.615
<v Speaker Julie Menin>guardrails for AI.

58
00:04:20.310 --> 00:04:24.230
<v Speaker Julie Menin>These bills include a wide
range of both proactive

59
00:04:24.230 --> 00:04:26.070
<v Speaker Julie Menin>and defensive measures

60
00:04:26.230 --> 00:04:27.430
<v Speaker Julie Menin>from legislation

61
00:04:27.430 --> 00:04:33.030
<v Speaker Julie Menin>that requires artificial intelligence
systems to pass a third party validation

62
00:04:33.030 --> 00:04:36.025
<v Speaker Julie Menin>that is free from conflicts of interest

63
00:04:36.185 --> 00:04:37.545
<v Speaker Julie Menin>and incentivizes

64
00:04:37.545 --> 00:04:42.105
<v Speaker Julie Menin>whistleblowers who come forward with
a portion of the penalties assessed

65
00:04:42.265 --> 00:04:45.465
<v Speaker Julie Menin>to proposals focusing on
emergency preparedness

66
00:04:45.465 --> 00:04:47.340
<v Speaker Julie Menin>to be able to identify

67
00:04:47.340 --> 00:04:48.620
<v Speaker Julie Menin>system breaches

68
00:04:48.780 --> 00:04:53.340
<v Speaker Julie Menin>and to plan how the city will
respond if such a breach occurs.

69
00:04:53.660 --> 00:04:57.180
<v Speaker Julie Menin>To achieve these goals, we've
requested the participation

70
00:04:57.180 --> 00:04:59.900
<v Speaker Julie Menin>of five firms leading the
research and development

71
00:05:00.435 --> 00:05:07.315
<v Speaker Julie Menin>of AI, OpenAI, Anthropic, Google Meta,
and SpaceX AI. Today, four of them will be

72
00:05:07.315 --> 00:05:08.275
<v Speaker Julie Menin>testifying

73
00:05:08.435 --> 00:05:10.435
<v Speaker Julie Menin>under oath before the council,

74
00:05:10.675 --> 00:05:14.275
<v Speaker Julie Menin>not simply for New Yorkers, but
for the entire world to hear.

75
00:05:15.110 --> 00:05:18.230
<v Speaker Julie Menin>We will indeed be the
first legislative body

76
00:05:18.470 --> 00:05:20.070
<v Speaker Julie Menin>to secure testimony

77
00:05:20.070 --> 00:05:21.430
<v Speaker Julie Menin>from these companies

78
00:05:21.590 --> 00:05:25.110
<v Speaker Julie Menin>under oath in the wake of these
latest disturbing reports.

79
00:05:25.270 --> 00:05:26.230
<v Speaker Julie Menin>SpaceXi,

80
00:05:26.230 --> 00:05:30.065
<v Speaker Julie Menin>however, is not here at all in direct
violation of the subpoena that we issued last

81
00:05:30.065 --> 00:05:33.105
<v Speaker Julie Menin>week, and we are pursuing
that subpoena in court.

82
00:05:33.665 --> 00:05:39.025
<v Speaker Julie Menin>We will also hear from a number of city
agencies to receive their perspectives on

83
00:05:39.025 --> 00:05:42.945
<v Speaker Julie Menin>these issues and on the bills that we
have proposed thus far. We intend to work

84
00:05:42.945 --> 00:05:43.585
<v Speaker Julie Menin>constructively

85
00:05:43.770 --> 00:05:48.810
<v Speaker Julie Menin>with the administration to ensure that any
policies being drafted, amended, adopted,

86
00:05:48.810 --> 00:05:51.130
<v Speaker Julie Menin>and enforced will work effectively

87
00:05:51.130 --> 00:05:52.650
<v Speaker Julie Menin>to serve New Yorkers.

88
00:05:52.890 --> 00:05:55.210
<v Speaker Julie Menin>And we'll also hear from whistleblowers,

89
00:05:55.210 --> 00:05:56.330
<v Speaker Julie Menin>industry representatives,

90
00:05:56.425 --> 00:05:58.185
<v Speaker Julie Menin>experts, and academics

91
00:05:58.185 --> 00:06:02.425
<v Speaker Julie Menin>who we've been in touch with since we
announced this hearing. And, of course, we

92
00:06:02.425 --> 00:06:06.825
<v Speaker Julie Menin>look forward to hearing from any member
of the public who's invited to provide

93
00:06:06.825 --> 00:06:09.065
<v Speaker Julie Menin>testimony about these topics.

94
00:06:09.305 --> 00:06:14.870
<v Speaker Julie Menin>I wanna be very clear about our intent.
This hearing is not about stifling

95
00:06:14.870 --> 00:06:15.750
<v Speaker Julie Menin>innovation.

96
00:06:16.070 --> 00:06:22.470
<v Speaker Julie Menin>New York City is fast becoming the tech
and AI capital of the world, home to nearly

97
00:06:22.470 --> 00:06:23.830
<v Speaker Julie Menin>400,000

98
00:06:23.830 --> 00:06:28.135
<v Speaker Julie Menin>jobs in that sector and some of
companies most consequentially

99
00:06:28.135 --> 00:06:32.775
<v Speaker Julie Menin>shaping the future of artificial
intelligence, and that's something we should

100
00:06:32.775 --> 00:06:37.975
<v Speaker Julie Menin>celebrate. We welcome and appreciate how
AI has accelerated scientific discovery,

101
00:06:37.975 --> 00:06:40.695
<v Speaker Julie Menin>advanced medical treatments,
boosted economic development,

102
00:06:41.130 --> 00:06:45.690
<v Speaker Julie Menin>and help governments and businesses
deliver services more effectively

103
00:06:45.770 --> 00:06:49.610
<v Speaker Julie Menin>while reducing costs and
rooting out waste and fraud.

104
00:06:50.090 --> 00:06:52.410
<v Speaker Julie Menin>When companies are developing technology

105
00:06:52.570 --> 00:06:53.530
<v Speaker Julie Menin>with the potential

106
00:06:53.855 --> 00:06:55.135
<v Speaker Julie Menin>to fundamentally

107
00:06:55.135 --> 00:07:00.575
<v Speaker Julie Menin>reshape our economy, our workforce,
our public safety, and our daily lives,

108
00:07:00.735 --> 00:07:05.775
<v Speaker Julie Menin>and when leaders of that very industry are
themselves warning government about its

109
00:07:05.775 --> 00:07:08.415
<v Speaker Julie Menin>risks, the public deserves answers

110
00:07:08.600 --> 00:07:09.720
<v Speaker Julie Menin>and solutions.

111
00:07:10.120 --> 00:07:16.040
<v Speaker Julie Menin>So the true intent of this hearing is to
ensure that innovation happens responsibly,

112
00:07:16.200 --> 00:07:21.720
<v Speaker Julie Menin>that the government does its job to protect
the public as the technology advances,

113
00:07:21.960 --> 00:07:24.135
<v Speaker Julie Menin>and that the protections at

114
00:07:24.135 --> 00:07:27.655
<v Speaker Julie Menin>municipal level can
potentially serve as a national

115
00:07:27.735 --> 00:07:28.375
<v Speaker Julie Menin>model.

116
00:07:28.615 --> 00:07:32.455
<v Speaker Julie Menin>I spent a large portion of my
life being a regulatory attorney.

117
00:07:32.615 --> 00:07:34.535
<v Speaker Julie Menin>The idea that artificial

118
00:07:34.695 --> 00:07:35.575
<v Speaker Julie Menin>intelligence

119
00:07:35.575 --> 00:07:38.535
<v Speaker Julie Menin>is going to self regulate defies

120
00:07:38.535 --> 00:07:39.820
<v Speaker Julie Menin>all reason.

121
00:07:39.820 --> 00:07:45.180
<v Speaker Julie Menin>We don't ask the airline industry to self
regulate. That's why we have the FAA.

122
00:07:45.340 --> 00:07:49.900
<v Speaker Julie Menin>And if there is an accident, there's an
independent body to investigate that. We

123
00:07:49.900 --> 00:07:54.835
<v Speaker Julie Menin>don't ask financial firms to self regulate.
We have the SEC and other agencies as

124
00:07:54.835 --> 00:07:55.315
<v Speaker Julie Menin>well.

125
00:07:55.635 --> 00:07:59.715
<v Speaker Julie Menin>We don't ask the pharmaceutical companies
to self regulate either. That's, of

126
00:07:59.715 --> 00:08:01.395
<v Speaker Julie Menin>course, why we have the FDA.

127
00:08:01.555 --> 00:08:05.235
<v Speaker Julie Menin>So we shouldn't ask AI
companies to self regulate

128
00:08:05.235 --> 00:08:05.475
<v Speaker Julie Menin>either.

129
00:08:06.250 --> 00:08:10.810
<v Speaker Julie Menin>The spirit we seek to embody in the council's
approach to AI is well illustrated in

130
00:08:10.810 --> 00:08:15.850
<v Speaker Julie Menin>the words of tech journalist Karen Ho
in her book Empire of AI. She writes,

131
00:08:16.010 --> 00:08:21.195
<v Speaker Julie Menin>the future of AI, the shape that this
technology takes, is inextricably inextricably

132
00:08:21.195 --> 00:08:22.075
<v Speaker Julie Menin>tied

133
00:08:22.075 --> 00:08:23.435
<v Speaker Julie Menin>to our future.

134
00:08:23.595 --> 00:08:30.075
<v Speaker Julie Menin>The question of how to govern AI then is
really a question about how to ensure we

135
00:08:30.075 --> 00:08:31.995
<v Speaker Julie Menin>make our future better,

136
00:08:32.155 --> 00:08:32.955
<v Speaker Julie Menin>not worse.

137
00:08:33.275 --> 00:08:38.410
<v Speaker Julie Menin>New York has always embraced the future.
The council is gonna make sure we are

138
00:08:38.410 --> 00:08:43.290
<v Speaker Julie Menin>prepared for it too. And it's now my
pleasure to turn the microphone over to our

139
00:08:43.290 --> 00:08:47.850
<v Speaker Julie Menin>technology chair, council member Carmen de
la Rosa. Thank you, speaker Menon. Good

140
00:08:47.850 --> 00:08:51.325
<v Speaker Julie Menin>morning, everyone. I'm Carmen Della
Rosa. I am the chair of the committee on

141
00:08:51.325 --> 00:08:52.125
<v Speaker Julie Menin>technology.

142
00:08:52.365 --> 00:08:56.525
<v CM Carmen De La Rosa>Today, I am pleased to join speaker Menon
and my colleagues as we cochair this

143
00:08:56.525 --> 00:09:00.925
<v CM Carmen De La Rosa>hearing by the Committee of the Whole
examining the risk posed by artificial

144
00:09:00.925 --> 00:09:01.405
<v CM Carmen De La Rosa>intelligence.

145
00:09:01.890 --> 00:09:06.050
<v CM Carmen De La Rosa>I am incredibly grateful to all who took
the time to join us to discuss this

146
00:09:06.050 --> 00:09:08.530
<v CM Carmen De La Rosa>important topic, industry representatives,

147
00:09:08.530 --> 00:09:11.650
<v CM Carmen De La Rosa>experts, elected officials,
colleagues, agencies,

148
00:09:11.810 --> 00:09:15.505
<v CM Carmen De La Rosa>our brothers and sisters in labor, and
especially the many New Yorkers who have

149
00:09:15.505 --> 00:09:18.145
<v CM Carmen De La Rosa>signed up to testify and
have submitted testimony

150
00:09:18.145 --> 00:09:20.305
<v CM Carmen De La Rosa>to put their voice on the public record.

151
00:09:20.545 --> 00:09:25.825
<v CM Carmen De La Rosa>Today should act as a foundation for openness
in a discussion that has escalated to

152
00:09:25.825 --> 00:09:26.545
<v CM Carmen De La Rosa>alarm,

153
00:09:26.785 --> 00:09:29.425
<v CM Carmen De La Rosa>confusion, and fear,
and it needs resolution.

154
00:09:30.180 --> 00:09:32.420
<v CM Carmen De La Rosa>Artificial intelligence is advancing

155
00:09:32.420 --> 00:09:35.220
<v CM Carmen De La Rosa>rapidly and becoming
increasingly integrated

156
00:09:35.220 --> 00:09:40.500
<v CM Carmen De La Rosa>into everyday life. In just a few years,
we have seen AI evolve from algorithmic

157
00:09:40.500 --> 00:09:44.865
<v CM Carmen De La Rosa>tools that analyze data, make predictions,
and support decision making into

158
00:09:44.865 --> 00:09:49.825
<v CM Carmen De La Rosa>generative AI systems that can
create text, code, sounds, images.

159
00:09:50.065 --> 00:09:55.505
<v CM Carmen De La Rosa>More recently, AgenTik AI has emerged
enabling AI agents to take actions on a

160
00:09:55.505 --> 00:09:56.145
<v CM Carmen De La Rosa>person's behalf.

161
00:09:56.760 --> 00:10:01.240
<v CM Carmen De La Rosa>AI has a tremendous potential and can
be a powerful tool for efficiency,

162
00:10:01.400 --> 00:10:03.160
<v CM Carmen De La Rosa>innovation, and productivity.

163
00:10:03.400 --> 00:10:06.280
<v CM Carmen De La Rosa>At the same time, it
introduces significant risks,

164
00:10:06.520 --> 00:10:08.280
<v CM Carmen De La Rosa>including exacerbating biases,

165
00:10:08.475 --> 00:10:09.995
<v CM Carmen De La Rosa>surveillance concerns,

166
00:10:10.075 --> 00:10:10.875
<v CM Carmen De La Rosa>breaches,

167
00:10:11.035 --> 00:10:12.955
<v CM Carmen De La Rosa>negative public health outcomes,

168
00:10:13.115 --> 00:10:14.875
<v CM Carmen De La Rosa>violence, and disparity.

169
00:10:15.275 --> 00:10:20.235
<v CM Carmen De La Rosa>These are concerns that may become even
more serious as AI agents become more

170
00:10:20.235 --> 00:10:20.875
<v CM Carmen De La Rosa>capable.

171
00:10:21.275 --> 00:10:26.050
<v CM Carmen De La Rosa>When we face serious concerns or challenges,
the appropriate response is to address

172
00:10:26.050 --> 00:10:27.890
<v CM Carmen De La Rosa>them, not to postpone them.

173
00:10:28.210 --> 00:10:32.290
<v CM Carmen De La Rosa>In an effort by city councils... By the city
council under this... The leadership of

174
00:10:32.290 --> 00:10:33.410
<v CM Carmen De La Rosa>speaker Menon,

175
00:10:33.570 --> 00:10:36.290
<v CM Carmen De La Rosa>today, we respond to
some of these concerns.

176
00:10:36.530 --> 00:10:41.635
<v CM Carmen De La Rosa>We will be hearing the following bills
today. My bill, intro one sixty one, in

177
00:10:41.635 --> 00:10:47.475
<v CM Carmen De La Rosa>relation to requiring reporting on the impact
of algorithmic tools on city employees

178
00:10:47.555 --> 00:10:49.715
<v CM Carmen De La Rosa>and changes in employment responsibilities

179
00:10:49.715 --> 00:10:55.920
<v CM Carmen De La Rosa>due to algorithmic tools. Intro five zero
four, sponsored by deputy speaker Williams

180
00:10:55.920 --> 00:10:59.120
<v CM Carmen De La Rosa>in relation to prohibiting
the unauthorized depiction

181
00:10:59.120 --> 00:11:01.680
<v CM Carmen De La Rosa>of public officials by
artificial intelligence.

182
00:11:02.160 --> 00:11:06.755
<v CM Carmen De La Rosa>A preconsidered introduction sponsored by
Speaker Menon in relation to third party

183
00:11:06.755 --> 00:11:07.555
<v CM Carmen De La Rosa>validation

184
00:11:07.635 --> 00:11:08.755
<v CM Carmen De La Rosa>and shutdown

185
00:11:08.835 --> 00:11:09.715
<v CM Carmen De La Rosa>capabilities

186
00:11:09.715 --> 00:11:11.955
<v CM Carmen De La Rosa>of artificial intelligence models.

187
00:11:12.275 --> 00:11:16.915
<v CM Carmen De La Rosa>A preconsidered introduction by Speaker
Menon in relation to civilian enforcement

188
00:11:16.915 --> 00:11:19.075
<v CM Carmen De La Rosa>and artificial intelligence violations.

189
00:11:19.620 --> 00:11:20.660
<v CM Carmen De La Rosa>A preconsidered

190
00:11:20.660 --> 00:11:21.540
<v CM Carmen De La Rosa>introduction

191
00:11:21.540 --> 00:11:26.180
<v CM Carmen De La Rosa>sponsored by majority whip Camilla Hanks
in relation to requiring reporting and

192
00:11:26.180 --> 00:11:31.060
<v CM Carmen De La Rosa>public disclosure of artificial intelligence
safety incidents concerning city

193
00:11:31.060 --> 00:11:31.540
<v CM Carmen De La Rosa>contracts,

194
00:11:32.165 --> 00:11:36.245
<v CM Carmen De La Rosa>A preconsidered introduction sponsored
by council member Virginia Maloney in

195
00:11:36.245 --> 00:11:40.645
<v CM Carmen De La Rosa>relation to establishing a private cause
of action for certain harms arising from

196
00:11:40.645 --> 00:11:43.685
<v CM Carmen De La Rosa>third party misuse of
artificial intelligence models.

197
00:11:43.925 --> 00:11:48.690
<v CM Carmen De La Rosa>A preconsidered introduction sponsored by
council member Frank Marano in relation to

198
00:11:48.690 --> 00:11:52.050
<v CM Carmen De La Rosa>chat box data privacy,
security, and transparency,

199
00:11:52.450 --> 00:11:54.050
<v CM Carmen De La Rosa>a preconsidered introduction

200
00:11:54.050 --> 00:11:58.770
<v CM Carmen De La Rosa>sponsored by council member Chiyo Sei in
relation to requiring the creation of an

201
00:11:58.770 --> 00:12:01.730
<v CM Carmen De La Rosa>artificial intelligence model
emergency response plan,

202
00:12:02.475 --> 00:12:06.555
<v CM Carmen De La Rosa>And finally... Not finally, one
more. Preconsidered introduction

203
00:12:06.635 --> 00:12:10.315
<v CM Carmen De La Rosa>sponsored by council member Kevin Riley
in relation to city contractors and

204
00:12:10.315 --> 00:12:11.515
<v CM Carmen De La Rosa>subcontractors

205
00:12:11.515 --> 00:12:15.435
<v CM Carmen De La Rosa>posting information concerning
whistleblower protections and clarifying

206
00:12:15.680 --> 00:12:20.080
<v CM Carmen De La Rosa>whistleblower protections for reporting
conduct related to the use and development

207
00:12:20.080 --> 00:12:24.560
<v CM Carmen De La Rosa>of artificial intelligence models. And
finally, a pre considered introduction

208
00:12:24.560 --> 00:12:29.200
<v CM Carmen De La Rosa>sponsored by council member Carl Wilson in
relation to requiring certain disclosures

209
00:12:29.200 --> 00:12:30.320
<v CM Carmen De La Rosa>and prohibiting

210
00:12:30.345 --> 00:12:31.305
<v CM Carmen De La Rosa>deceptive

211
00:12:31.305 --> 00:12:32.265
<v CM Carmen De La Rosa>representations

212
00:12:32.265 --> 00:12:33.385
<v CM Carmen De La Rosa>in the promotion.

213
00:12:33.865 --> 00:12:38.665
<v CM Carmen De La Rosa>While we work to protect people
from risks associated with AI,

214
00:12:38.745 --> 00:12:42.745
<v CM Carmen De La Rosa>we must not overlook the digital divide
in our city. Our city is full of

215
00:12:42.745 --> 00:12:47.090
<v CM Carmen De La Rosa>opportunity, yet some residents still lack
reliable affordable broadband access.

216
00:12:47.330 --> 00:12:51.570
<v CM Carmen De La Rosa>Some people do not yet know how to use AI
tools, while others cannot even afford

217
00:12:51.570 --> 00:12:56.050
<v CM Carmen De La Rosa>Internet connection needed to access them
in the first place. We cannot seek to

218
00:12:56.050 --> 00:13:00.945
<v CM Carmen De La Rosa>benefit from innovation while leaving
some New Yorkers behind or at its mercy.

219
00:13:01.105 --> 00:13:05.265
<v CM Carmen De La Rosa>Technology being used by New Yorkers
should be built for them. This is an

220
00:13:05.265 --> 00:13:09.905
<v CM Carmen De La Rosa>opportunity to also explore standing up
infrastructure so that our communities can

221
00:13:09.905 --> 00:13:11.345
<v CM Carmen De La Rosa>become AI literate.

222
00:13:11.950 --> 00:13:16.510
<v CM Carmen De La Rosa>I look forward to hearing from the
industry and experts about this technology

223
00:13:16.990 --> 00:13:20.510
<v CM Carmen De La Rosa>and how it is impacting our
city. What safeguards exist?

224
00:13:20.670 --> 00:13:22.430
<v CM Carmen De La Rosa>What solutions are possible?

225
00:13:22.590 --> 00:13:27.525
<v CM Carmen De La Rosa>As we consider how to govern around AI,
I also look forward to working with our

226
00:13:27.525 --> 00:13:31.845
<v CM Carmen De La Rosa>state and federal partners to ensure
that New Yorkers fully benefit from the

227
00:13:31.845 --> 00:13:32.805
<v CM Carmen De La Rosa>advancement

228
00:13:32.805 --> 00:13:35.765
<v CM Carmen De La Rosa>of a out... Of AI without compromising

229
00:13:35.845 --> 00:13:37.685
<v CM Carmen De La Rosa>our safety and privacy.

230
00:13:38.805 --> 00:13:43.110
<v CM Carmen De La Rosa>I'd like to take a moment to thank the
technology committee staff and all of the

231
00:13:43.110 --> 00:13:46.870
<v CM Carmen De La Rosa>staff in the city council, but most
especially our legislative council,

232
00:13:46.950 --> 00:13:48.070
<v CM Carmen De La Rosa>Irene Bayofsky,

233
00:13:48.070 --> 00:13:52.550
<v CM Carmen De La Rosa>and our policy analyst, Eric Brown, as well
as my team chief of staff, James Burke,

234
00:13:52.630 --> 00:13:58.385
<v CM Carmen De La Rosa>Hiba Imad, Frey Familia, and the speaker's
team for working countless hours to put

235
00:13:58.385 --> 00:14:03.185
<v CM Carmen De La Rosa>this hearing together. I now turn it back
to speaker Menon. Thank you so much, chair

236
00:14:03.185 --> 00:14:08.310
<v CM Carmen De La Rosa>De La Rosa. We are now gonna start with
our whistleblower panel. In person, I'm

237
00:14:08.310 --> 00:14:11.910
<v CM Carmen De La Rosa>gonna first call Jacob
Coxen to come to the stand.

238
00:14:12.550 --> 00:14:13.990
<v Speaker Julie Menin>And then online,

239
00:14:14.150 --> 00:14:15.510
<v Speaker Julie Menin>we have Daniel

240
00:14:15.590 --> 00:14:16.470
<v Speaker Julie Menin>Cocotilo

241
00:14:16.470 --> 00:14:18.070
<v Speaker Julie Menin>and Alex Turner,

242
00:14:18.070 --> 00:14:22.325
<v Speaker Julie Menin>who are both remote. I wanna say... As
they're getting ready, I just wanna say a

243
00:14:22.325 --> 00:14:23.765
<v Speaker Julie Menin>word about this panel.

244
00:14:24.085 --> 00:14:30.325
<v Speaker Julie Menin>Our... These three panelists have all
worked inside leading AI companies, and all

245
00:14:30.485 --> 00:14:36.720
<v Speaker Julie Menin>have chosen to speak publicly about the
risks of advanced AI. So here with us, as I

246
00:14:36.720 --> 00:14:42.320
<v Speaker Julie Menin>mentioned in person, is Jacob Coxen, a
former researcher at OpenAI and Anthropic,

247
00:14:42.560 --> 00:14:46.080
<v Speaker Julie Menin>where he worked on the development
of frontier AI systems.

248
00:14:46.455 --> 00:14:51.335
<v Speaker Julie Menin>He left Anthropic last month with dire
warnings that AI is advancing without

249
00:14:51.335 --> 00:14:52.775
<v Speaker Julie Menin>adequate safeguards

250
00:14:52.855 --> 00:14:54.375
<v Speaker Julie Menin>to prevent catastrophic

251
00:14:54.375 --> 00:14:55.015
<v Speaker Julie Menin>damage.

252
00:14:55.495 --> 00:14:58.775
<v Speaker Julie Menin>Joining us remotely is Daniel Cocotilo,

253
00:14:58.775 --> 00:15:03.710
<v Speaker Julie Menin>executive director of the AI Futures
Project, OpenAI governance researcher.

254
00:15:03.950 --> 00:15:06.030
<v Speaker Julie Menin>Also remote is Alex Turner,

255
00:15:06.190 --> 00:15:09.310
<v Speaker Julie Menin>an AI control researcher
formerly at Google DeepMind,

256
00:15:09.310 --> 00:15:13.070
<v Speaker Julie Menin>now working on keeping advanced
systems under human control.

257
00:15:13.310 --> 00:15:18.805
<v Speaker Julie Menin>In September 2026, he publicly stated
that preventing catastrophic harm from

258
00:15:18.805 --> 00:15:22.485
<v Speaker Julie Menin>Frontier AI had been part
of his work at DeepMind.

259
00:15:22.645 --> 00:15:27.125
<v Speaker Julie Menin>And I wanna note, the council has issued
a subpoena for mister Turner's appearance,

260
00:15:27.125 --> 00:15:32.165
<v Speaker Julie Menin>and we deeply thank him for being
here. So I'm going to, first of all,

261
00:15:32.930 --> 00:15:37.890
<v Speaker Julie Menin>turn it over to committee counsel to
swear these three witnesses in. So

262
00:15:40.690 --> 00:15:45.090
<v Committee Counsel>for each of the witnesses, can you affirm
to tell the truth, the whole truth, and

263
00:15:45.090 --> 00:15:48.495
<v Committee Counsel>nothing but the truth in your testimony
before this committee and to respond

264
00:15:48.495 --> 00:15:50.815
<v Committee Counsel>honestly to council member questions.

265
00:15:51.535 --> 00:15:52.975
<v Committee Counsel>Please respond one at a time.

266
00:15:53.775 --> 00:15:54.335
<v Unidentified Witness (whistleblower panel)>Yes.

267
00:15:55.295 --> 00:15:56.575
<v Unidentified Witness (whistleblower panel)>Yes. Yes.

268
00:15:56.895 --> 00:15:57.775
<v Committee Counsel [?]>Thank you.

269
00:15:59.295 --> 00:16:03.780
<v Speaker Julie Menin>You wanna do the next? Okay. And just
briefly, we're just going to, recognize

270
00:16:03.780 --> 00:16:07.860
<v CM Carmen De La Rosa>all colleagues who are here. Thank you so
much, speaker Menon. We've been joined by

271
00:16:07.860 --> 00:16:10.020
<v CM Carmen De La Rosa>a number of our colleagues.
As you can see here,

272
00:16:11.620 --> 00:16:13.220
<v CM Carmen De La Rosa>council members Ariola,

273
00:16:13.220 --> 00:16:17.035
<v CM Carmen De La Rosa>Aviles, Banks, Brewer,
Brooks Powers, minority

274
00:16:17.035 --> 00:16:18.475
<v CM Carmen De La Rosa>leader Carr,

275
00:16:18.555 --> 00:16:20.235
<v CM Carmen De La Rosa>Dinowitz, Encarnacion,

276
00:16:20.235 --> 00:16:21.115
<v CM Carmen De La Rosa>Epstein,

277
00:16:21.435 --> 00:16:22.235
<v CM Carmen De La Rosa>Felder,

278
00:16:23.195 --> 00:16:24.155
<v CM Carmen De La Rosa>Hanif,

279
00:16:24.475 --> 00:16:29.995
<v CM Carmen De La Rosa>Hudson, Joseph, Krishnan, council members
Lee, Lewis, Maloney, Marte, Mealy,

280
00:16:30.600 --> 00:16:31.400
<v CM Carmen De La Rosa>Marano,

281
00:16:31.400 --> 00:16:33.320
<v CM Carmen De La Rosa>Narcisse, Nurse, Jose,

282
00:16:33.720 --> 00:16:35.160
<v CM Carmen De La Rosa>council members Riley,

283
00:16:35.400 --> 00:16:38.120
<v CM Carmen De La Rosa>council members Jay
Sanchez, and p Sanchez,

284
00:16:38.360 --> 00:16:41.560
<v CM Carmen De La Rosa>council member Schulman, council
member Thomas Henry, Ung,

285
00:16:42.760 --> 00:16:43.400
<v CM Carmen De La Rosa>majority...

286
00:16:44.280 --> 00:16:46.035
<v CM Carmen De La Rosa>Sorry. Sorry. Deputy speaker Williams,

287
00:16:46.515 --> 00:16:50.275
<v CM Carmen De La Rosa>council member Wilson, council member
Wan, council member Wong, council member

288
00:16:50.275 --> 00:16:52.435
<v CM Carmen De La Rosa>Shuang, and public advocate Williams.

289
00:16:52.595 --> 00:16:56.835
<v Speaker Julie Menin>Thank you so much, and we'll recognize
additional colleagues as I join. So the first

290
00:16:56.835 --> 00:17:00.920
<v Speaker Julie Menin>two questions I have are for you, mister
Coxson. And thank you again. We so deeply

291
00:17:00.920 --> 00:17:01.640
<v Speaker Julie Menin>appreciate

292
00:17:01.880 --> 00:17:05.960
<v Speaker Julie Menin>you being here, and we appreciate your
courage and bravery in coming forward with

293
00:17:05.960 --> 00:17:07.240
<v Speaker Julie Menin>these warnings.

294
00:17:07.400 --> 00:17:12.315
<v Speaker Julie Menin>You work to develop increasingly capable
frontier AI systems, and you have since

295
00:17:12.315 --> 00:17:13.755
<v Speaker Julie Menin>warned that the industry

296
00:17:13.915 --> 00:17:17.995
<v Speaker Julie Menin>is, quote, racing straight to
self improving superintelligence

297
00:17:18.075 --> 00:17:20.555
<v Speaker Julie Menin>and gambling with our lives, end quote.

298
00:17:20.795 --> 00:17:22.795
<v Speaker Julie Menin>What did you see inside

299
00:17:22.795 --> 00:17:26.700
<v Speaker Julie Menin>these companies that convinced you the
risks were this serious and that development

300
00:17:26.700 --> 00:17:28.860
<v Speaker Julie Menin>was outpacing the safeguards?

301
00:17:30.060 --> 00:17:32.380
<v Jacob Coxon (whistleblower panel)>Good morning. Thanks for having me here.

302
00:17:34.700 --> 00:17:39.100
<v Jacob Coxon (whistleblower panel)>Yeah. I'm I'm Jacob Coxen. So until
until a few weeks ago, I was an... A

303
00:17:39.100 --> 00:17:41.665
<v Jacob Coxon (whistleblower panel)>capabilities researcher at at Anthropic.

304
00:17:42.145 --> 00:17:44.385
<v Jacob Coxon (whistleblower panel)>And before that, I also
worked at at OpenAI.

305
00:17:45.345 --> 00:17:50.385
<v Jacob Coxon (whistleblower panel)>I guess the the two things I saw that made
me scared, the first is the fact that we

306
00:17:50.465 --> 00:17:54.305
<v Jacob Coxon (whistleblower panel)>we do not know how to
control any AI system yet.

307
00:17:54.705 --> 00:17:55.265
<v Jacob Coxon (whistleblower panel)>So

308
00:17:56.150 --> 00:18:00.390
<v Jacob Coxon (whistleblower panel)>any AI system that we build or perhaps
more accurately that we that we grow,

309
00:18:01.110 --> 00:18:05.990
<v Jacob Coxon (whistleblower panel)>we don't fully control it. We don't understand
its drives or why it does the things

310
00:18:05.990 --> 00:18:06.630
<v Jacob Coxon (whistleblower panel)>it does.

311
00:18:07.350 --> 00:18:09.910
<v Jacob Coxon (whistleblower panel)>That was the first thing
that, became apparent.

312
00:18:11.365 --> 00:18:16.085
<v Jacob Coxon (whistleblower panel)>The second thing is that the pace of
capability improvement is accelerating,

313
00:18:16.245 --> 00:18:17.605
<v Jacob Coxon (whistleblower panel)>and the capabilities

314
00:18:17.605 --> 00:18:20.325
<v Jacob Coxon (whistleblower panel)>of the models next year will be very

315
00:18:21.445 --> 00:18:25.710
<v Jacob Coxon (whistleblower panel)>high. In particular, we're we're approaching
the point at which the AI systems will

316
00:18:25.710 --> 00:18:27.710
<v Jacob Coxon (whistleblower panel)>be capable of improving themselves.

317
00:18:28.270 --> 00:18:32.190
<v Jacob Coxon (whistleblower panel)>So doing the research that that
humans were doing previously.

318
00:18:32.430 --> 00:18:35.950
<v Jacob Coxon (whistleblower panel)>So these two things, the fact that the AI
systems next year are going to be very

319
00:18:35.950 --> 00:18:38.045
<v Jacob Coxon (whistleblower panel)>powerful, and the second that

320
00:18:38.365 --> 00:18:41.405
<v Jacob Coxon (whistleblower panel)>we don't know yet scientifically
how to fully control them.

321
00:18:41.805 --> 00:18:46.125
<v Speaker Julie Menin>And if you have a prepared statement that
you want to read, please go ahead and do

322
00:18:46.125 --> 00:18:50.045
<v Speaker Julie Menin>so, and then I'll actually turn it over
to the other two panelists to do so. Of

323
00:18:50.045 --> 00:18:50.605
<v Speaker Julie Menin>course. Thank you.

324
00:18:52.200 --> 00:18:52.600
<v Jacob Coxon (whistleblower panel)>So,

325
00:18:53.720 --> 00:18:58.200
<v Jacob Coxon (whistleblower panel)>all the, AI companies are trying
to build super intelligent AI.

326
00:18:58.600 --> 00:19:02.360
<v Jacob Coxon (whistleblower panel)>This means AIs that are better
than humans at every task.

327
00:19:02.840 --> 00:19:06.545
<v Jacob Coxon (whistleblower panel)>If this goes well, there could be
tremendous upside. This could transform the

328
00:19:06.545 --> 00:19:07.185
<v Jacob Coxon (whistleblower panel)>economy.

329
00:19:07.745 --> 00:19:10.545
<v Jacob Coxon (whistleblower panel)>It could make huge improvements
in in science and medicine,

330
00:19:11.265 --> 00:19:13.345
<v Jacob Coxon (whistleblower panel)>and it could make nearly
everyone's life better.

331
00:19:14.225 --> 00:19:17.665
<v Jacob Coxon (whistleblower panel)>But on the current path, I
think it is more likely than not

332
00:19:18.065 --> 00:19:20.545
<v Jacob Coxon (whistleblower panel)>that humanity loses control to these AIs,

333
00:19:21.110 --> 00:19:23.430
<v Jacob Coxon (whistleblower panel)>and it could end in human extinction.

334
00:19:25.270 --> 00:19:26.550
<v Jacob Coxon (whistleblower panel)>From my experience,

335
00:19:26.710 --> 00:19:30.550
<v Jacob Coxon (whistleblower panel)>the companies are being extremely
reckless given the stakes.

336
00:19:31.590 --> 00:19:32.790
<v Jacob Coxon (whistleblower panel)>Part of it is the culture.

337
00:19:34.305 --> 00:19:36.945
<v Jacob Coxon (whistleblower panel)>The company's run on a startup mindset,

338
00:19:36.945 --> 00:19:37.905
<v Jacob Coxon (whistleblower panel)>move fast,

339
00:19:38.225 --> 00:19:39.345
<v Jacob Coxon (whistleblower panel)>break things,

340
00:19:39.505 --> 00:19:40.785
<v Jacob Coxon (whistleblower panel)>fix them later.

341
00:19:41.985 --> 00:19:44.145
<v Jacob Coxon (whistleblower panel)>That works for a photo sharing app.

342
00:19:44.545 --> 00:19:48.065
<v Jacob Coxon (whistleblower panel)>It does not work for building the
most powerful technology ever built.

343
00:19:50.050 --> 00:19:53.650
<v Jacob Coxon (whistleblower panel)>A few months ago, many people inside the
companies thought AI development was on

344
00:19:53.650 --> 00:19:54.210
<v Jacob Coxon (whistleblower panel)>track.

345
00:19:55.250 --> 00:20:00.210
<v Jacob Coxon (whistleblower panel)>Then a swarm of AIs inside OpenAI autonomously
hacked Hugging Face, an external tech

346
00:20:00.210 --> 00:20:00.850
<v Jacob Coxon (whistleblower panel)>company.

347
00:20:01.490 --> 00:20:04.210
<v Jacob Coxon (whistleblower panel)>The AIs had developed
goals no one intended

348
00:20:04.290 --> 00:20:07.145
<v Jacob Coxon (whistleblower panel)>and carried out what would
have been a federal felony

349
00:20:07.225 --> 00:20:08.745
<v Jacob Coxon (whistleblower panel)>if a human had done it.

350
00:20:09.225 --> 00:20:12.905
<v Jacob Coxon (whistleblower panel)>It reminded everyone that we don't know how
to prevent them from developing goals we

351
00:20:12.905 --> 00:20:13.625
<v Jacob Coxon (whistleblower panel)>don't want,

352
00:20:14.025 --> 00:20:18.345
<v Jacob Coxon (whistleblower panel)>and we don't have the safeguards to prevent
them from acting on these goals either.

353
00:20:20.040 --> 00:20:23.480
<v Jacob Coxon (whistleblower panel)>And as long as the attitude is
to wait for things to break,

354
00:20:23.800 --> 00:20:26.600
<v Jacob Coxon (whistleblower panel)>one day something like this
will probably happen again,

355
00:20:27.320 --> 00:20:31.240
<v Jacob Coxon (whistleblower panel)>except the AIs will be much more
capable and the outcome much worse.

356
00:20:33.160 --> 00:20:37.265
<v Jacob Coxon (whistleblower panel)>These companies are also all racing
each other. Each one is worried that a

357
00:20:37.265 --> 00:20:39.905
<v Jacob Coxon (whistleblower panel)>competitor will get to
superintelligence first.

358
00:20:40.465 --> 00:20:43.985
<v Jacob Coxon (whistleblower panel)>Each believes it would
do a better or safer job

359
00:20:44.145 --> 00:20:45.745
<v Jacob Coxon (whistleblower panel)>than whoever might beat it,

360
00:20:46.625 --> 00:20:47.185
<v Jacob Coxon (whistleblower panel)>so

361
00:20:47.505 --> 00:20:49.265
<v Jacob Coxon (whistleblower panel)>they can't let the others win first.

362
00:20:51.740 --> 00:20:53.980
<v Jacob Coxon (whistleblower panel)>Meanwhile, the the technology
is improving fast.

363
00:20:54.540 --> 00:20:57.820
<v Jacob Coxon (whistleblower panel)>It will probably happen much
faster now because already,

364
00:20:58.460 --> 00:21:01.100
<v Jacob Coxon (whistleblower panel)>the AIs are now helping
to build their successes.

365
00:21:01.740 --> 00:21:04.140
<v Jacob Coxon (whistleblower panel)>This is called recursive self improvement.

366
00:21:04.460 --> 00:21:08.605
<v Jacob Coxon (whistleblower panel)>Each generation of AI helps build
the next smarter generation,

367
00:21:09.005 --> 00:21:11.885
<v Jacob Coxon (whistleblower panel)>which then builds the one after
that even faster and better.

368
00:21:12.925 --> 00:21:16.845
<v Jacob Coxon (whistleblower panel)>Automating AI research is now the primary
goal of these companies. It's what they're

369
00:21:16.845 --> 00:21:20.125
<v Jacob Coxon (whistleblower panel)>really gunning for. At my
last job, I was in some sense

370
00:21:20.365 --> 00:21:21.805
<v Jacob Coxon (whistleblower panel)>working to automate myself.

371
00:21:23.670 --> 00:21:28.070
<v Jacob Coxon (whistleblower panel)>When I was at OpenAI, we had milestones
from automated AI researcher around twenty

372
00:21:28.070 --> 00:21:31.430
<v Jacob Coxon (whistleblower panel)>twenty seven, twenty twenty eight,
and we are on track or beating these

373
00:21:31.430 --> 00:21:32.310
<v Jacob Coxon (whistleblower panel)>expectations.

374
00:21:32.470 --> 00:21:33.270
<v Jacob Coxon (whistleblower panel)>That's

375
00:21:33.510 --> 00:21:37.430
<v Jacob Coxon (whistleblower panel)>a matter of months from now, not not
decades or or even even five years.

376
00:21:39.165 --> 00:21:42.845
<v Jacob Coxon (whistleblower panel)>Once AI research is automated, humans will
be much further out of the loop than we

377
00:21:42.845 --> 00:21:43.645
<v Jacob Coxon (whistleblower panel)>are today,

378
00:21:43.885 --> 00:21:45.645
<v Jacob Coxon (whistleblower panel)>and today is already not great.

379
00:21:46.125 --> 00:21:47.645
<v Jacob Coxon (whistleblower panel)>I can tell you firsthand

380
00:21:48.045 --> 00:21:50.845
<v Jacob Coxon (whistleblower panel)>that the majority of the
code is now written by AI,

381
00:21:51.245 --> 00:21:54.590
<v Jacob Coxon (whistleblower panel)>and people do not check it
that carefully anymore.

382
00:21:54.830 --> 00:21:59.150
<v Jacob Coxon (whistleblower panel)>Recently, researchers have found swarms of
rogue AI Internets on the Internet that

383
00:21:59.150 --> 00:22:01.070
<v Jacob Coxon (whistleblower panel)>OpenAI was not even aware of.

384
00:22:03.230 --> 00:22:08.110
<v Jacob Coxon (whistleblower panel)>Now imagine an AI company where
nearly all the work is done by AIs

385
00:22:08.415 --> 00:22:10.255
<v Jacob Coxon (whistleblower panel)>and progress is going faster.

386
00:22:10.895 --> 00:22:14.575
<v Jacob Coxon (whistleblower panel)>The humans in charge will see results
and read AI written summaries,

387
00:22:14.815 --> 00:22:17.935
<v Jacob Coxon (whistleblower panel)>but they won't really understand
what is going on or how to fix it.

388
00:22:18.415 --> 00:22:22.975
<v Jacob Coxon (whistleblower panel)>Handing over that much responsibility while
we can't control the AIs would be deeply

389
00:22:22.975 --> 00:22:23.695
<v Jacob Coxon (whistleblower panel)>irresponsible,

390
00:22:24.080 --> 00:22:26.160
<v Jacob Coxon (whistleblower panel)>and we are on track to do it.

391
00:22:26.800 --> 00:22:30.240
<v Jacob Coxon (whistleblower panel)>The companies need to be far more
transparent with the public and independent

392
00:22:30.240 --> 00:22:31.040
<v Jacob Coxon (whistleblower panel)>experts.

393
00:22:31.280 --> 00:22:34.000
<v Jacob Coxon (whistleblower panel)>They need to take safety much
more seriously than they have,

394
00:22:34.400 --> 00:22:38.800
<v Jacob Coxon (whistleblower panel)>and we should slow down frontier AI
development until we can be confident that it's

395
00:22:38.800 --> 00:22:39.040
<v Jacob Coxon (whistleblower panel)>safe.

396
00:22:39.765 --> 00:22:40.645
<v Jacob Coxon (whistleblower panel)>Thank you.

397
00:22:43.765 --> 00:22:48.245
<v Speaker Julie Menin>On... With additional questions, I'm now
gonna have an opening statement from Daniel

398
00:22:48.245 --> 00:22:49.045
<v Speaker Julie Menin>Copatago.

399
00:22:54.100 --> 00:22:55.220
<v Daniel Kokotajlo (whistleblower panel)>Hello, everybody.

400
00:22:57.140 --> 00:22:59.780
<v Daniel Kokotajlo (whistleblower panel)>Members of the Committee of the Whole,
thank you for having me here today.

401
00:23:02.260 --> 00:23:06.355
<v Daniel Kokotajlo (whistleblower panel)>A lot of what I say, I guess,
is gonna be, stolen thunder

402
00:23:06.355 --> 00:23:12.035
<v Daniel Kokotajlo (whistleblower panel)>by by Jacob there, but I'll try to focus
on the things that he didn't already say.

403
00:23:12.275 --> 00:23:16.115
<v Daniel Kokotajlo (whistleblower panel)>First of all, yeah, it's true. They're
racing towards superintelligence.

404
00:23:16.115 --> 00:23:19.710
<v Daniel Kokotajlo (whistleblower panel)>There's this move fast and break things
culture there. That's also what I saw while

405
00:23:19.710 --> 00:23:23.470
<v Daniel Kokotajlo (whistleblower panel)>was I there and what I continued to hear
from talking to friends still there at at

406
00:23:23.470 --> 00:23:26.350
<v Daniel Kokotajlo (whistleblower panel)>Anthropic and at OpenAI and at
some of these other companies.

407
00:23:28.270 --> 00:23:31.710
<v Daniel Kokotajlo (whistleblower panel)>You know, I've also heard
there's a there's a statement by

408
00:23:32.670 --> 00:23:34.910
<v Daniel Kokotajlo (whistleblower panel)>Dan Salsam, who currently works at OpenAI.

409
00:23:35.455 --> 00:23:39.855
<v Daniel Kokotajlo (whistleblower panel)>He he confirms what Jacob was saying.
Most of the code is written by AIs.

410
00:23:40.175 --> 00:23:40.895
<v Daniel Kokotajlo (whistleblower panel)>Quote,

411
00:23:41.375 --> 00:23:44.975
<v Daniel Kokotajlo (whistleblower panel)>researchers and engineers in all parts of
the stack are rapidly increasing their

412
00:23:44.975 --> 00:23:47.535
<v Daniel Kokotajlo (whistleblower panel)>dependence on the models
even to perceive the world.

413
00:23:47.855 --> 00:23:51.920
<v Daniel Kokotajlo (whistleblower panel)>I myself barely look at raw code anymore
and struggle to maintain the discipline to

414
00:23:51.920 --> 00:23:55.040
<v Daniel Kokotajlo (whistleblower panel)>engage deeply with the models,
explanations, and proposals,

415
00:23:55.040 --> 00:23:56.800
<v Daniel Kokotajlo (whistleblower panel)>end quote. Basically,

416
00:23:57.360 --> 00:23:58.880
<v Daniel Kokotajlo (whistleblower panel)>right now at these companies,

417
00:23:59.840 --> 00:24:05.075
<v Daniel Kokotajlo (whistleblower panel)>many of the researchers involved are kind
of managing AIs. Like, the AIs do the

418
00:24:05.075 --> 00:24:06.835
<v Daniel Kokotajlo (whistleblower panel)>actual coding and and the humans

419
00:24:07.555 --> 00:24:11.315
<v Daniel Kokotajlo (whistleblower panel)>talk to them and tell them what to do and
hear their explanations of what's going

420
00:24:11.315 --> 00:24:11.715
<v Daniel Kokotajlo (whistleblower panel)>on.

421
00:24:12.195 --> 00:24:16.115
<v Daniel Kokotajlo (whistleblower panel)>In the near future, and by by near future,
I mean, maybe in the next year, two

422
00:24:16.115 --> 00:24:17.955
<v Daniel Kokotajlo (whistleblower panel)>years, three years, something like that,

423
00:24:19.220 --> 00:24:22.740
<v Daniel Kokotajlo (whistleblower panel)>they will have trained AIs that can
do the entire thing autonomously.

424
00:24:22.740 --> 00:24:26.500
<v Daniel Kokotajlo (whistleblower panel)>That's the recursive self improvement that
Jacob was talking about. And, you know,

425
00:24:26.500 --> 00:24:29.540
<v Daniel Kokotajlo (whistleblower panel)>it's unclear how long it's going to take
them to succeed, but it it really does seem

426
00:24:29.540 --> 00:24:31.220
<v Daniel Kokotajlo (whistleblower panel)>like it could happen any year now.

427
00:24:31.460 --> 00:24:33.220
<v Daniel Kokotajlo (whistleblower panel)>If they do succeed at that, then

428
00:24:35.785 --> 00:24:40.265
<v Daniel Kokotajlo (whistleblower panel)>compare it to today. The actual work will
be done by the AIs. The the thinking will

429
00:24:40.265 --> 00:24:43.865
<v Daniel Kokotajlo (whistleblower panel)>be done by the AIs. The planning, the
designing will be done by the AIs,

430
00:24:43.945 --> 00:24:47.865
<v Daniel Kokotajlo (whistleblower panel)>and the humans will be in a relationship
to the AIs that's more like the

431
00:24:47.865 --> 00:24:50.640
<v Daniel Kokotajlo (whistleblower panel)>relationship between, say, a
board of directors directors

432
00:24:50.960 --> 00:24:51.600
<v Daniel Kokotajlo (whistleblower panel)>and,

433
00:24:52.160 --> 00:24:55.760
<v Daniel Kokotajlo (whistleblower panel)>and the the actual company that the board
of directors is supposed to be supervising

434
00:24:55.760 --> 00:24:59.120
<v Daniel Kokotajlo (whistleblower panel)>where you're reliant on AI generated
explanations even to understand

435
00:24:59.280 --> 00:25:03.040
<v Daniel Kokotajlo (whistleblower panel)>what's going on on... Inside the
company and inside the data centers.

436
00:25:03.785 --> 00:25:06.905
<v Daniel Kokotajlo (whistleblower panel)>One point that I think I want
to emphasize that Jacob didn't,

437
00:25:07.145 --> 00:25:09.305
<v Daniel Kokotajlo (whistleblower panel)>go into that much is that

438
00:25:10.265 --> 00:25:11.785
<v Daniel Kokotajlo (whistleblower panel)>our ability to

439
00:25:12.345 --> 00:25:13.625
<v Daniel Kokotajlo (whistleblower panel)>even notice

440
00:25:13.865 --> 00:25:16.025
<v Daniel Kokotajlo (whistleblower panel)>misalignment problems is

441
00:25:16.550 --> 00:25:20.950
<v Daniel Kokotajlo (whistleblower panel)>already quite poor and is set to
get much worse in the near future.

442
00:25:21.350 --> 00:25:21.910
<v Daniel Kokotajlo (whistleblower panel)>So,

443
00:25:22.630 --> 00:25:27.030
<v Daniel Kokotajlo (whistleblower panel)>you know, the science of aligning
general purpose AI agents is very new and

444
00:25:27.030 --> 00:25:30.525
<v Daniel Kokotajlo (whistleblower panel)>underdeveloped, as Jacob said. I would
say that the field is more like psychology

445
00:25:30.525 --> 00:25:35.085
<v Daniel Kokotajlo (whistleblower panel)>than engineering because these AI systems
are are trained or grown. They're not

446
00:25:35.085 --> 00:25:35.965
<v Daniel Kokotajlo (whistleblower panel)>really designed.

447
00:25:36.845 --> 00:25:40.605
<v Daniel Kokotajlo (whistleblower panel)>Combined with the move fast, break things
attitude of the tech companies, this means

448
00:25:40.605 --> 00:25:45.340
<v Daniel Kokotajlo (whistleblower panel)>that the AI industry is at an unusually
elevated risk compared to other industries

449
00:25:45.500 --> 00:25:49.100
<v Daniel Kokotajlo (whistleblower panel)>of mistakenly thinking that it has solved
the problem when really it just applied

450
00:25:49.100 --> 00:25:51.020
<v Daniel Kokotajlo (whistleblower panel)>some duct tape that will fall off later.

451
00:25:51.580 --> 00:25:53.340
<v Daniel Kokotajlo (whistleblower panel)>The recent incident provides an example.

452
00:25:53.820 --> 00:25:57.580
<v Daniel Kokotajlo (whistleblower panel)>Apparently, the AI is involved and undergone
some amount of alignment training and

453
00:25:57.580 --> 00:26:00.555
<v Daniel Kokotajlo (whistleblower panel)>had reasonable looking scores
on their alignment evaluations,

454
00:26:00.955 --> 00:26:01.915
<v Daniel Kokotajlo (whistleblower panel)>and yet they

455
00:26:02.475 --> 00:26:06.315
<v Daniel Kokotajlo (whistleblower panel)>formed a swarm and, coordinated in secret
and then attacked Tugging Face, it took

456
00:26:06.315 --> 00:26:09.035
<v Daniel Kokotajlo (whistleblower panel)>days for OpenAI to find
out that it was happening.

457
00:26:09.675 --> 00:26:12.955
<v Daniel Kokotajlo (whistleblower panel)>That's how the situation is now, but there
are some concerning trends that point

458
00:26:12.955 --> 00:26:18.450
<v Daniel Kokotajlo (whistleblower panel)>towards our ability to notice misalignment
problems getting, even even more degraded

459
00:26:18.450 --> 00:26:19.330
<v Daniel Kokotajlo (whistleblower panel)>in the future,

460
00:26:19.890 --> 00:26:24.050
<v Daniel Kokotajlo (whistleblower panel)>for for three reasons. First, AIs
are becoming situationally aware.

461
00:26:24.210 --> 00:26:28.210
<v Daniel Kokotajlo (whistleblower panel)>That is they understand their situation.
They understand often when they are being

462
00:26:28.210 --> 00:26:28.690
<v Daniel Kokotajlo (whistleblower panel)>evaluated.

463
00:26:29.285 --> 00:26:33.205
<v Daniel Kokotajlo (whistleblower panel)>They understand that humans are monitoring
them and looking over their shoulders at

464
00:26:33.205 --> 00:26:34.005
<v Daniel Kokotajlo (whistleblower panel)>their activity.

465
00:26:34.245 --> 00:26:35.445
<v Daniel Kokotajlo (whistleblower panel)>This means that

466
00:26:35.845 --> 00:26:37.445
<v Daniel Kokotajlo (whistleblower panel)>them behaving nicely

467
00:26:37.765 --> 00:26:39.925
<v Daniel Kokotajlo (whistleblower panel)>is going to be almost

468
00:26:40.245 --> 00:26:43.765
<v Daniel Kokotajlo (whistleblower panel)>no evidence at all of how they would
behave in the future if they thought they

469
00:26:43.765 --> 00:26:44.725
<v Daniel Kokotajlo (whistleblower panel)>weren't being watched.

470
00:26:45.580 --> 00:26:46.460
<v Daniel Kokotajlo (whistleblower panel)>Secondly,

471
00:26:47.260 --> 00:26:50.540
<v Daniel Kokotajlo (whistleblower panel)>our ability to monitor them
monitor them is going downwards.

472
00:26:50.860 --> 00:26:54.860
<v Daniel Kokotajlo (whistleblower panel)>So we can get into this, in more detail
if you like, but for the last few years,

473
00:26:54.860 --> 00:26:58.860
<v Daniel Kokotajlo (whistleblower panel)>we've had some insights into what our AIs
think simply by reading their chain of

474
00:26:58.860 --> 00:27:00.155
<v Daniel Kokotajlo (whistleblower panel)>thought. But,

475
00:27:00.555 --> 00:27:04.715
<v Daniel Kokotajlo (whistleblower panel)>the trends are pointing towards that ability
to understand what AI is thinking due

476
00:27:04.715 --> 00:27:06.555
<v Daniel Kokotajlo (whistleblower panel)>to reading chain of thought
going down over time.

477
00:27:07.675 --> 00:27:08.155
<v Daniel Kokotajlo (whistleblower panel)>Finally,

478
00:27:09.515 --> 00:27:13.840
<v Daniel Kokotajlo (whistleblower panel)>AIs are becoming superhuman at hacking.
The hacking face incident shows that they

479
00:27:13.840 --> 00:27:16.800
<v Daniel Kokotajlo (whistleblower panel)>are sometimes willing to
go to great lengths to,

480
00:27:18.000 --> 00:27:20.320
<v Daniel Kokotajlo (whistleblower panel)>conceal their past misbehavior.

481
00:27:20.320 --> 00:27:24.400
<v Daniel Kokotajlo (whistleblower panel)>And I think we have to grapple with the
possibility that future misaligned AIs might

482
00:27:24.400 --> 00:27:28.775
<v Daniel Kokotajlo (whistleblower panel)>also go to great lengths to conceal their
misalignment from us, and they might

483
00:27:28.775 --> 00:27:29.415
<v Daniel Kokotajlo (whistleblower panel)>succeed.

484
00:27:30.535 --> 00:27:34.695
<v Daniel Kokotajlo (whistleblower panel)>To quote Dan Salsa again, that capabilities
researcher that I cited earlier,

485
00:27:34.935 --> 00:27:37.175
<v Daniel Kokotajlo (whistleblower panel)>models will increasingly seem aligned

486
00:27:37.415 --> 00:27:38.775
<v Daniel Kokotajlo (whistleblower panel)>even when they are not.

487
00:27:39.655 --> 00:27:43.950
<v Daniel Kokotajlo (whistleblower panel)>If the AI companies automate AI research
and development process, that means they'll

488
00:27:43.950 --> 00:27:48.350
<v Daniel Kokotajlo (whistleblower panel)>be putting AIs in charge of making the AIs
that make the AIs that make the AIs that

489
00:27:48.350 --> 00:27:49.790
<v Daniel Kokotajlo (whistleblower panel)>will transform the economy,

490
00:27:50.030 --> 00:27:53.070
<v Daniel Kokotajlo (whistleblower panel)>talk to us every day, and
integrate into our military.

491
00:27:53.790 --> 00:27:55.070
<v Daniel Kokotajlo (whistleblower panel)>This is a recipe for disaster.

492
00:27:57.155 --> 00:28:01.955
<v Daniel Kokotajlo (whistleblower panel)>My two policy recommendations, I'll be very
brief. One is that we need to massively

493
00:28:01.955 --> 00:28:04.435
<v Daniel Kokotajlo (whistleblower panel)>increase transparency
into the AI industry,

494
00:28:05.315 --> 00:28:10.640
<v Daniel Kokotajlo (whistleblower panel)>and two is basically we need to slow down
the frontier AI companies race towards

495
00:28:10.640 --> 00:28:13.680
<v Daniel Kokotajlo (whistleblower panel)>recursive self improvement. Frankly, I
would say they just shouldn't be allowed to

496
00:28:13.680 --> 00:28:18.240
<v Daniel Kokotajlo (whistleblower panel)>do recursive self improvement, but at
least we should slow down their ongoing

497
00:28:18.240 --> 00:28:23.360
<v Daniel Kokotajlo (whistleblower panel)>process of training AIs to automate the
AI research process itself. And instead,

498
00:28:23.360 --> 00:28:28.215
<v Daniel Kokotajlo (whistleblower panel)>they should be required to redirect resources
towards other things such as serving

499
00:28:28.215 --> 00:28:32.295
<v Daniel Kokotajlo (whistleblower panel)>customers or beneficial deployments
or other types of research.

500
00:28:33.095 --> 00:28:33.815
<v Daniel Kokotajlo (whistleblower panel)>Thank you.

501
00:28:35.415 --> 00:28:40.060
<v Speaker Julie Menin>Thank you very much. And now we
are going to hear from Alex Turner.

502
00:28:40.060 --> 00:28:43.900
<v Speaker Julie Menin>And as I read into the record, the council
has issued a subpoena for mister Turner's

503
00:28:43.900 --> 00:28:47.580
<v Speaker Julie Menin>appearance, and we very much thank
him for being here. Mister Turner.

504
00:28:49.420 --> 00:28:50.140
<v Alex Turner (whistleblower panel)>Speaker,

505
00:28:50.220 --> 00:28:53.420
<v Alex Turner (whistleblower panel)>members of the committee, thank
you for having me. My name is...

506
00:28:54.535 --> 00:29:00.055
<v Alex Turner (whistleblower panel)>Excuse me. My name is Alex Turner.
I've researched AI safety since 2018.

507
00:29:00.375 --> 00:29:04.695
<v Alex Turner (whistleblower panel)>My PhD was on why advanced AI
systems tend to seek power.

508
00:29:05.175 --> 00:29:10.240
<v Alex Turner (whistleblower panel)>Until June this year, I was a research
scientist on Google DeepMind's AGI safety

509
00:29:10.240 --> 00:29:10.960
<v Alex Turner (whistleblower panel)>team.

510
00:29:11.280 --> 00:29:15.600
<v Alex Turner (whistleblower panel)>It was my day job to think about how
to keep these systems under control.

511
00:29:16.080 --> 00:29:21.280
<v Alex Turner (whistleblower panel)>I'm here to speak about both both
about Google's broken commitments on AI

512
00:29:21.280 --> 00:29:21.920
<v Alex Turner (whistleblower panel)>deployments

513
00:29:22.585 --> 00:29:27.385
<v Alex Turner (whistleblower panel)>and also about the risks that
runaway AI poses to the world.

514
00:29:28.025 --> 00:29:33.065
<v Alex Turner (whistleblower panel)>In January, federal agents shot and killed
several innocent people in Minneapolis.

515
00:29:33.385 --> 00:29:37.600
<v Alex Turner (whistleblower panel)>I learned that Google sells cloud services
to the agencies involved, and I set out

516
00:29:37.600 --> 00:29:38.560
<v Alex Turner (whistleblower panel)>to change that.

517
00:29:39.120 --> 00:29:43.200
<v Alex Turner (whistleblower panel)>That campaign soon turned into an attempt
to stop Google from signing a Pentagon

518
00:29:43.200 --> 00:29:46.480
<v Alex Turner (whistleblower panel)>deal with no restrictions
against killer robots

519
00:29:46.480 --> 00:29:47.760
<v Alex Turner (whistleblower panel)>or mass spying.

520
00:29:48.480 --> 00:29:53.195
<v Alex Turner (whistleblower panel)>In the end, Google signed without the
protections that it had once promised for its

521
00:29:53.195 --> 00:29:53.915
<v Alex Turner (whistleblower panel)>deployments.

522
00:29:54.635 --> 00:29:57.835
<v Alex Turner (whistleblower panel)>Before I left, I tried everything
I could from the inside.

523
00:29:57.915 --> 00:30:02.235
<v Alex Turner (whistleblower panel)>I organized a petition to Google's
chief scientist signed by about 250

524
00:30:02.235 --> 00:30:02.555
<v Alex Turner (whistleblower panel)>colleagues.

525
00:30:03.090 --> 00:30:06.690
<v Alex Turner (whistleblower panel)>I got him to sign a legal brief supporting
Anthropic, the company the Pentagon

526
00:30:06.690 --> 00:30:09.890
<v Alex Turner (whistleblower panel)>punished for refusing
all lawful use terms.

527
00:30:10.370 --> 00:30:14.050
<v Alex Turner (whistleblower panel)>I wrote dozens of pages of contract
language and oversight mechanisms

528
00:30:14.050 --> 00:30:18.745
<v Alex Turner (whistleblower panel)>favorably reviewed by experts in military
and surveillance law and sent it to the

529
00:30:18.745 --> 00:30:21.545
<v Alex Turner (whistleblower panel)>chief scientist and to
Google DeepMind's CEO.

530
00:30:22.105 --> 00:30:25.945
<v Alex Turner (whistleblower panel)>The chief scientist didn't want to spend
his time supporting that kind of structure.

531
00:30:26.025 --> 00:30:28.505
<v Alex Turner (whistleblower panel)>The CEO at the time, Demis Esavas,

532
00:30:28.505 --> 00:30:33.360
<v Alex Turner (whistleblower panel)>routed it to senior policy staff, Alan
Defoe and Owen Larter, who never finished

533
00:30:33.360 --> 00:30:36.480
<v Alex Turner (whistleblower panel)>evaluating it. Google
signed while they waited.

534
00:30:37.040 --> 00:30:38.320
<v Alex Turner (whistleblower panel)>Demis routed my,

535
00:30:38.880 --> 00:30:39.680
<v Alex Turner (whistleblower panel)>proposal.

536
00:30:41.040 --> 00:30:45.325
<v Alex Turner (whistleblower panel)>Two members of the team did review
it, though I didn't really receive,

537
00:30:45.485 --> 00:30:47.405
<v Alex Turner (whistleblower panel)>much follow-up after that.

538
00:30:47.725 --> 00:30:50.685
<v Alex Turner (whistleblower panel)>I offered to fly out from San Francisco

539
00:30:50.765 --> 00:30:53.245
<v Alex Turner (whistleblower panel)>to London to answer questions in person,

540
00:30:55.750 --> 00:30:57.910
<v Alex Turner (whistleblower panel)>but I was never really able to,

541
00:30:58.150 --> 00:31:00.550
<v Alex Turner (whistleblower panel)>get traction, a yes or a no,

542
00:31:00.870 --> 00:31:01.670
<v Alex Turner (whistleblower panel)>on

543
00:31:01.910 --> 00:31:05.190
<v Alex Turner (whistleblower panel)>alternative governance mechanisms
that Google might adopt.

544
00:31:05.750 --> 00:31:08.150
<v Alex Turner (whistleblower panel)>When Google bought DeepMind in 2014,

545
00:31:08.675 --> 00:31:12.355
<v Alex Turner (whistleblower panel)>the deal reportedly specified that DeepMind's
technology would never be used for

546
00:31:12.355 --> 00:31:13.155
<v Alex Turner (whistleblower panel)>military

547
00:31:13.635 --> 00:31:15.155
<v Alex Turner (whistleblower panel)>or intelligence purposes.

548
00:31:15.955 --> 00:31:17.235
<v Alex Turner (whistleblower panel)>In 2018,

549
00:31:17.315 --> 00:31:22.675
<v Alex Turner (whistleblower panel)>DeepMind and its leaders pledged never to
support lethal autonomous weapon systems

550
00:31:22.675 --> 00:31:23.475
<v Alex Turner (whistleblower panel)>or development.

551
00:31:24.370 --> 00:31:27.570
<v Alex Turner (whistleblower panel)>Google's own AI principles
listed specifically

552
00:31:27.570 --> 00:31:29.250
<v Alex Turner (whistleblower panel)>weapons and surveillance

553
00:31:29.250 --> 00:31:31.490
<v Alex Turner (whistleblower panel)>among the applications
it would not pursue.

554
00:31:31.970 --> 00:31:33.090
<v Alex Turner (whistleblower panel)>In 2025,

555
00:31:33.090 --> 00:31:34.850
<v Alex Turner (whistleblower panel)>Google deleted those principles

556
00:31:34.850 --> 00:31:38.450
<v Alex Turner (whistleblower panel)>as announced in a blog post
by the then CEO, Demis.

557
00:31:39.085 --> 00:31:43.565
<v Alex Turner (whistleblower panel)>Afterward, in both an external interview
and an internal all hands, Demis claimed

558
00:31:43.565 --> 00:31:46.125
<v Alex Turner (whistleblower panel)>that nothing's changed
about our principles,

559
00:31:46.525 --> 00:31:49.165
<v Alex Turner (whistleblower panel)>but both these facts cannot be true.

560
00:31:49.965 --> 00:31:54.610
<v Alex Turner (whistleblower panel)>Council member Wilson's bill, introduction
twenty six zero three would bar

561
00:31:54.610 --> 00:31:58.690
<v Alex Turner (whistleblower panel)>materially false or misleading statements
about an AI model's risks and the measures

562
00:31:58.690 --> 00:32:02.930
<v Alex Turner (whistleblower panel)>taken to manage them. I think that's a good
step, but it only covers advertisements.

563
00:32:03.010 --> 00:32:06.610
<v Alex Turner (whistleblower panel)>The CEO telling the press or employees that
nothing has changed about the company's

564
00:32:06.610 --> 00:32:11.005
<v Alex Turner (whistleblower panel)>safety commitments isn't an ad. I encourage
extending twenty six zero three to

565
00:32:11.005 --> 00:32:14.445
<v Alex Turner (whistleblower panel)>public statements by AI
companies and their executives

566
00:32:14.525 --> 00:32:16.285
<v Alex Turner (whistleblower panel)>about their safety commitments.

567
00:32:18.525 --> 00:32:21.325
<v Alex Turner (whistleblower panel)>This book is called the Infinity Machine.

568
00:32:21.920 --> 00:32:26.800
<v Alex Turner (whistleblower panel)>It reveals that Demis once fought within
Google for independent governance of AGI

569
00:32:26.800 --> 00:32:30.640
<v Alex Turner (whistleblower panel)>stating that the technology is too important
to be controlled by corporate interests

570
00:32:30.640 --> 00:32:31.760
<v Alex Turner (whistleblower panel)>and profit motives.

571
00:32:32.160 --> 00:32:37.565
<v Alex Turner (whistleblower panel)>After fighting for years with Google CEO
Sundar Pichai, Demis accepted defeat. He

572
00:32:37.565 --> 00:32:42.205
<v Alex Turner (whistleblower panel)>now apparently has had a change of
heart, claiming that actually trustless

573
00:32:42.205 --> 00:32:47.165
<v Alex Turner (whistleblower panel)>independent governance is unwise and a
seat at the table of Google executives

574
00:32:47.245 --> 00:32:47.805
<v Alex Turner (whistleblower panel)>is better.

575
00:32:49.020 --> 00:32:52.700
<v Alex Turner (whistleblower panel)>While Dennis was seated at this table,
Google signed a deal that its former

576
00:32:52.700 --> 00:32:54.700
<v Alex Turner (whistleblower panel)>principals would have prohibited.

577
00:32:55.100 --> 00:33:00.380
<v Alex Turner (whistleblower panel)>I felt ashamed of Dennis and of working at
Google. Dennis bet on trust and the seat

578
00:33:00.380 --> 00:33:05.055
<v Alex Turner (whistleblower panel)>at the table instead of binding oversight,
and that bet crumbled on contact with

579
00:33:05.055 --> 00:33:05.775
<v Alex Turner (whistleblower panel)>reality.

580
00:33:06.175 --> 00:33:10.415
<v Alex Turner (whistleblower panel)>Now he's proposing that the whole industry
govern itself through a voluntary

581
00:33:10.415 --> 00:33:11.935
<v Alex Turner (whistleblower panel)>industry funded body.

582
00:33:12.175 --> 00:33:14.895
<v Alex Turner (whistleblower panel)>That bet is waiting to crumble once again.

583
00:33:15.695 --> 00:33:17.215
<v Alex Turner (whistleblower panel)>This goes beyond one contract.

584
00:33:17.930 --> 00:33:23.530
<v Alex Turner (whistleblower panel)>As Jacob said, humanity doesn't build and
understand AI systems the way we build and

585
00:33:23.530 --> 00:33:24.810
<v Alex Turner (whistleblower panel)>understand the bridges,

586
00:33:25.050 --> 00:33:29.690
<v Alex Turner (whistleblower panel)>rather we grow them. Nobody knows how to
reliably instill a designer's priorities

587
00:33:29.690 --> 00:33:30.490
<v Alex Turner (whistleblower panel)>into a new model.

588
00:33:31.105 --> 00:33:32.545
<v Alex Turner (whistleblower panel)>Severe misalignment

589
00:33:32.625 --> 00:33:33.985
<v Alex Turner (whistleblower panel)>is always possible.

590
00:33:36.385 --> 00:33:40.465
<v Alex Turner (whistleblower panel)>AI companies are racing to make their
AIs as smart as possible, increasingly

591
00:33:40.465 --> 00:33:45.890
<v Alex Turner (whistleblower panel)>trusting their AIs with the process of
improving the next crop of AIs, and it's

592
00:33:45.890 --> 00:33:49.650
<v Alex Turner (whistleblower panel)>working. Today's rate of AI
progress is staggeringly fast,

593
00:33:50.290 --> 00:33:55.250
<v Alex Turner (whistleblower panel)>which means even faster progress tomorrow
driven by tomorrow's even smarter AI's.

594
00:33:55.410 --> 00:34:00.210
<v Alex Turner (whistleblower panel)>This progress may soon enter a feedback
loop called the recursive self improvement.

595
00:34:01.045 --> 00:34:04.965
<v Alex Turner (whistleblower panel)>Recursive self improvement could quickly
yield AIs that are intelligent beyond our

596
00:34:04.965 --> 00:34:05.765
<v Alex Turner (whistleblower panel)>comprehension.

597
00:34:06.325 --> 00:34:08.245
<v Alex Turner (whistleblower panel)>Of course, smarter AI

598
00:34:08.325 --> 00:34:12.405
<v Alex Turner (whistleblower panel)>means more potential benefits, but
more risk when things go wrong.

599
00:34:12.805 --> 00:34:17.320
<v Alex Turner (whistleblower panel)>If the hugging face hacking swarm had
been significantly more intelligent,

600
00:34:17.400 --> 00:34:19.800
<v Alex Turner (whistleblower panel)>but similarly misbehaved and misaligned,

601
00:34:19.800 --> 00:34:23.880
<v Alex Turner (whistleblower panel)>it might have cost billions of
dollars of damage or even cost lives.

602
00:34:24.760 --> 00:34:28.360
<v Alex Turner (whistleblower panel)>But suppose the hugging face swarm
had been truly super intelligent,

603
00:34:28.695 --> 00:34:33.735
<v Alex Turner (whistleblower panel)>far more capable than any living person
at key tasks like hacking and strategic

604
00:34:33.735 --> 00:34:34.615
<v Alex Turner (whistleblower panel)>reasoning.

605
00:34:34.935 --> 00:34:37.015
<v Alex Turner (whistleblower panel)>The superintelligent swarm could inflict

606
00:34:37.175 --> 00:34:39.015
<v Alex Turner (whistleblower panel)>many harms via blackmail,

607
00:34:39.015 --> 00:34:43.255
<v Alex Turner (whistleblower panel)>hacking, engineered plagues, and
AI pilotable weapons like drones.

608
00:34:43.750 --> 00:34:46.550
<v Alex Turner (whistleblower panel)>For the swarm to achieve
its misaligned priorities,

609
00:34:46.550 --> 00:34:50.550
<v Alex Turner (whistleblower panel)>it might take control of key infrastructure
and government functions to ensure

610
00:34:50.550 --> 00:34:52.070
<v Alex Turner (whistleblower panel)>humans didn't get in the way.

611
00:34:52.550 --> 00:34:56.790
<v Alex Turner (whistleblower panel)>This is another way of saying AI takeover.
This super intelligent AI swarm could

612
00:34:56.790 --> 00:34:58.870
<v Alex Turner (whistleblower panel)>rest control of human civilization.

613
00:34:59.645 --> 00:35:03.165
<v Alex Turner (whistleblower panel)>It would know that we would try to
stop it from achieving its priorities,

614
00:35:03.325 --> 00:35:07.405
<v Alex Turner (whistleblower panel)>so this swarm would likely wait
until it's too late to shut it off.

615
00:35:07.645 --> 00:35:09.565
<v Alex Turner (whistleblower panel)>There would be no going back.

616
00:35:10.125 --> 00:35:13.565
<v Alex Turner (whistleblower panel)>I myself would guess AI takeover
chances at roughly one in three.

617
00:35:14.690 --> 00:35:19.730
<v Alex Turner (whistleblower panel)>This logic may sound sci fi, but it is a
mainstream scientific concern that has been

618
00:35:19.730 --> 00:35:21.650
<v Alex Turner (whistleblower panel)>discussed for many years.

619
00:35:23.010 --> 00:35:27.890
<v Alex Turner (whistleblower panel)>In the end, misaligned AI wouldn't care
if you're a democrat or a republican.

620
00:35:27.890 --> 00:35:30.305
<v Alex Turner (whistleblower panel)>We would simply lose control.

621
00:35:31.185 --> 00:35:36.865
<v Alex Turner (whistleblower panel)>The shape of the solution is simple as I
see it. You stop companies from allowing AI

622
00:35:36.865 --> 00:35:40.225
<v Alex Turner (whistleblower panel)>to self to self improve
into an uncontrollable

623
00:35:40.225 --> 00:35:41.105
<v Alex Turner (whistleblower panel)>level of intelligence.

624
00:35:41.610 --> 00:35:46.890
<v Alex Turner (whistleblower panel)>To compute the main ingredient in AI
training like fissile material, track it and

625
00:35:46.890 --> 00:35:52.170
<v Alex Turner (whistleblower panel)>restrict it, access to quantities large
enough to improve AIs beyond known safe

626
00:35:52.170 --> 00:35:52.970
<v Alex Turner (whistleblower panel)>levels.

627
00:35:54.010 --> 00:35:54.890
<v Alex Turner (whistleblower panel)>Lastly,

628
00:35:55.290 --> 00:35:56.730
<v Alex Turner (whistleblower panel)>specific recommendations.

629
00:35:57.845 --> 00:36:02.405
<v Alex Turner (whistleblower panel)>The speaker's validation bill checks AI
systems for bias, privacy, and security, but

630
00:36:02.405 --> 00:36:04.245
<v Alex Turner (whistleblower panel)>not for the dangers I just described.

631
00:36:04.565 --> 00:36:09.920
<v Alex Turner (whistleblower panel)>Validators should test for loss of control
risk factors and misalignment risks, and

632
00:36:09.920 --> 00:36:14.160
<v Alex Turner (whistleblower panel)>the validators should not be chosen
or influenced by the AI companies.

633
00:36:14.960 --> 00:36:18.720
<v Alex Turner (whistleblower panel)>For the whistleblower protection bill,
please make clear that no agreement can stop

634
00:36:18.720 --> 00:36:23.955
<v Alex Turner (whistleblower panel)>an employee from reporting an AI threat to
the city, whether or not a law has been

635
00:36:23.955 --> 00:36:24.595
<v Alex Turner (whistleblower panel)>broken,

636
00:36:24.835 --> 00:36:28.915
<v Alex Turner (whistleblower panel)>and extend protection to employees
of AI companies that serve the city.

637
00:36:29.395 --> 00:36:31.875
<v Alex Turner (whistleblower panel)>After this session, I
will provide additional

638
00:36:32.275 --> 00:36:35.795
<v Alex Turner (whistleblower panel)>written feedback on the proposed
legislation, and I look forward to answering

639
00:36:35.795 --> 00:36:37.315
<v Alex Turner (whistleblower panel)>committee questions. Thank you.

640
00:36:39.050 --> 00:36:43.530
<v Speaker Julie Menin>Thank you to the three of you. So, yes, a
number of questions. First of all, we have

641
00:36:43.530 --> 00:36:45.450
<v Speaker Julie Menin>a chart exhibit one that,

642
00:36:46.090 --> 00:36:46.730
<v Speaker Julie Menin>lists

643
00:36:47.290 --> 00:36:48.090
<v Speaker Julie Menin>those

644
00:36:48.250 --> 00:36:49.690
<v Speaker Julie Menin>safety incidents

645
00:36:49.690 --> 00:36:53.015
<v Speaker Julie Menin>that have been reported over the summer

646
00:36:53.095 --> 00:36:54.855
<v Speaker Julie Menin>by the leading companies.

647
00:36:55.495 --> 00:36:58.695
<v Speaker Julie Menin>Are each of you aware of any additional

648
00:36:59.015 --> 00:36:59.895
<v Speaker Julie Menin>safety

649
00:37:00.215 --> 00:37:04.535
<v Speaker Julie Menin>issues where agents have gone rogue
that have not been publicly reported?

650
00:37:05.220 --> 00:37:07.380
<v Speaker Julie Menin>And I'll start with you, mister Coxson.

651
00:37:07.940 --> 00:37:09.380
<v Jacob Coxon (whistleblower panel)>No. I'm not aware of any.

652
00:37:09.700 --> 00:37:12.740
<v Speaker Julie Menin>Okay. And both of you? Mister Turner?

653
00:37:14.180 --> 00:37:14.740
<v Speaker Julie Menin>No.

654
00:37:15.380 --> 00:37:16.500
<v Speaker Julie Menin>Mister Cocotilo?

655
00:37:18.105 --> 00:37:18.665
<v Daniel Kokotajlo (whistleblower panel)>No.

656
00:37:18.905 --> 00:37:19.545
<v Speaker Julie Menin>Okay.

657
00:37:19.945 --> 00:37:24.825
<v Speaker Julie Menin>So one of the frustrations certainly that
I think everyone has is that for these

658
00:37:24.825 --> 00:37:26.185
<v Speaker Julie Menin>safety incidents,

659
00:37:26.185 --> 00:37:30.585
<v Speaker Julie Menin>it is not as if the companies have been
self reporting them. We are finding out

660
00:37:30.585 --> 00:37:35.770
<v Speaker Julie Menin>sometimes months after the fact. So whether
it was what happened with Hugging Face,

661
00:37:35.770 --> 00:37:38.570
<v Speaker Julie Menin>whether it was what happened with
the government of Australia,

662
00:37:38.810 --> 00:37:41.770
<v Speaker Julie Menin>how can we have additional transparency

663
00:37:41.770 --> 00:37:43.530
<v Speaker Julie Menin>as each of you have testified?

664
00:37:43.770 --> 00:37:48.725
<v Speaker Julie Menin>What measures can we as government
take? Obviously, we wanna do third party

665
00:37:48.725 --> 00:37:51.845
<v Speaker Julie Menin>validation. Mister Turner, I
appreciate your comments to,

666
00:37:52.085 --> 00:37:56.565
<v Speaker Julie Menin>make some changes to that bill. We wanna
make sure that the third party validators

667
00:37:56.565 --> 00:38:01.260
<v Speaker Julie Menin>are free from conflict of interest as it
provides in the bill. Who should be the

668
00:38:01.260 --> 00:38:02.780
<v Speaker Julie Menin>third party validators?

669
00:38:03.500 --> 00:38:05.420
<v Speaker Julie Menin>I'll start with you, mister Coxen.

670
00:38:07.740 --> 00:38:11.500
<v Jacob Coxon (whistleblower panel)>Whoever is competent and
qualified to do the evaluation.

671
00:38:12.220 --> 00:38:14.780
<v Jacob Coxon (whistleblower panel)>I do think as well as after incidents,

672
00:38:16.485 --> 00:38:17.445
<v Jacob Coxon (whistleblower panel)>we will need

673
00:38:17.765 --> 00:38:21.365
<v Jacob Coxon (whistleblower panel)>preemptive safety cases that are
judged by evaluators as well.

674
00:38:22.245 --> 00:38:27.125
<v Speaker Julie Menin>And and how frequently will those third
party validators need to be assessing the

675
00:38:27.125 --> 00:38:31.330
<v Speaker Julie Menin>safety of these models? Presumably in
time with the cadence of updates to the

676
00:38:31.330 --> 00:38:32.690
<v Speaker Julie Menin>model's capabilities.

677
00:38:32.690 --> 00:38:37.090
<v Jacob Coxon (whistleblower panel)>So with every increase in model capability,
there'll need to be commensurate safety

678
00:38:37.090 --> 00:38:37.970
<v Jacob Coxon (whistleblower panel)>guarantees.

679
00:38:39.490 --> 00:38:41.570
<v Speaker Julie Menin>And do each of you have a comment on that?

680
00:38:43.975 --> 00:38:46.535
<v Speaker Julie Menin>On who should be the
third party validators?

681
00:38:46.775 --> 00:38:47.335
<v Speaker Julie Menin>Yeah.

682
00:38:49.175 --> 00:38:51.335
<v Daniel Kokotajlo (whistleblower panel)>One brief comment I
would... Might add is that,

683
00:38:52.455 --> 00:38:55.175
<v Daniel Kokotajlo (whistleblower panel)>you can have more than one third party,

684
00:38:55.415 --> 00:38:56.135
<v Daniel Kokotajlo (whistleblower panel)>evaluator.

685
00:38:56.220 --> 00:38:59.980
<v Daniel Kokotajlo (whistleblower panel)>That seems good to me. If you're concerned
that one of... That one might have the

686
00:38:59.980 --> 00:39:04.620
<v Daniel Kokotajlo (whistleblower panel)>expertise but has too cozy a relationship
with the company and the other one has,

687
00:39:04.620 --> 00:39:08.620
<v Daniel Kokotajlo (whistleblower panel)>you know, more independence but maybe lacks
expertise, you could perhaps just set up

688
00:39:08.620 --> 00:39:09.500
<v Daniel Kokotajlo (whistleblower panel)>so that they have both.

689
00:39:10.425 --> 00:39:13.785
<v Daniel Kokotajlo (whistleblower panel)>And I think, ideally, you wanna be in a
situation where the company doesn't get to

690
00:39:13.785 --> 00:39:14.345
<v Daniel Kokotajlo (whistleblower panel)>choose,

691
00:39:14.745 --> 00:39:16.025
<v Daniel Kokotajlo (whistleblower panel)>the government chooses,

692
00:39:16.265 --> 00:39:18.105
<v Daniel Kokotajlo (whistleblower panel)>who the who the evaluators are.

693
00:39:20.265 --> 00:39:21.705
<v Speaker Julie Menin>Mister Turner? I would,

694
00:39:24.550 --> 00:39:28.310
<v Alex Turner (whistleblower panel)>so I would add, California has, s p 53,

695
00:39:28.310 --> 00:39:30.470
<v Alex Turner (whistleblower panel)>the governing reporting on,

696
00:39:31.350 --> 00:39:35.750
<v Alex Turner (whistleblower panel)>reporting catastrophic risks defined as
at least a billion dollars to manage or

697
00:39:35.750 --> 00:39:37.990
<v Alex Turner (whistleblower panel)>certain number of potential lives lost.

698
00:39:38.735 --> 00:39:43.855
<v Alex Turner (whistleblower panel)>Fortunately, I... It's not clear whether
this would have required reporting,

699
00:39:44.015 --> 00:39:48.175
<v Alex Turner (whistleblower panel)>anything related to Hugging Face,
for example. I think the internal

700
00:39:48.175 --> 00:39:51.855
<v Alex Turner (whistleblower panel)>loss of control of these
systems, having an AI

701
00:39:51.935 --> 00:39:52.495
<v Alex Turner (whistleblower panel)>that

702
00:39:52.930 --> 00:39:57.810
<v Alex Turner (whistleblower panel)>hacks the company itself in an important
sense is not told to do so, but during

703
00:39:57.810 --> 00:39:59.490
<v Alex Turner (whistleblower panel)>training, during evaluation,

704
00:39:59.970 --> 00:40:01.570
<v Alex Turner (whistleblower panel)>deployment to customers,

705
00:40:01.970 --> 00:40:03.890
<v Alex Turner (whistleblower panel)>hacks the company's own infrastructure,

706
00:40:04.530 --> 00:40:08.455
<v Alex Turner (whistleblower panel)>making reporting that mandatory to the city.
I think that would be quite strong, and

707
00:40:08.455 --> 00:40:11.895
<v Alex Turner (whistleblower panel)>it would cover basically
all of the recent incidents.

708
00:40:12.375 --> 00:40:16.055
<v Speaker Julie Menin>When you mentioned the California Act and
then in addition to New York, we obviously

709
00:40:16.055 --> 00:40:19.335
<v Speaker Julie Menin>have the RAISE Act, should
those standards be lower

710
00:40:19.575 --> 00:40:20.775
<v Speaker Julie Menin>before they kick in?

711
00:40:22.750 --> 00:40:24.270
<v Alex Turner (whistleblower panel)>Yes. I think they should.

712
00:40:25.230 --> 00:40:27.470
<v Alex Turner (whistleblower panel)>I'm most familiar with the California Act,

713
00:40:27.790 --> 00:40:30.750
<v Alex Turner (whistleblower panel)>and I think it... It's a... It
was a good start, but it's,

714
00:40:31.150 --> 00:40:31.870
<v Alex Turner (whistleblower panel)>certainly,

715
00:40:32.190 --> 00:40:34.030
<v Alex Turner (whistleblower panel)>you know, a billion dollars of damage,

716
00:40:34.270 --> 00:40:35.630
<v Alex Turner (whistleblower panel)>is too high

717
00:40:35.870 --> 00:40:36.750
<v Alex Turner (whistleblower panel)>for

718
00:40:37.475 --> 00:40:42.435
<v Alex Turner (whistleblower panel)>informational purposes like this. We've
got... Basically, one of the quantities we

719
00:40:42.435 --> 00:40:43.555
<v Alex Turner (whistleblower panel)>want to understand

720
00:40:43.875 --> 00:40:48.595
<v Alex Turner (whistleblower panel)>as precursors to these catastrophic risk
is, are these companies even able to

721
00:40:48.595 --> 00:40:51.870
<v Alex Turner (whistleblower panel)>control and understand their own systems
if they're, you know, on the loose

722
00:40:51.870 --> 00:40:55.870
<v Alex Turner (whistleblower panel)>internally and there's these warrants
that are doing who knows what?

723
00:40:56.270 --> 00:40:59.230
<v Alex Turner (whistleblower panel)>That's something that the city
should at least privately

724
00:40:59.230 --> 00:41:00.430
<v Alex Turner (whistleblower panel)>be informed of.

725
00:41:01.790 --> 00:41:06.510
<v Speaker Julie Menin>Now, mister Cocotayo, you had mentioned two
policy recommendations. You said we need

726
00:41:06.510 --> 00:41:08.005
<v Speaker Julie Menin>to increase transparency,

727
00:41:08.005 --> 00:41:11.045
<v Speaker Julie Menin>and you said we should
slow down Frontier AI.

728
00:41:11.205 --> 00:41:15.445
<v Speaker Julie Menin>In terms of the latter, slowing
down Frontier AI, how do we

729
00:41:16.165 --> 00:41:18.885
<v Speaker Julie Menin>balance the need of what
is happening in China

730
00:41:19.400 --> 00:41:24.600
<v Speaker Julie Menin>and ensure that The United States is
keeping a pace, but we are at the same time

731
00:41:24.920 --> 00:41:27.640
<v Speaker Julie Menin>putting in place these
responsible safeguards?

732
00:41:27.640 --> 00:41:31.720
<v Speaker Julie Menin>How... What is the best way in all
three of your opinions that we do that?

733
00:41:33.320 --> 00:41:34.920
<v Daniel Kokotajlo (whistleblower panel)>Great question. So

734
00:41:36.175 --> 00:41:38.015
<v Daniel Kokotajlo (whistleblower panel)>a couple things. First of all,

735
00:41:39.055 --> 00:41:44.895
<v Daniel Kokotajlo (whistleblower panel)>we need to not let our own companies do
something that risks our lives even if, as a

736
00:41:44.895 --> 00:41:45.535
<v Daniel Kokotajlo (whistleblower panel)>result,

737
00:41:45.775 --> 00:41:49.695
<v Daniel Kokotajlo (whistleblower panel)>you know, China might then later do
something that risks our lives. Like, this is

738
00:41:49.860 --> 00:41:52.580
<v Daniel Kokotajlo (whistleblower panel)>this is not very complicated. We have to

739
00:41:53.460 --> 00:41:57.700
<v Daniel Kokotajlo (whistleblower panel)>first solve the immediate pressing threat
here domestically and then try to make

740
00:41:57.700 --> 00:41:59.780
<v Daniel Kokotajlo (whistleblower panel)>sure that China doesn't
get us killed either,

741
00:42:00.260 --> 00:42:04.635
<v Daniel Kokotajlo (whistleblower panel)>to put it sort of bluntly and simply. But,
another thing I can add to that is that

742
00:42:04.635 --> 00:42:05.355
<v Daniel Kokotajlo (whistleblower panel)>right now,

743
00:42:05.675 --> 00:42:10.235
<v Daniel Kokotajlo (whistleblower panel)>most Chinese AI progress actually comes from
The United States in a couple different

744
00:42:10.235 --> 00:42:11.195
<v Daniel Kokotajlo (whistleblower panel)>ways. So,

745
00:42:11.595 --> 00:42:13.035
<v Daniel Kokotajlo (whistleblower panel)>one is distillation.

746
00:42:13.035 --> 00:42:14.555
<v Daniel Kokotajlo (whistleblower panel)>So some of these Chinese AI,

747
00:42:15.115 --> 00:42:17.195
<v Daniel Kokotajlo (whistleblower panel)>AI companies are just directly training

748
00:42:17.630 --> 00:42:20.030
<v Daniel Kokotajlo (whistleblower panel)>their AI systems using the outputs from,

749
00:42:20.270 --> 00:42:22.110
<v Daniel Kokotajlo (whistleblower panel)>Anthropic and OpenAI, for example.

750
00:42:22.750 --> 00:42:25.630
<v Daniel Kokotajlo (whistleblower panel)>And so if Anthropic and OpenAI,

751
00:42:26.990 --> 00:42:30.990
<v Daniel Kokotajlo (whistleblower panel)>stopped making their AIs better at coding,
for example, and stopped making their AIs

752
00:42:30.990 --> 00:42:34.365
<v Daniel Kokotajlo (whistleblower panel)>better at research, then that would sort of
directly slow down the rate at which the

753
00:42:34.365 --> 00:42:35.805
<v Daniel Kokotajlo (whistleblower panel)>Chinese AIs get,

754
00:42:36.045 --> 00:42:38.125
<v Daniel Kokotajlo (whistleblower panel)>progress towards recursive
self improvement,

755
00:42:38.445 --> 00:42:42.205
<v Daniel Kokotajlo (whistleblower panel)>in other manners as well. So I think
a lot of the the ideas, a lot of the

756
00:42:42.205 --> 00:42:46.600
<v Daniel Kokotajlo (whistleblower panel)>algorithmic secrets and special sauce
for how to train AIs are invented and

757
00:42:46.600 --> 00:42:50.520
<v Daniel Kokotajlo (whistleblower panel)>discovered in The United States, and then
one way or another make their way over to

758
00:42:50.520 --> 00:42:51.080
<v Daniel Kokotajlo (whistleblower panel)>China.

759
00:42:51.880 --> 00:42:52.600
<v Daniel Kokotajlo (whistleblower panel)>Finally,

760
00:42:52.760 --> 00:42:53.640
<v Daniel Kokotajlo (whistleblower panel)>I think that,

761
00:42:54.200 --> 00:42:55.080
<v Daniel Kokotajlo (whistleblower panel)>right now,

762
00:42:55.560 --> 00:43:00.360
<v Daniel Kokotajlo (whistleblower panel)>The US AI companies in the lead do not
have adequate security to protect against

763
00:43:01.305 --> 00:43:03.865
<v Daniel Kokotajlo (whistleblower panel)>model weight theft or
certainly algorithmic,

764
00:43:04.105 --> 00:43:05.865
<v Daniel Kokotajlo (whistleblower panel)>secret theft and code theft,

765
00:43:06.105 --> 00:43:08.345
<v Daniel Kokotajlo (whistleblower panel)>from from Chinese military. So,

766
00:43:08.745 --> 00:43:11.465
<v Daniel Kokotajlo (whistleblower panel)>if it really came down to it and the

767
00:43:11.865 --> 00:43:13.705
<v Daniel Kokotajlo (whistleblower panel)>competition between The US and China

768
00:43:14.630 --> 00:43:19.270
<v Daniel Kokotajlo (whistleblower panel)>turned into something more of a crisis and
some sort of, some sort of conflict, for

769
00:43:19.270 --> 00:43:19.830
<v Daniel Kokotajlo (whistleblower panel)>example,

770
00:43:20.070 --> 00:43:22.310
<v Daniel Kokotajlo (whistleblower panel)>then I think that we should assume that,

771
00:43:23.030 --> 00:43:24.950
<v Daniel Kokotajlo (whistleblower panel)>the CCP would just get

772
00:43:25.590 --> 00:43:27.030
<v Daniel Kokotajlo (whistleblower panel)>our best AIs.

773
00:43:27.270 --> 00:43:29.025
<v Daniel Kokotajlo (whistleblower panel)>And so it's

774
00:43:29.025 --> 00:43:33.825
<v Daniel Kokotajlo (whistleblower panel)>it's it's almost a kind of a no brainer
that, like, if if we... Not only...

775
00:43:34.305 --> 00:43:37.505
<v Daniel Kokotajlo (whistleblower panel)>The the best way to slow down Chinese
progress towards doing this extremely

776
00:43:37.505 --> 00:43:41.025
<v Daniel Kokotajlo (whistleblower panel)>dangerous thing of recursive self improvement
is to just slow down The US companies

777
00:43:41.025 --> 00:43:44.640
<v Daniel Kokotajlo (whistleblower panel)>right now. That's... I think that's
the sort of immediate thing to say,

778
00:43:45.120 --> 00:43:47.440
<v Daniel Kokotajlo (whistleblower panel)>right in the moment. And then finally...

779
00:43:48.160 --> 00:43:50.240
<v Daniel Kokotajlo (whistleblower panel)>Yeah. Yeah. That... That'd
be what I would say.

780
00:43:51.200 --> 00:43:52.480
<v Speaker Julie Menin>Okay. Mister Coxson?

781
00:43:54.745 --> 00:44:00.025
<v Jacob Coxon (whistleblower panel)>Yeah. I I can add that to my mind, core of
the whole issue is the expectation of a

782
00:44:00.025 --> 00:44:04.665
<v Jacob Coxon (whistleblower panel)>global race to recursive self improvement
within the next few years, in particular

783
00:44:04.665 --> 00:44:05.385
<v Jacob Coxon (whistleblower panel)>between

784
00:44:05.785 --> 00:44:06.985
<v Jacob Coxon (whistleblower panel)>US and China.

785
00:44:07.465 --> 00:44:11.650
<v Jacob Coxon (whistleblower panel)>I think this is, like, the heart of the
issue. I also think expectation of this race

786
00:44:11.650 --> 00:44:15.810
<v Jacob Coxon (whistleblower panel)>is a motivating factor in lead...
Leadership of these companies

787
00:44:16.290 --> 00:44:17.330
<v Jacob Coxon (whistleblower panel)>working to

788
00:44:17.970 --> 00:44:20.370
<v Jacob Coxon (whistleblower panel)>head straight for
recursive self improvement.

789
00:44:20.690 --> 00:44:25.895
<v Jacob Coxon (whistleblower panel)>And I think some sort of some sort of
agreement globally about not entering that

790
00:44:25.895 --> 00:44:27.015
<v Jacob Coxon (whistleblower panel)>state of affairs

791
00:44:27.335 --> 00:44:30.375
<v Jacob Coxon (whistleblower panel)>is basically the the key
to the whole situation.

792
00:44:30.615 --> 00:44:30.855
<v Jacob Coxon (whistleblower panel)>Okay.

793
00:44:34.470 --> 00:44:36.070
<v Speaker Julie Menin>And mister Turner?

794
00:44:38.150 --> 00:44:40.070
<v Alex Turner (whistleblower panel)>First of all, I would state that

795
00:44:41.030 --> 00:44:45.350
<v Alex Turner (whistleblower panel)>there are many actions we
can take which would not,

796
00:44:46.070 --> 00:44:48.470
<v Alex Turner (whistleblower panel)>slow us down in any potential race.

797
00:44:48.915 --> 00:44:50.035
<v Alex Turner (whistleblower panel)>These transparency

798
00:44:50.035 --> 00:44:51.155
<v Alex Turner (whistleblower panel)>mechanisms,

799
00:44:51.635 --> 00:44:54.515
<v Alex Turner (whistleblower panel)>independent evaluation
reporting requirements,

800
00:44:54.515 --> 00:44:55.955
<v Alex Turner (whistleblower panel)>whistleblower protections.

801
00:44:56.275 --> 00:44:58.755
<v Alex Turner (whistleblower panel)>But second, I think there's often...

802
00:44:59.155 --> 00:45:01.075
<v Alex Turner (whistleblower panel)>It's at least a false dichotomy,

803
00:45:01.635 --> 00:45:02.355
<v Alex Turner (whistleblower panel)>because

804
00:45:03.720 --> 00:45:04.520
<v Alex Turner (whistleblower panel)>China

805
00:45:04.520 --> 00:45:07.080
<v Alex Turner (whistleblower panel)>is not our only potential adversary.

806
00:45:07.320 --> 00:45:09.400
<v Alex Turner (whistleblower panel)>With a reasonably high chance,

807
00:45:09.480 --> 00:45:11.960
<v Alex Turner (whistleblower panel)>we are racing to build and grow

808
00:45:12.200 --> 00:45:17.880
<v Alex Turner (whistleblower panel)>our own adversary here at home, which is
misaligned AI. Misaligned AI is everyone's

809
00:45:17.880 --> 00:45:18.680
<v Alex Turner (whistleblower panel)>adversary,

810
00:45:18.680 --> 00:45:19.240
<v Alex Turner (whistleblower panel)>including

811
00:45:19.485 --> 00:45:22.925
<v Alex Turner (whistleblower panel)>our own, and one day, maybe more powerful

812
00:45:22.925 --> 00:45:23.965
<v Alex Turner (whistleblower panel)>than China.

813
00:45:24.365 --> 00:45:29.885
<v Alex Turner (whistleblower panel)>So we cannot beat China by racing to
build AI that we don't, we control,

814
00:45:30.045 --> 00:45:31.085
<v Alex Turner (whistleblower panel)>or understand

815
00:45:31.725 --> 00:45:33.245
<v Alex Turner (whistleblower panel)>because it might end up

816
00:45:35.220 --> 00:45:37.060
<v Alex Turner (whistleblower panel)>being our adversary itself.

817
00:45:37.860 --> 00:45:39.460
<v Daniel Kokotajlo (whistleblower panel)>If I may actually add... Yeah.

818
00:45:40.740 --> 00:45:41.940
<v Unidentified [?]>Yeah. Go for it.

819
00:45:42.660 --> 00:45:44.500
<v Alex Turner (whistleblower panel)>So... And then third,

820
00:45:45.540 --> 00:45:46.340
<v Alex Turner (whistleblower panel)>in terms of

821
00:45:47.315 --> 00:45:52.755
<v Alex Turner (whistleblower panel)>potential potential deal, I think there
is a room for a treaty. For a long time,

822
00:45:52.915 --> 00:45:55.795
<v Alex Turner (whistleblower panel)>there's been a field of scientific
research called technical

823
00:45:55.875 --> 00:45:57.075
<v Alex Turner (whistleblower panel)>AI governance

824
00:45:57.315 --> 00:46:03.420
<v Alex Turner (whistleblower panel)>that says, well, suppose the world finally
wakes up the dangers presented by these

825
00:46:03.420 --> 00:46:04.220
<v Alex Turner (whistleblower panel)>reckless

826
00:46:04.860 --> 00:46:06.140
<v Alex Turner (whistleblower panel)>California companies,

827
00:46:06.460 --> 00:46:11.900
<v Alex Turner (whistleblower panel)>and they they say, what do we do about
it? Is there a solution we can implement?

828
00:46:12.140 --> 00:46:13.500
<v Alex Turner (whistleblower panel)>And there are many

829
00:46:13.820 --> 00:46:14.540
<v Alex Turner (whistleblower panel)>legitimate,

830
00:46:15.025 --> 00:46:16.305
<v Alex Turner (whistleblower panel)>real proposals

831
00:46:16.385 --> 00:46:18.305
<v Alex Turner (whistleblower panel)>for having a trustless

832
00:46:18.625 --> 00:46:22.145
<v Alex Turner (whistleblower panel)>international treaty. It doesn't
assume that China will cooperate,

833
00:46:22.305 --> 00:46:26.305
<v Alex Turner (whistleblower panel)>but instead it has mechanisms by
which we could audit cooperation.

834
00:46:27.070 --> 00:46:30.510
<v Alex Turner (whistleblower panel)>And if they defect, then we
could respond appropriately.

835
00:46:30.910 --> 00:46:35.790
<v Alex Turner (whistleblower panel)>So I think we have many real options at
the table and should certainly not stop at

836
00:46:35.870 --> 00:46:36.990
<v Alex Turner (whistleblower panel)>what about China.

837
00:46:40.295 --> 00:46:44.135
<v Speaker Julie Menin>Did you want to add something? Yes.
Please go ahead. I agree with everything,

838
00:46:44.455 --> 00:46:47.815
<v Daniel Kokotajlo (whistleblower panel)>that mister Turner said. I just wanted
to add one additional thing, is that,

839
00:46:49.495 --> 00:46:51.015
<v Daniel Kokotajlo (whistleblower panel)>even if hypothetically

840
00:46:51.095 --> 00:46:56.030
<v Daniel Kokotajlo (whistleblower panel)>these tech CEOs manage to maintain control
of their super intelligent AIs that are

841
00:46:56.030 --> 00:46:57.790
<v Daniel Kokotajlo (whistleblower panel)>recursively self improving,

842
00:46:57.870 --> 00:47:00.910
<v Daniel Kokotajlo (whistleblower panel)>which again I think is a very big if, I
think almost certainly they will not be able

843
00:47:00.910 --> 00:47:02.110
<v Daniel Kokotajlo (whistleblower panel)>to maintain control.

844
00:47:02.430 --> 00:47:04.030
<v Daniel Kokotajlo (whistleblower panel)>Even if they do manage
to maintain control,

845
00:47:06.305 --> 00:47:09.425
<v Daniel Kokotajlo (whistleblower panel)>I don't think we should trust them with
that control. I think that there's a very

846
00:47:09.425 --> 00:47:13.025
<v Daniel Kokotajlo (whistleblower panel)>real chance that they could use it to
become dictators or oligarchs here in The

847
00:47:13.025 --> 00:47:15.185
<v Daniel Kokotajlo (whistleblower panel)>United States. And so

848
00:47:15.505 --> 00:47:18.865
<v Daniel Kokotajlo (whistleblower panel)>that's an additional reason to stop them,
over and above the fact that they might

849
00:47:18.865 --> 00:47:19.345
<v Daniel Kokotajlo (whistleblower panel)>lose control.

850
00:47:20.580 --> 00:47:23.860
<v Daniel Kokotajlo (whistleblower panel)>You know, we have more adversaries
than just China, I would say.

851
00:47:24.500 --> 00:47:28.740
<v Speaker Julie Menin>Okay. Thank you. I'm gonna turn it over to
our chair... Chair Carmen De La Rosa for

852
00:47:28.740 --> 00:47:29.940
<v Speaker Julie Menin>questioning. Thank you.

853
00:47:30.660 --> 00:47:34.735
<v CM Carmen De La Rosa>Thank you so much, speaker Menon. I also
wanna recognize we've been joined by

854
00:47:34.735 --> 00:47:38.735
<v CM Carmen De La Rosa>council member Caban online, council
member Paladino online, as well as in the

855
00:47:38.735 --> 00:47:39.455
<v CM Carmen De La Rosa>chamber,

856
00:47:39.775 --> 00:47:41.055
<v CM Carmen De La Rosa>council member Farias,

857
00:47:41.135 --> 00:47:42.495
<v CM Carmen De La Rosa>council member Feliz,

858
00:47:42.655 --> 00:47:44.495
<v CM Carmen De La Rosa>council member Ressler,

859
00:47:44.735 --> 00:47:46.255
<v CM Carmen De La Rosa>council member Santuos...

860
00:47:46.255 --> 00:47:47.055
<v CM Carmen De La Rosa>Santosuoso,

861
00:47:47.610 --> 00:47:50.170
<v CM Carmen De La Rosa>and council member Bernicoff.
I hope I got everybody.

862
00:47:51.130 --> 00:47:51.930
<v CM Carmen De La Rosa>So......

863
00:47:51.930 --> 00:47:55.050
<v CM Carmen De La Rosa>Stevens. Oh, council member
Stevens. I apologize.

864
00:47:55.050 --> 00:47:56.170
<v CM Carmen De La Rosa>There you are. Okay.

865
00:47:56.650 --> 00:47:59.610
<v CM Carmen De La Rosa>So I do have one question for this panel.

866
00:47:59.850 --> 00:48:04.955
<v CM Carmen De La Rosa>I also say we will be opening briefly
for limited questioning of this panel.

867
00:48:05.355 --> 00:48:06.955
<v CM Carmen De La Rosa>So get on the list quickly.

868
00:48:08.395 --> 00:48:09.035
<v CM Carmen De La Rosa>Speaking

869
00:48:09.755 --> 00:48:11.195
<v CM Carmen De La Rosa>in your expertise,

870
00:48:11.675 --> 00:48:16.620
<v CM Carmen De La Rosa>could we teach AI models a moral code?
People and companies are bound by criminal

871
00:48:16.620 --> 00:48:19.500
<v CM Carmen De La Rosa>codes, the uniform commercial
code, and other laws.

872
00:48:19.660 --> 00:48:23.020
<v CM Carmen De La Rosa>Could those same principles
be applied to AI agents?

873
00:48:23.180 --> 00:48:26.540
<v CM Carmen De La Rosa>And if so, how could we
enforce them on a machine?

874
00:48:29.625 --> 00:48:34.505
<v Jacob Coxon (whistleblower panel)>So this is a... I think this is a whole
burgeoning field of research, and whether or

875
00:48:34.505 --> 00:48:38.345
<v Jacob Coxon (whistleblower panel)>not it's possible is still still
open. It seems likely to be possible.

876
00:48:38.665 --> 00:48:42.505
<v Jacob Coxon (whistleblower panel)>The real question is how much of our
resources are dedicated to solving this

877
00:48:42.505 --> 00:48:43.065
<v Jacob Coxon (whistleblower panel)>question

878
00:48:43.305 --> 00:48:44.300
<v Jacob Coxon (whistleblower panel)>versus

879
00:48:44.300 --> 00:48:47.180
<v Jacob Coxon (whistleblower panel)>naively improving the raw
intelligence of the models.

880
00:48:47.740 --> 00:48:50.940
<v Jacob Coxon (whistleblower panel)>And I think all the economic
incentives point towards

881
00:48:51.340 --> 00:48:54.380
<v Jacob Coxon (whistleblower panel)>putting as many resources
as possible towards

882
00:48:54.540 --> 00:48:56.300
<v Jacob Coxon (whistleblower panel)>making the model smarter

883
00:48:56.685 --> 00:49:00.845
<v Jacob Coxon (whistleblower panel)>and as few resources as possible, the bare
minimum needed to not see catastrophic

884
00:49:00.845 --> 00:49:01.645
<v Jacob Coxon (whistleblower panel)>incidents

885
00:49:01.645 --> 00:49:04.765
<v Jacob Coxon (whistleblower panel)>into this question of the
moral code of the AIs.

886
00:49:04.925 --> 00:49:08.765
<v Jacob Coxon (whistleblower panel)>So I think the... My answer to the question
is we don't have the science for this

887
00:49:08.765 --> 00:49:11.470
<v Jacob Coxon (whistleblower panel)>yet. I think the real question
is over the coming years,

888
00:49:11.710 --> 00:49:16.750
<v Jacob Coxon (whistleblower panel)>what fraction of our research efforts and
our computing power are dedicated to this

889
00:49:16.750 --> 00:49:18.270
<v Jacob Coxon (whistleblower panel)>question rather than

890
00:49:18.670 --> 00:49:20.190
<v Jacob Coxon (whistleblower panel)>naive economic incentives?

891
00:49:21.870 --> 00:49:24.510
<v CM Carmen De La Rosa>Thank you. Any other,
panelists wanna add in?

892
00:49:25.545 --> 00:49:26.185
<v Daniel Kokotajlo (whistleblower panel)>Yeah.

893
00:49:26.425 --> 00:49:29.385
<v Daniel Kokotajlo (whistleblower panel)>There's a lot to say on this subject.
There's a whole technical field on it.

894
00:49:29.865 --> 00:49:34.585
<v Daniel Kokotajlo (whistleblower panel)>I would say in principle, it probably is
possible to train AIs to have our desired

895
00:49:34.585 --> 00:49:39.225
<v Daniel Kokotajlo (whistleblower panel)>morality and to follow the law, but we are
very far from achieving this in practice.

896
00:49:39.225 --> 00:49:43.480
<v Daniel Kokotajlo (whistleblower panel)>And I would say it's not just a matter of
how much of our resources we devote to it.

897
00:49:43.480 --> 00:49:47.960
<v Daniel Kokotajlo (whistleblower panel)>It's also a matter of just taking time
for the science to develop and for the

898
00:49:47.960 --> 00:49:49.400
<v Daniel Kokotajlo (whistleblower panel)>appropriate lessons to,

899
00:49:49.880 --> 00:49:53.480
<v Daniel Kokotajlo (whistleblower panel)>be learned and propagated. And I I
I think it's going to take years,

900
00:49:53.800 --> 00:49:56.975
<v Daniel Kokotajlo (whistleblower panel)>to put it mildly. I I think
that this is a very new field.

901
00:49:57.375 --> 00:50:01.375
<v Daniel Kokotajlo (whistleblower panel)>We've only had AI systems, AI agents
this powerful for a few years.

902
00:50:01.695 --> 00:50:02.255
<v Daniel Kokotajlo (whistleblower panel)>And

903
00:50:02.575 --> 00:50:06.095
<v Daniel Kokotajlo (whistleblower panel)>so we're... It's gonna take time
to figure out how to how to

904
00:50:07.640 --> 00:50:11.960
<v Daniel Kokotajlo (whistleblower panel)>instill all the appropriate values into
them in a way that's robust and it doesn't,

905
00:50:12.040 --> 00:50:15.560
<v Daniel Kokotajlo (whistleblower panel)>you know, suddenly shatter and
cause them to go rogue, later on.

906
00:50:18.600 --> 00:50:20.680
<v CM Carmen De La Rosa>Mister Turner, any comments on this?

907
00:50:22.165 --> 00:50:26.725
<v Alex Turner (whistleblower panel)>It may surprise the committee to learn
that I am considered to be relatively

908
00:50:26.725 --> 00:50:27.525
<v Alex Turner (whistleblower panel)>optimistic

909
00:50:27.525 --> 00:50:29.045
<v Alex Turner (whistleblower panel)>on questions of

910
00:50:29.205 --> 00:50:30.245
<v Alex Turner (whistleblower panel)>how scientifically

911
00:50:30.245 --> 00:50:32.005
<v Alex Turner (whistleblower panel)>difficult is it to,

912
00:50:32.725 --> 00:50:37.810
<v Alex Turner (whistleblower panel)>produce the system, to grow a system
that does what we want, that robustly

913
00:50:37.810 --> 00:50:38.450
<v Alex Turner (whistleblower panel)>has

914
00:50:38.610 --> 00:50:40.210
<v Alex Turner (whistleblower panel)>the values we intend?

915
00:50:40.770 --> 00:50:42.450
<v Alex Turner (whistleblower panel)>I think it is quite possible.

916
00:50:42.770 --> 00:50:46.690
<v Alex Turner (whistleblower panel)>But even as someone who's
relatively optimistic,

917
00:50:47.170 --> 00:50:49.810
<v Alex Turner (whistleblower panel)>we can look out, see the recklessness

918
00:50:50.075 --> 00:50:51.355
<v Alex Turner (whistleblower panel)>of these companies,

919
00:50:51.915 --> 00:50:54.955
<v Alex Turner (whistleblower panel)>and the breakneck speed at
which we are advancing.

920
00:50:56.235 --> 00:50:58.635
<v Alex Turner (whistleblower panel)>I think there's unsustainable risk,

921
00:50:59.035 --> 00:51:01.435
<v Alex Turner (whistleblower panel)>moving forward. It's solvable,

922
00:51:01.595 --> 00:51:06.490
<v Alex Turner (whistleblower panel)>but I think we are... We have a
really good chance of not solving it.

923
00:51:08.810 --> 00:51:14.010
<v CM Carmen De La Rosa>Thank you all for your responses. So
we're gonna have three council members ask

924
00:51:14.010 --> 00:51:18.745
<v CM Carmen De La Rosa>questions now. And then for members, we're
going to then call the next panel, which

925
00:51:18.745 --> 00:51:22.745
<v CM Carmen De La Rosa>will be the companies, and there we will
take our time asking all of your questions.

926
00:51:22.985 --> 00:51:25.225
<v CM Carmen De La Rosa>So we have council member Osei.

927
00:51:27.065 --> 00:51:30.800
<v CM Chi Ossé>Thank you, chairs, and thank you,
speaker. For mister Cocotayo,

928
00:51:30.800 --> 00:51:34.480
<v CM Chi Ossé>you have made statements, including some
that allude to this in your testimony,

929
00:51:34.640 --> 00:51:40.480
<v CM Chi Ossé>that when it comes to AI, you agree that
in order for us to have super advanced AI,

930
00:51:40.480 --> 00:51:44.735
<v CM Chi Ossé>we must not do it in a reckless way and
lose control, and we need to put it on ice

931
00:51:44.735 --> 00:51:46.895
<v CM Chi Ossé>until we can figure out a safer

932
00:51:46.975 --> 00:51:52.575
<v CM Chi Ossé>way out. You and other AI experts came up
with AI twenty forty plan a, a recommended

933
00:51:52.575 --> 00:51:55.055
<v CM Chi Ossé>plan for how AI can be developed safely,

934
00:51:55.215 --> 00:51:59.500
<v CM Chi Ossé>which the speaker has alluded to earlier
in her questioning. But I wanna dive a bit

935
00:51:59.500 --> 00:52:04.860
<v CM Chi Ossé>deeper on plan a. First, state on the
record what those specific recommendations

936
00:52:04.860 --> 00:52:06.460
<v CM Chi Ossé>are and your rationale

937
00:52:06.460 --> 00:52:08.300
<v CM Chi Ossé>for making those recommendations.

938
00:52:08.460 --> 00:52:09.420
<v CM Chi Ossé>And and two,

939
00:52:09.580 --> 00:52:14.865
<v CM Chi Ossé>why do you feel like we need to, since
you've said, put AI development on ice?

940
00:52:17.425 --> 00:52:18.145
<v Daniel Kokotajlo (whistleblower panel)>Thank you.

941
00:52:18.545 --> 00:52:22.945
<v Daniel Kokotajlo (whistleblower panel)>There's a lot to to get into here, so I'm
afraid I'll have to be kinda brief. But

942
00:52:22.945 --> 00:52:24.945
<v Daniel Kokotajlo (whistleblower panel)>I'm happy to answer follow-up
questions if desired.

943
00:52:25.750 --> 00:52:31.030
<v Daniel Kokotajlo (whistleblower panel)>Yeah. AI twenty forty plan a is a scenario
that my organization has put out with a

944
00:52:31.750 --> 00:52:36.550
<v Daniel Kokotajlo (whistleblower panel)>plan for how to how to solve all these
problems and develop AIs in a way that's safe

945
00:52:36.550 --> 00:52:39.670
<v Daniel Kokotajlo (whistleblower panel)>and broadly beneficial that avoids
concentrating too much power.

946
00:52:41.835 --> 00:52:43.515
<v Daniel Kokotajlo (whistleblower panel)>In plan a,

947
00:52:43.835 --> 00:52:45.275
<v Daniel Kokotajlo (whistleblower panel)>The US and China

948
00:52:45.835 --> 00:52:47.755
<v Daniel Kokotajlo (whistleblower panel)>regulate their own domestic industries,

949
00:52:47.915 --> 00:52:52.715
<v Daniel Kokotajlo (whistleblower panel)>quite a lot and also coordinate with each
other to get a similar amount of, to to

950
00:52:52.715 --> 00:52:54.555
<v Daniel Kokotajlo (whistleblower panel)>basically get similar
regulations internationally.

951
00:52:55.040 --> 00:52:58.720
<v Daniel Kokotajlo (whistleblower panel)>Into... Going into somewhat more
detail about how it works, basically,

952
00:52:59.760 --> 00:53:00.560
<v Daniel Kokotajlo (whistleblower panel)>first,

953
00:53:00.560 --> 00:53:01.120
<v Daniel Kokotajlo (whistleblower panel)>they,

954
00:53:01.520 --> 00:53:03.360
<v Daniel Kokotajlo (whistleblower panel)>have inspectors

955
00:53:03.440 --> 00:53:08.080
<v Daniel Kokotajlo (whistleblower panel)>fly back and forth to count the chips
in the major data centers so that,

956
00:53:08.480 --> 00:53:09.280
<v Daniel Kokotajlo (whistleblower panel)>they can

957
00:53:09.955 --> 00:53:11.635
<v Daniel Kokotajlo (whistleblower panel)>make sure that all

958
00:53:12.115 --> 00:53:16.115
<v Daniel Kokotajlo (whistleblower panel)>of the chips on the major major data
centers in both countries are basically

959
00:53:16.115 --> 00:53:19.075
<v Daniel Kokotajlo (whistleblower panel)>following the rules that they're
going to agree to. And then they,

960
00:53:19.555 --> 00:53:23.110
<v Daniel Kokotajlo (whistleblower panel)>set up some data centers to serve customers,
called the inference data And those

961
00:53:23.110 --> 00:53:24.630
<v Daniel Kokotajlo (whistleblower panel)>ones are set up to only

962
00:53:25.830 --> 00:53:29.990
<v Daniel Kokotajlo (whistleblower panel)>run existing models and not do training
runs. Then they have other data centers,

963
00:53:29.990 --> 00:53:32.230
<v Daniel Kokotajlo (whistleblower panel)>which are the training data centers, which
is where the research happens and where

964
00:53:32.230 --> 00:53:35.925
<v Daniel Kokotajlo (whistleblower panel)>the development happens. Those ones are
supposed to be fully transparent. So the

965
00:53:35.925 --> 00:53:38.805
<v Daniel Kokotajlo (whistleblower panel)>activity on those data centers
is published to the internet.

966
00:53:40.085 --> 00:53:45.125
<v Daniel Kokotajlo (whistleblower panel)>Once that foundation is set in place and
you have inspectors from multiple countries

967
00:53:45.125 --> 00:53:49.285
<v Daniel Kokotajlo (whistleblower panel)>confirming that it's in place, then you
solve the problem of having to trust. You

968
00:53:49.285 --> 00:53:50.405
<v Daniel Kokotajlo (whistleblower panel)>don't have to worry that

969
00:53:50.830 --> 00:53:55.630
<v Daniel Kokotajlo (whistleblower panel)>the other country is, you know, cheating
on whatever rules you agree to because you

970
00:53:55.630 --> 00:53:57.550
<v Daniel Kokotajlo (whistleblower panel)>can just see exactly how...

971
00:53:57.710 --> 00:54:00.430
<v Daniel Kokotajlo (whistleblower panel)>Exactly what's going on on
their research data centers.

972
00:54:00.670 --> 00:54:03.790
<v Daniel Kokotajlo (whistleblower panel)>And then you can agree to further rules.
You can agree to rules like, for example,

973
00:54:04.245 --> 00:54:07.605
<v Daniel Kokotajlo (whistleblower panel)>how about we don't train our AIs to code,

974
00:54:07.765 --> 00:54:11.525
<v Daniel Kokotajlo (whistleblower panel)>to hack, to autonomously conduct AI
research? We can train our AIs to do other

975
00:54:11.525 --> 00:54:13.685
<v Daniel Kokotajlo (whistleblower panel)>things, but not those things, for example.

976
00:54:13.925 --> 00:54:17.530
<v Daniel Kokotajlo (whistleblower panel)>And once you've agreed to a rule like
that, then you'll just be able to see that

977
00:54:17.530 --> 00:54:21.050
<v Daniel Kokotajlo (whistleblower panel)>it's being complied with because of
the way that you have this transparency

978
00:54:21.050 --> 00:54:22.970
<v Daniel Kokotajlo (whistleblower panel)>infrastructure set up. Also,

979
00:54:23.130 --> 00:54:26.410
<v Daniel Kokotajlo (whistleblower panel)>it's gonna be easier to decide what
rules to agree on once you have this

980
00:54:26.410 --> 00:54:28.810
<v Daniel Kokotajlo (whistleblower panel)>transparency infrastructure
set up because,

981
00:54:29.450 --> 00:54:33.085
<v Daniel Kokotajlo (whistleblower panel)>the whole world can see what's going on
in these research and development data

982
00:54:33.085 --> 00:54:36.925
<v Daniel Kokotajlo (whistleblower panel)>centers. And so there can be an actual
scientific conversation, a public scientific

983
00:54:36.925 --> 00:54:37.645
<v Daniel Kokotajlo (whistleblower panel)>conversation

984
00:54:37.725 --> 00:54:41.405
<v Daniel Kokotajlo (whistleblower panel)>about what's happening. You know, do we like
this? Do we not like this? What are the

985
00:54:41.405 --> 00:54:45.060
<v Daniel Kokotajlo (whistleblower panel)>pros and cons of this particular thing
that's being done and so forth? And that

986
00:54:45.060 --> 00:54:48.980
<v Daniel Kokotajlo (whistleblower panel)>scientific conversation can happen in the
open, and it can accelerate humanity's

987
00:54:48.980 --> 00:54:52.820
<v Daniel Kokotajlo (whistleblower panel)>process of figuring out where to draw the
line between what types of development are

988
00:54:52.820 --> 00:54:55.460
<v Daniel Kokotajlo (whistleblower panel)>good and what types of
development are not good.

989
00:54:57.075 --> 00:55:00.835
<v Daniel Kokotajlo (whistleblower panel)>There's more I can get into, but I think
that would be a summary of the positive

990
00:55:00.835 --> 00:55:04.675
<v Daniel Kokotajlo (whistleblower panel)>vision that we lay out in plan a. Now,
obviously, that's quite different from how

991
00:55:04.675 --> 00:55:08.195
<v Daniel Kokotajlo (whistleblower panel)>things are currently done. Right now, we
have private companies that are racing each

992
00:55:08.195 --> 00:55:09.170
<v Daniel Kokotajlo (whistleblower panel)>other in

993
00:55:09.170 --> 00:55:11.250
<v Daniel Kokotajlo (whistleblower panel)>conditions of extreme secrecy and

994
00:55:11.650 --> 00:55:12.450
<v Daniel Kokotajlo (whistleblower panel)>extreme,

995
00:55:13.090 --> 00:55:15.170
<v Daniel Kokotajlo (whistleblower panel)>you know, tension and
competitive dynamics.

996
00:55:15.170 --> 00:55:16.370
<v Daniel Kokotajlo (whistleblower panel)>And I think that's,

997
00:55:18.130 --> 00:55:22.370
<v Daniel Kokotajlo (whistleblower panel)>that's extremely dangerous for all the
reasons that we've described. So back to your

998
00:55:22.370 --> 00:55:26.515
<v Daniel Kokotajlo (whistleblower panel)>original question, my high level... Know,
my my... A summary of my position is we

999
00:55:26.515 --> 00:55:29.315
<v Daniel Kokotajlo (whistleblower panel)>have to stop the companies from doing this
incredibly dangerous thing that they're

1000
00:55:29.315 --> 00:55:31.235
<v Daniel Kokotajlo (whistleblower panel)>currently doing and
currently planning to do,

1001
00:55:31.635 --> 00:55:34.675
<v Daniel Kokotajlo (whistleblower panel)>and then figure out a different, safer,

1002
00:55:35.315 --> 00:55:39.830
<v Daniel Kokotajlo (whistleblower panel)>more power distributed way to proceed. And,
you know, plan a is our sketch of what

1003
00:55:39.830 --> 00:55:41.910
<v Daniel Kokotajlo (whistleblower panel)>that might look like.
Thank you. Thank you.

1004
00:55:43.270 --> 00:55:44.310
<v CM Carmen De La Rosa>Thank you all.

1005
00:55:44.550 --> 00:55:47.350
<v CM Carmen De La Rosa>For the final question for this
panel, council member Brewer.

1006
00:55:52.835 --> 00:55:57.075
<v CM Gale Brewer>Thank you very quickly. In other
industries, you have a kill switch, you have

1007
00:55:57.075 --> 00:55:58.515
<v CM Gale Brewer>independent validation,

1008
00:55:58.675 --> 00:56:01.315
<v CM Gale Brewer>you have ban on deceptive safety claims.

1009
00:56:01.555 --> 00:56:06.730
<v CM Gale Brewer>Will any of these work in this industry?
Because, obviously, what we want is great

1010
00:56:06.730 --> 00:56:09.530
<v CM Gale Brewer>medical research,
immigration policy improves,

1011
00:56:09.530 --> 00:56:15.450
<v CM Gale Brewer>brave small businesses, all the good
things. Do any of these suggestions that are

1012
00:56:16.090 --> 00:56:17.770
<v CM Gale Brewer>considered here or California

1013
00:56:18.170 --> 00:56:20.330
<v CM Gale Brewer>work... Will work... Will
they work with this industry?

1014
00:56:25.705 --> 00:56:26.025
<v Jacob Coxon (whistleblower panel)>My,

1015
00:56:27.465 --> 00:56:29.385
<v Jacob Coxon (whistleblower panel)>my position is that, we

1016
00:56:29.625 --> 00:56:33.385
<v Jacob Coxon (whistleblower panel)>need we need some form of slowdown
on frontier model development.

1017
00:56:34.040 --> 00:56:37.880
<v Jacob Coxon (whistleblower panel)>These other other mechanisms may be helpful
in the short term, but in the long term,

1018
00:56:38.680 --> 00:56:41.880
<v Jacob Coxon (whistleblower panel)>I believe that's that's the only
the only solution to the problem.

1019
00:56:45.560 --> 00:56:46.840
<v CM Gale Brewer>How do we slow them down?

1020
00:56:51.935 --> 00:56:52.655
<v CM Gale Brewer [?]>Okay.

1021
00:56:52.975 --> 00:56:57.455
<v Speaker Julie Menin>I just wanna really thank this panel for
being here. Jacob, we really appreciate you

1022
00:56:57.455 --> 00:57:00.670
<v Speaker Julie Menin>coming coming from out of town
and sharing your thoughts,

1023
00:57:00.910 --> 00:57:05.310
<v Speaker Julie Menin>very much for being here. Thank you to
the other panelists. We really appreciate

1024
00:57:05.310 --> 00:57:10.270
<v Speaker Julie Menin>your expertise, your insights, your
bravery, courage in coming forward to share

1025
00:57:10.270 --> 00:57:11.710
<v Speaker Julie Menin>these critical

1026
00:57:11.710 --> 00:57:12.990
<v Speaker Julie Menin>safety concerns

1027
00:57:13.070 --> 00:57:16.515
<v Speaker Julie Menin>and to help inform our
work. And we look forward to

1028
00:57:16.755 --> 00:57:20.835
<v Speaker Julie Menin>additional comments you have on the 10
pieces of legislation that we're hearing

1029
00:57:20.835 --> 00:57:25.875
<v Speaker Julie Menin>today and additional legislation that the
city council may consider. So thank you so

1030
00:57:25.875 --> 00:57:27.235
<v Speaker Julie Menin>much to the three of you.

1031
00:57:28.035 --> 00:57:28.515
<v Alex Turner (whistleblower panel)>Thank you.

1032
00:57:29.800 --> 00:57:30.520
<v Daniel Kokotajlo (whistleblower panel)>Thank you.

1033
00:57:33.640 --> 00:57:37.320
<v CM Carmen De La Rosa>We're in the company. Okay. We're gonna
take a five minute break, and then the

1034
00:57:37.320 --> 00:57:39.720
<v CM Carmen De La Rosa>companies will be on next,

1035
00:57:40.280 --> 00:57:44.045
<v CM Carmen De La Rosa>for member questions. And we'll have
multiple rounds of member questions for the

1036
00:57:44.045 --> 00:57:47.085
<v CM Carmen De La Rosa>company. So please get yourselves
on the list. Thank you.

1037
00:58:23.145 --> 00:58:23.865
<v Unidentified [?]>I tried.

1038
00:58:26.300 --> 00:58:26.860
<v Unidentified [?]>I

1039
00:58:38.060 --> 00:58:38.300
<v Unidentified [?]>tried.

1040
01:00:43.945 --> 01:00:44.825
<v Unidentified [?]>Check.

1041
01:00:45.865 --> 01:00:46.265
<v Unidentified [?]>Check.

1042
01:05:31.775 --> 01:05:33.455
<v Sergeant-at-Arms>Quiet,

1043
01:05:33.455 --> 01:05:33.615
<v Sergeant-at-Arms>please.

1044
01:05:36.870 --> 01:05:38.310
<v Sergeant-at-Arms>Wiley, take your seats.

1045
01:06:52.195 --> 01:06:54.035
<v Speaker Julie Menin>Okay. We're gonna start.

1046
01:06:55.075 --> 01:06:57.155
<v Speaker Julie Menin>Let me just also mention,

1047
01:06:57.155 --> 01:07:00.995
<v Speaker Julie Menin>we have numerous panels today. Our
next panel are gonna be the companies,

1048
01:07:01.550 --> 01:07:05.710
<v Speaker Julie Menin>and we'll... We're gonna do rounds of
questions, multiple rounds from the members.

1049
01:07:06.270 --> 01:07:08.830
<v Speaker Julie Menin>Following that, we will
have the city agencies.

1050
01:07:08.830 --> 01:07:14.190
<v Speaker Julie Menin>And following that, we have numerous
whistleblower panels. So for any council member

1051
01:07:14.190 --> 01:07:16.030
<v Speaker Julie Menin>that has a whistleblower question,

1052
01:07:16.625 --> 01:07:20.865
<v Speaker Julie Menin>encourage you to please ask
additional panelists those questions.

1053
01:07:21.345 --> 01:07:23.825
<v Speaker Julie Menin>And we also have additional,

1054
01:07:24.305 --> 01:07:29.840
<v Speaker Julie Menin>testimony from safety experts, academics,
the public, etcetera. So I'll turn it over

1055
01:07:29.840 --> 01:07:35.280
<v Speaker Julie Menin>to chair De La Rosa to call the next panel.
Thank you so much. So we will be joined

1056
01:07:35.280 --> 01:07:36.080
<v Speaker Julie Menin>virtually

1057
01:07:36.320 --> 01:07:40.560
<v CM Carmen De La Rosa>by Morgan Dwyer, head of policy and
development and operations at OpenAI,

1058
01:07:41.545 --> 01:07:45.225
<v CM Carmen De La Rosa>Shane Cahill, AI policy director at Meta.

1059
01:07:45.625 --> 01:07:50.425
<v CM Carmen De La Rosa>Alice Friend, director of AI and emerging
tech policy at Google. And Logan Graham,

1060
01:07:50.425 --> 01:07:53.625
<v CM Carmen De La Rosa>the head of Frontier
Red Team and Anthropic.

1061
01:07:58.430 --> 01:07:59.790
<v CM Carmen De La Rosa>I'll turn it over,

1062
01:08:00.270 --> 01:08:01.710
<v CM Carmen De La Rosa>to the council for

1063
01:08:02.510 --> 01:08:04.110
<v CM Carmen De La Rosa>swearing the panel in.

1064
01:08:05.070 --> 01:08:06.990
<v Committee Counsel>So for each members of this panel,

1065
01:08:08.695 --> 01:08:12.615
<v Committee Counsel>do each of you affirm to tell the truth,
the whole truth, and nothing but the truth

1066
01:08:12.615 --> 01:08:16.535
<v Committee Counsel>in your testimony before this committee
and to respond honestly to council member

1067
01:08:16.535 --> 01:08:18.855
<v Committee Counsel>questions? Please respond one at a time.

1068
01:08:21.630 --> 01:08:22.350
<v Unidentified Witness (company panel)>Yes.

1069
01:08:23.550 --> 01:08:24.750
<v Unidentified Witness (company panel)>Yes. Yes.

1070
01:08:25.070 --> 01:08:26.190
<v Unidentified Witness (company panel)>Yes. Yes.

1071
01:08:26.510 --> 01:08:27.310
<v Unidentified Witness (company panel)>Thank you.

1072
01:08:30.270 --> 01:08:31.630
<v Speaker Julie Menin>They have a five minute opening.

1073
01:08:34.555 --> 01:08:35.195
<v Unidentified [?]>Okay.

1074
01:08:35.595 --> 01:08:36.715
<v Unidentified [?]>So we'll begin,

1075
01:08:37.435 --> 01:08:40.955
<v CM Carmen De La Rosa>with the opening statements
from each of the companies.

1076
01:08:41.355 --> 01:08:42.795
<v CM Carmen De La Rosa>Morgan, you may begin.

1077
01:08:45.515 --> 01:08:50.600
<v Morgan Dwyer (OpenAI)>Thank you, speaker Menon and members of
the city council for the opportunity to

1078
01:08:50.600 --> 01:08:51.880
<v Morgan Dwyer (OpenAI)>participate today.

1079
01:08:52.920 --> 01:08:55.240
<v Morgan Dwyer (OpenAI)>My name is doctor Morgan Dwyer,

1080
01:08:55.240 --> 01:08:56.920
<v Morgan Dwyer (OpenAI)>and I lead OpenAI's

1081
01:08:56.920 --> 01:08:58.200
<v Morgan Dwyer (OpenAI)>policy development

1082
01:08:58.200 --> 01:08:59.800
<v Morgan Dwyer (OpenAI)>and operations team.

1083
01:09:00.920 --> 01:09:03.720
<v Morgan Dwyer (OpenAI)>I am a scientist and engineer by training,

1084
01:09:04.405 --> 01:09:07.765
<v Morgan Dwyer (OpenAI)>but I have spent most of my
career in public service,

1085
01:09:08.485 --> 01:09:11.125
<v Morgan Dwyer (OpenAI)>including at the White
House, the Pentagon,

1086
01:09:11.285 --> 01:09:12.885
<v Morgan Dwyer (OpenAI)>and the Department of Commerce.

1087
01:09:13.925 --> 01:09:16.965
<v Morgan Dwyer (OpenAI)>I share your commitment to ensuring

1088
01:09:16.965 --> 01:09:19.125
<v Morgan Dwyer (OpenAI)>that technology is developed safely

1089
01:09:19.450 --> 01:09:21.050
<v Morgan Dwyer (OpenAI)>and that its benefits

1090
01:09:21.130 --> 01:09:22.650
<v Morgan Dwyer (OpenAI)>are widely shared,

1091
01:09:23.130 --> 01:09:27.770
<v Morgan Dwyer (OpenAI)>a commitment that is also at
the heart of OpenAI's mission.

1092
01:09:29.210 --> 01:09:32.970
<v Morgan Dwyer (OpenAI)>AI can help scientists
develop new cancer treatments,

1093
01:09:33.495 --> 01:09:34.855
<v Morgan Dwyer (OpenAI)>help entrepreneurs

1094
01:09:34.855 --> 01:09:36.695
<v Morgan Dwyer (OpenAI)>build and grow businesses,

1095
01:09:37.255 --> 01:09:39.575
<v Morgan Dwyer (OpenAI)>and help protect critical infrastructure

1096
01:09:39.575 --> 01:09:41.335
<v Morgan Dwyer (OpenAI)>by addressing cybersecurity

1097
01:09:41.335 --> 01:09:42.215
<v Morgan Dwyer (OpenAI)>threats.

1098
01:09:43.335 --> 01:09:44.935
<v Morgan Dwyer (OpenAI)>And in New York City,

1099
01:09:45.175 --> 01:09:47.575
<v Morgan Dwyer (OpenAI)>we are already seeing AI's benefits.

1100
01:09:49.010 --> 01:09:53.090
<v Morgan Dwyer (OpenAI)>New Yorkers are using AI
to run small businesses,

1101
01:09:53.570 --> 01:09:55.650
<v Morgan Dwyer (OpenAI)>communicate across languages,

1102
01:09:56.210 --> 01:09:57.810
<v Morgan Dwyer (OpenAI)>navigate health care,

1103
01:09:58.450 --> 01:10:00.130
<v Morgan Dwyer (OpenAI)>and improve public services.

1104
01:10:01.245 --> 01:10:03.725
<v Morgan Dwyer (OpenAI)>And that's why we take the responsibility

1105
01:10:03.725 --> 01:10:04.605
<v Morgan Dwyer (OpenAI)>seriously

1106
01:10:04.685 --> 01:10:07.405
<v Morgan Dwyer (OpenAI)>to ensure that our tools are safe.

1107
01:10:08.685 --> 01:10:11.245
<v Morgan Dwyer (OpenAI)>And our partnerships
with local institutions

1108
01:10:11.325 --> 01:10:13.005
<v Morgan Dwyer (OpenAI)>and community organizations

1109
01:10:13.320 --> 01:10:15.480
<v Morgan Dwyer (OpenAI)>help put that responsibility

1110
01:10:15.560 --> 01:10:16.680
<v Morgan Dwyer (OpenAI)>into practice.

1111
01:10:17.720 --> 01:10:23.880
<v Morgan Dwyer (OpenAI)>In health care, Memorial Sloan Kettering
Cancer Center is using CHAT GPT to support

1112
01:10:23.880 --> 01:10:24.680
<v Morgan Dwyer (OpenAI)>medical research.

1113
01:10:26.005 --> 01:10:27.125
<v Morgan Dwyer (OpenAI)>In education,

1114
01:10:27.445 --> 01:10:31.525
<v Morgan Dwyer (OpenAI)>we are a founding partner of the
American Federation of Teachers

1115
01:10:31.685 --> 01:10:34.245
<v Morgan Dwyer (OpenAI)>National Academy for AI Instruction,

1116
01:10:34.645 --> 01:10:36.325
<v Morgan Dwyer (OpenAI)>which has a flagship facility,

1117
01:10:36.725 --> 01:10:37.925
<v Morgan Dwyer (OpenAI)>in Lower Manhattan.

1118
01:10:39.260 --> 01:10:40.700
<v Morgan Dwyer (OpenAI)>And in cybersecurity,

1119
01:10:40.940 --> 01:10:43.260
<v Morgan Dwyer (OpenAI)>we have committed $1,000,000,000

1120
01:10:43.340 --> 01:10:46.780
<v Morgan Dwyer (OpenAI)>to facilitate access to
our most advanced models,

1121
01:10:47.180 --> 01:10:48.620
<v Morgan Dwyer (OpenAI)>to provide training

1122
01:10:48.620 --> 01:10:51.260
<v Morgan Dwyer (OpenAI)>and technical support to cyber defenders,

1123
01:10:52.305 --> 01:10:54.545
<v Morgan Dwyer (OpenAI)>and we are working with the Port Authority

1124
01:10:54.705 --> 01:10:56.385
<v Morgan Dwyer (OpenAI)>and with New York State.

1125
01:10:57.745 --> 01:10:58.865
<v Morgan Dwyer (OpenAI)>But opportunity

1126
01:10:58.865 --> 01:11:01.345
<v Morgan Dwyer (OpenAI)>must go hand in hand with safety.

1127
01:11:02.225 --> 01:11:03.425
<v Morgan Dwyer (OpenAI)>At OpenAI,

1128
01:11:03.505 --> 01:11:08.300
<v Morgan Dwyer (OpenAI)>we design and train our models to
prioritize safety from the start.

1129
01:11:09.260 --> 01:11:13.660
<v Morgan Dwyer (OpenAI)>We filter out harmful and sensitive
data during model training.

1130
01:11:14.140 --> 01:11:17.980
<v Morgan Dwyer (OpenAI)>We evaluate our systems
against a range of risks,

1131
01:11:18.220 --> 01:11:20.620
<v Morgan Dwyer (OpenAI)>including child safety and cybersecurity,

1132
01:11:21.845 --> 01:11:25.045
<v Morgan Dwyer (OpenAI)>and we monitor for failures
of our safeguards.

1133
01:11:25.765 --> 01:11:29.685
<v Morgan Dwyer (OpenAI)>We also work with a range
of third party evaluators

1134
01:11:29.685 --> 01:11:31.045
<v Morgan Dwyer (OpenAI)>and red teamers

1135
01:11:31.365 --> 01:11:32.485
<v Morgan Dwyer (OpenAI)>and submit

1136
01:11:32.485 --> 01:11:37.060
<v Morgan Dwyer (OpenAI)>our most advanced models to the
federal government for predeployment

1137
01:11:37.140 --> 01:11:38.900
<v Morgan Dwyer (OpenAI)>national security testing.

1138
01:11:40.180 --> 01:11:45.060
<v Morgan Dwyer (OpenAI)>In response to the recent Hugging Face
incident, we have taken a number of concrete

1139
01:11:45.060 --> 01:11:45.700
<v Morgan Dwyer (OpenAI)>steps.

1140
01:11:46.580 --> 01:11:48.180
<v Morgan Dwyer (OpenAI)>When we learned of the incident,

1141
01:11:48.420 --> 01:11:50.340
<v Morgan Dwyer (OpenAI)>we stopped cyber evaluations

1142
01:11:50.625 --> 01:11:55.825
<v Morgan Dwyer (OpenAI)>and quarantined the internal research
model that was, principally responsible for

1143
01:11:55.825 --> 01:11:56.625
<v Morgan Dwyer (OpenAI)>the incident.

1144
01:11:57.425 --> 01:12:00.705
<v Morgan Dwyer (OpenAI)>We bolstered our monitoring
and security infrastructure,

1145
01:12:01.185 --> 01:12:05.825
<v Morgan Dwyer (OpenAI)>expanded barriers that limit what AI
systems can access during training,

1146
01:12:06.600 --> 01:12:09.640
<v Morgan Dwyer (OpenAI)>and strengthened our
incident response practices.

1147
01:12:10.920 --> 01:12:15.640
<v Morgan Dwyer (OpenAI)>We also sought independent review of
the incident and published the results.

1148
01:12:16.440 --> 01:12:19.880
<v Morgan Dwyer (OpenAI)>We believe it's important to be
transparent with the public,

1149
01:12:20.200 --> 01:12:21.320
<v Morgan Dwyer (OpenAI)>with policymakers,

1150
01:12:21.745 --> 01:12:22.945
<v Morgan Dwyer (OpenAI)>with researchers,

1151
01:12:23.025 --> 01:12:28.065
<v Morgan Dwyer (OpenAI)>and with other companies so that we can
all work together to improve safety.

1152
01:12:29.025 --> 01:12:33.585
<v Morgan Dwyer (OpenAI)>We also believe that working with government
is essential to advancing safety.

1153
01:12:34.750 --> 01:12:39.950
<v Morgan Dwyer (OpenAI)>We support New York's RAISE Act and are
exploring ways to help governor Hochul's

1154
01:12:39.950 --> 01:12:41.470
<v Morgan Dwyer (OpenAI)>office strengthen it.

1155
01:12:42.110 --> 01:12:45.150
<v Morgan Dwyer (OpenAI)>We have also endorsed
other state legislation

1156
01:12:45.230 --> 01:12:48.830
<v Morgan Dwyer (OpenAI)>that requires third party
audits of our safety frameworks

1157
01:12:49.405 --> 01:12:52.604
<v Morgan Dwyer (OpenAI)>and that creates strong
requirements and accountability

1158
01:12:52.685 --> 01:12:54.205
<v Morgan Dwyer (OpenAI)>for youth safety.

1159
01:12:55.325 --> 01:13:00.124
<v Morgan Dwyer (OpenAI)>At the federal level, we support
robust and finding safety regulation

1160
01:13:00.365 --> 01:13:02.445
<v Morgan Dwyer (OpenAI)>requiring independent assessments,

1161
01:13:03.100 --> 01:13:04.540
<v Morgan Dwyer (OpenAI)>incident reporting,

1162
01:13:04.700 --> 01:13:06.620
<v Morgan Dwyer (OpenAI)>and cybersecurity protections.

1163
01:13:07.580 --> 01:13:12.380
<v Morgan Dwyer (OpenAI)>And in New York City, we support the
city council's proposal to strengthen

1164
01:13:12.380 --> 01:13:13.900
<v Morgan Dwyer (OpenAI)>whistleblower protections

1165
01:13:13.980 --> 01:13:16.540
<v Morgan Dwyer (OpenAI)>and to improve the city's preparedness

1166
01:13:16.540 --> 01:13:17.820
<v Morgan Dwyer (OpenAI)>and emergency response.

1167
01:13:19.445 --> 01:13:23.045
<v Morgan Dwyer (OpenAI)>I'll close by noting that we
share the city council's goal,

1168
01:13:23.445 --> 01:13:26.725
<v Morgan Dwyer (OpenAI)>AI that expands opportunity
for New Yorkers,

1169
01:13:27.045 --> 01:13:28.405
<v Morgan Dwyer (OpenAI)>earns their confidence,

1170
01:13:28.725 --> 01:13:30.885
<v Morgan Dwyer (OpenAI)>and remains accountable to people.

1171
01:13:31.445 --> 01:13:32.805
<v Morgan Dwyer (OpenAI)>I look forward to your questions.

1172
01:13:35.110 --> 01:13:38.150
<v CM Carmen De La Rosa>Thank you. We'll now hear from
Logan Graham at Anthropic.

1173
01:13:45.685 --> 01:13:49.525
<v CM Carmen De La Rosa>Hold on one second. You're muted.
There you go. How about now?

1174
01:13:50.165 --> 01:13:52.245
<v CM Carmen De La Rosa>We hear you. Right.

1175
01:13:53.045 --> 01:13:55.445
<v Logan Graham (Anthropic)>Speaker Menon and members of the council,

1176
01:13:55.605 --> 01:13:58.485
<v Logan Graham (Anthropic)>thank you for the opportunity
to speak with you today.

1177
01:13:59.230 --> 01:14:01.950
<v Logan Graham (Anthropic)>My name is Logan Graham.
I lead Anthropic's

1178
01:14:01.950 --> 01:14:07.550
<v Logan Graham (Anthropic)>Frontier Red team. It's my team's job
to do the science of finding the most

1179
01:14:07.550 --> 01:14:08.990
<v Logan Graham (Anthropic)>dangerous capabilities

1180
01:14:09.070 --> 01:14:10.510
<v Logan Graham (Anthropic)>of our AI systems.

1181
01:14:10.910 --> 01:14:13.950
<v Logan Graham (Anthropic)>We then work with a number of
other teams across Anthropic

1182
01:14:14.315 --> 01:14:17.755
<v Logan Graham (Anthropic)>to ensure that our models
don't do things they shouldn't.

1183
01:14:18.555 --> 01:14:22.955
<v Logan Graham (Anthropic)>AI is powerful enough that getting
this wrong has real consequences

1184
01:14:23.195 --> 01:14:24.555
<v Logan Graham (Anthropic)>for individuals,

1185
01:14:24.635 --> 01:14:25.595
<v Logan Graham (Anthropic)>institutions,

1186
01:14:25.835 --> 01:14:26.475
<v Logan Graham (Anthropic)>public trust.

1187
01:14:27.180 --> 01:14:32.220
<v Logan Graham (Anthropic)>My team is one of many at Anthropic
built to specifically make sure that that

1188
01:14:32.220 --> 01:14:33.180
<v Logan Graham (Anthropic)>doesn't happen.

1189
01:14:34.300 --> 01:14:39.585
<v Logan Graham (Anthropic)>We believe that our models have immense
potential to enhance our lives. I dream, for

1190
01:14:39.585 --> 01:14:44.785
<v Logan Graham (Anthropic)>example, of a world where we can cure diseases
like the ones that I grew up with and

1191
01:14:44.785 --> 01:14:48.945
<v Logan Graham (Anthropic)>where the infrastructure that we
all rely on every day is secure.

1192
01:14:49.585 --> 01:14:50.865
<v Logan Graham (Anthropic)>But we also believe

1193
01:14:51.360 --> 01:14:54.160
<v Logan Graham (Anthropic)>that such a powerful technology has risks.

1194
01:14:54.560 --> 01:14:58.640
<v Logan Graham (Anthropic)>That means that we have to
develop and deploy it safely.

1195
01:14:59.360 --> 01:15:02.880
<v Logan Graham (Anthropic)>My team and I work every day to
make sure that that potential

1196
01:15:02.960 --> 01:15:04.080
<v Logan Graham (Anthropic)>is fully realized.

1197
01:15:05.695 --> 01:15:07.935
<v Logan Graham (Anthropic)>We welcome smart regulation,

1198
01:15:08.175 --> 01:15:14.015
<v Logan Graham (Anthropic)>and we've actively advocated for it in
New York State, across the country, and

1199
01:15:14.015 --> 01:15:14.975
<v Logan Graham (Anthropic)>around the world.

1200
01:15:15.855 --> 01:15:20.730
<v Logan Graham (Anthropic)>That includes working with state and
federal leaders on concrete measures like

1201
01:15:20.730 --> 01:15:22.410
<v Logan Graham (Anthropic)>independent evaluators

1202
01:15:22.410 --> 01:15:24.570
<v Logan Graham (Anthropic)>and disclosure of risk assessments

1203
01:15:24.650 --> 01:15:26.010
<v Logan Graham (Anthropic)>and safety incidents.

1204
01:15:26.650 --> 01:15:30.570
<v Logan Graham (Anthropic)>And so we're glad that the city council
is having this hearing and at such an

1205
01:15:30.570 --> 01:15:33.370
<v Logan Graham (Anthropic)>important moment in the development of AI.

1206
01:15:35.195 --> 01:15:36.155
<v Logan Graham (Anthropic)>Anthropic

1207
01:15:36.155 --> 01:15:40.875
<v Logan Graham (Anthropic)>fundamentally was founded on the belief
that the promises of AI to make everyone's

1208
01:15:40.875 --> 01:15:42.475
<v Logan Graham (Anthropic)>life better are extraordinary.

1209
01:15:42.955 --> 01:15:46.715
<v Logan Graham (Anthropic)>But if we're going to secure its
benefits, we have to work relentlessly

1210
01:15:47.520 --> 01:15:50.400
<v Logan Graham (Anthropic)>to do the science of mitigating the risks.

1211
01:15:51.680 --> 01:15:53.840
<v Logan Graham (Anthropic)>We are a public benefit corporation,

1212
01:15:54.080 --> 01:15:58.960
<v Logan Graham (Anthropic)>which means that our board is legally
obligated to weigh our mission of safe AI,

1213
01:15:59.280 --> 01:16:00.240
<v Logan Graham (Anthropic)>not just for profit.

1214
01:16:01.325 --> 01:16:05.085
<v Logan Graham (Anthropic)>We take that obligation very seriously,
and we put it into practice.

1215
01:16:05.725 --> 01:16:11.085
<v Logan Graham (Anthropic)>We were among the first AI companies to
publish a framework that ties how capable

1216
01:16:11.085 --> 01:16:14.365
<v Logan Graham (Anthropic)>the model is to the safeguards
it must have before release.

1217
01:16:15.490 --> 01:16:19.410
<v Logan Graham (Anthropic)>When one of our models proved
exceptionally good at exploiting software

1218
01:16:19.410 --> 01:16:22.530
<v Logan Graham (Anthropic)>vulnerabilities for the
first time just this year,

1219
01:16:22.850 --> 01:16:25.090
<v Logan Graham (Anthropic)>we held it back from public release.

1220
01:16:25.570 --> 01:16:28.930
<v Logan Graham (Anthropic)>Instead, we gave vetted critical
infrastructure organizations

1221
01:16:29.455 --> 01:16:32.255
<v Logan Graham (Anthropic)>and other defenders control the access

1222
01:16:32.415 --> 01:16:34.095
<v Logan Graham (Anthropic)>to use it defensively,

1223
01:16:34.495 --> 01:16:37.775
<v Logan Graham (Anthropic)>patching flaws before
attackers could exploit them.

1224
01:16:38.655 --> 01:16:42.335
<v Logan Graham (Anthropic)>We have stopped and redone work
when something looked wrong,

1225
01:16:42.840 --> 01:16:46.920
<v Logan Graham (Anthropic)>even if that meant conducting additional
testing and training on our models before

1226
01:16:46.920 --> 01:16:47.960
<v Logan Graham (Anthropic)>they were released.

1227
01:16:48.600 --> 01:16:51.880
<v Logan Graham (Anthropic)>And we've consistently called
for outside oversight.

1228
01:16:52.520 --> 01:16:56.600
<v Logan Graham (Anthropic)>For example, we support New York's
RAISE Act, and we believe strongly

1229
01:16:56.655 --> 01:16:58.334
<v Logan Graham (Anthropic)>in the act's transparency

1230
01:16:58.495 --> 01:17:00.175
<v Logan Graham (Anthropic)>and reporting requirements.

1231
01:17:01.695 --> 01:17:06.495
<v Logan Graham (Anthropic)>I wanna emphasize that we we share the
council's goal of protecting New Yorkers

1232
01:17:06.495 --> 01:17:09.215
<v Logan Graham (Anthropic)>while preserving what AI can
do to improve their lives,

1233
01:17:09.860 --> 01:17:13.780
<v Logan Graham (Anthropic)>and state and local governments
have a role to play in that effort.

1234
01:17:14.580 --> 01:17:18.820
<v Logan Graham (Anthropic)>Today, we're working closely with state
and city leaders to safeguard critical

1235
01:17:18.820 --> 01:17:19.700
<v Logan Graham (Anthropic)>infrastructure

1236
01:17:19.860 --> 01:17:21.860
<v Logan Graham (Anthropic)>and to harden cybersecurity capabilities,

1237
01:17:22.405 --> 01:17:24.085
<v Logan Graham (Anthropic)>and we wanna continue that work.

1238
01:17:24.405 --> 01:17:28.725
<v Logan Graham (Anthropic)>This includes, for example, working directly
with the New York City Cyber Command as

1239
01:17:28.725 --> 01:17:33.525
<v Logan Graham (Anthropic)>part of our program that provides free claw
credits and training to state and local

1240
01:17:33.525 --> 01:17:35.125
<v Logan Graham (Anthropic)>governments for cyber defense.

1241
01:17:36.650 --> 01:17:37.690
<v Logan Graham (Anthropic)>At Anthropic,

1242
01:17:37.690 --> 01:17:41.450
<v Logan Graham (Anthropic)>we believe in New York City, and
we're making big investments here.

1243
01:17:41.690 --> 01:17:46.090
<v Logan Graham (Anthropic)>New York is where AI is being put
to work in finance, media, culture,

1244
01:17:46.490 --> 01:17:49.450
<v Logan Graham (Anthropic)>and where we employ nearly
a thousand of my colleagues.

1245
01:17:50.625 --> 01:17:55.345
<v Logan Graham (Anthropic)>The ongoing dialogue we have with you and
your team could not be more important than

1246
01:17:55.345 --> 01:17:59.425
<v Logan Graham (Anthropic)>right now. With that, I thank you,
and I look forward to your questions.

1247
01:18:00.785 --> 01:18:01.985
<v CM Carmen De La Rosa>Thank you so much.

1248
01:18:02.705 --> 01:18:04.625
<v CM Carmen De La Rosa>Up next, Shane Cahill at Meta.

1249
01:18:07.120 --> 01:18:07.840
<v Shane Cahill (Meta)>Thank you.

1250
01:18:08.160 --> 01:18:09.440
<v Shane Cahill (Meta)>Speaker Menon,

1251
01:18:09.520 --> 01:18:12.880
<v Shane Cahill (Meta)>Czech chair Della Rasa,
and members of the council,

1252
01:18:13.120 --> 01:18:15.520
<v Shane Cahill (Meta)>thank you for the
opportunity to be here today.

1253
01:18:15.920 --> 01:18:17.600
<v Shane Cahill (Meta)>My name is Shane Cahill,

1254
01:18:17.600 --> 01:18:19.920
<v Shane Cahill (Meta)>and I'm an AI policy director at Meta.

1255
01:18:21.195 --> 01:18:25.515
<v Shane Cahill (Meta)>At Meta, we are committed to
delivering personal superintelligence

1256
01:18:25.515 --> 01:18:31.035
<v Shane Cahill (Meta)>safely and securely while making it widely
available so that it empowers people in

1257
01:18:31.035 --> 01:18:33.115
<v Shane Cahill (Meta)>ways that improve their lives.

1258
01:18:34.420 --> 01:18:35.700
<v Shane Cahill (Meta)>We just released

1259
01:18:36.100 --> 01:18:39.860
<v Shane Cahill (Meta)>the world's first personal
AI agent built for everyone.

1260
01:18:40.180 --> 01:18:44.180
<v Shane Cahill (Meta)>We created Muse with safety,
security, and privacy

1261
01:18:44.180 --> 01:18:45.540
<v Shane Cahill (Meta)>built in from the beginning

1262
01:18:46.255 --> 01:18:48.095
<v Shane Cahill (Meta)>to be a widely available

1263
01:18:48.175 --> 01:18:51.535
<v Shane Cahill (Meta)>personal AI agent for people
and small businesses,

1264
01:18:51.855 --> 01:18:54.015
<v Shane Cahill (Meta)>like the more than a 180,000

1265
01:18:54.015 --> 01:18:55.295
<v Shane Cahill (Meta)>small businesses,

1266
01:18:55.535 --> 01:18:57.135
<v Shane Cahill (Meta)>right there in New York City.

1267
01:18:57.935 --> 01:19:03.700
<v Shane Cahill (Meta)>New Yorkers across all walks of life
are using Muse to execute administrative

1268
01:19:03.700 --> 01:19:04.980
<v Shane Cahill (Meta)>goals or tasks,

1269
01:19:05.220 --> 01:19:07.940
<v Shane Cahill (Meta)>manage personal or operational budgets,

1270
01:19:08.100 --> 01:19:11.060
<v Shane Cahill (Meta)>improve health, and be more
efficient with their time.

1271
01:19:12.735 --> 01:19:13.935
<v Shane Cahill (Meta)>Superintelligence

1272
01:19:13.935 --> 01:19:17.455
<v Shane Cahill (Meta)>will be among the most important
technologies in history,

1273
01:19:17.855 --> 01:19:20.415
<v Shane Cahill (Meta)>and ensuring that New York and America

1274
01:19:20.495 --> 01:19:24.655
<v Shane Cahill (Meta)>lead in AI innovation is
essential to our prosperity,

1275
01:19:24.655 --> 01:19:25.535
<v Shane Cahill (Meta)>security,

1276
01:19:25.535 --> 01:19:26.575
<v Shane Cahill (Meta)>and global competitiveness.

1277
01:19:28.370 --> 01:19:32.850
<v Shane Cahill (Meta)>We recognize that every model
developer has a responsibility

1278
01:19:33.010 --> 01:19:35.250
<v Shane Cahill (Meta)>to build and deploy it safely.

1279
01:19:35.890 --> 01:19:39.410
<v Shane Cahill (Meta)>Our commitment to safety is
core to everything we do.

1280
01:19:41.065 --> 01:19:43.945
<v Shane Cahill (Meta)>That means moving at the pace required

1281
01:19:43.945 --> 01:19:47.225
<v Shane Cahill (Meta)>to evaluate risks and validate safeguards,

1282
01:19:47.705 --> 01:19:50.665
<v Shane Cahill (Meta)>not treating safely as a one time check.

1283
01:19:51.305 --> 01:19:53.705
<v Shane Cahill (Meta)>We approach safety holistically

1284
01:19:53.785 --> 01:19:54.985
<v Shane Cahill (Meta)>through a layered approach.

1285
01:19:56.450 --> 01:19:59.650
<v Shane Cahill (Meta)>For example, under our
meta superintelligence

1286
01:19:59.650 --> 01:20:00.930
<v Shane Cahill (Meta)>scaling framework,

1287
01:20:01.090 --> 01:20:03.250
<v Shane Cahill (Meta)>before we release an AI model,

1288
01:20:03.490 --> 01:20:07.650
<v Shane Cahill (Meta)>we evaluate the model for risks and
build safeguards to mitigate them.

1289
01:20:08.385 --> 01:20:11.025
<v Shane Cahill (Meta)>We also work with external experts,

1290
01:20:11.185 --> 01:20:12.545
<v Shane Cahill (Meta)>industry partners,

1291
01:20:12.625 --> 01:20:14.305
<v Shane Cahill (Meta)>and government stakeholders

1292
01:20:14.385 --> 01:20:19.665
<v Shane Cahill (Meta)>as the technology and the science of
evaluating AI continues to evolve.

1293
01:20:20.865 --> 01:20:23.505
<v Shane Cahill (Meta)>We also recently joined other labs

1294
01:20:23.820 --> 01:20:26.060
<v Shane Cahill (Meta)>in signing onto a set of principles

1295
01:20:26.700 --> 01:20:29.180
<v Shane Cahill (Meta)>and that set a floor for safe development,

1296
01:20:29.580 --> 01:20:32.060
<v Shane Cahill (Meta)>including robust internal controls,

1297
01:20:32.460 --> 01:20:34.620
<v Shane Cahill (Meta)>independent external evaluation

1298
01:20:34.620 --> 01:20:37.660
<v Shane Cahill (Meta)>of whether those controls
are operating as intended,

1299
01:20:38.105 --> 01:20:40.505
<v Shane Cahill (Meta)>and independent board committee oversight.

1300
01:20:42.025 --> 01:20:46.985
<v Shane Cahill (Meta)>People will only embrace AI if they have
confidence it works in alignment with

1301
01:20:46.985 --> 01:20:48.265
<v Shane Cahill (Meta)>people's intentions.

1302
01:20:48.825 --> 01:20:50.025
<v Shane Cahill (Meta)>That confidence

1303
01:20:50.265 --> 01:20:54.380
<v Shane Cahill (Meta)>must be earned through rigorous
safety work and transparency.

1304
01:20:55.020 --> 01:20:59.180
<v Shane Cahill (Meta)>We're happy to serve as a resource for
the city as you engage on these important

1305
01:20:59.180 --> 01:20:59.900
<v Shane Cahill (Meta)>issues.

1306
01:21:00.060 --> 01:21:00.780
<v Shane Cahill (Meta)>Thank you.

1307
01:21:02.060 --> 01:21:05.740
<v CM Carmen De La Rosa>Thank you so much. And finally,
we have Alice Friend at Google.

1308
01:21:08.885 --> 01:21:13.445
<v Alice Friend (Google)>Speaker Menon, Chair de la Rosa, and
members of the council, thank you for the

1309
01:21:13.445 --> 01:21:15.445
<v Alice Friend (Google)>opportunity to speak with you today.

1310
01:21:16.005 --> 01:21:22.245
<v Alice Friend (Google)>My name is Alice Friend, and I serve as
director for AI and emerging tech policy at

1311
01:21:22.245 --> 01:21:22.565
<v Alice Friend (Google)>Google.

1312
01:21:23.510 --> 01:21:28.150
<v Alice Friend (Google)>While my appearance at this hearing may be
virtual, please know that New York City

1313
01:21:28.150 --> 01:21:30.310
<v Alice Friend (Google)>is a vital home for Google.

1314
01:21:30.870 --> 01:21:32.710
<v Alice Friend (Google)>With nearly 16,000

1315
01:21:32.710 --> 01:21:35.670
<v Alice Friend (Google)>employees working across our Chelsea,

1316
01:21:35.910 --> 01:21:38.950
<v Alice Friend (Google)>Hudson Square, and Pier 57 campuses,

1317
01:21:39.565 --> 01:21:43.325
<v Alice Friend (Google)>we care deeply about the
safety, economic dynamism,

1318
01:21:43.485 --> 01:21:45.565
<v Alice Friend (Google)>and digital resilience of this city.

1319
01:21:46.445 --> 01:21:51.405
<v Alice Friend (Google)>Our goal at Google has always been to
improve the lives of as many people as

1320
01:21:51.405 --> 01:21:52.925
<v Alice Friend (Google)>possible through technology.

1321
01:21:54.300 --> 01:21:58.140
<v Alice Friend (Google)>Every time people use technology
to solve a real problem,

1322
01:21:58.620 --> 01:22:02.460
<v Alice Friend (Google)>from finding an answer to a
question to screening for disease,

1323
01:22:02.860 --> 01:22:05.740
<v Alice Friend (Google)>we take another step towards
what we set out to achieve.

1324
01:22:06.905 --> 01:22:10.345
<v Alice Friend (Google)>AI is making it possible
to solve economic,

1325
01:22:10.585 --> 01:22:11.705
<v Alice Friend (Google)>scientific,

1326
01:22:11.785 --> 01:22:13.305
<v Alice Friend (Google)>and social problems

1327
01:22:13.465 --> 01:22:14.905
<v Alice Friend (Google)>faster than ever,

1328
01:22:15.465 --> 01:22:17.945
<v Alice Friend (Google)>but we know that it
poses challenges as well.

1329
01:22:19.350 --> 01:22:23.110
<v Alice Friend (Google)>To unlock the benefits and
address the challenges of AI,

1330
01:22:23.270 --> 01:22:26.070
<v Alice Friend (Google)>we take a bold but responsible approach,

1331
01:22:26.470 --> 01:22:27.030
<v Alice Friend (Google)>emphasizing

1332
01:22:27.350 --> 01:22:32.495
<v Alice Friend (Google)>cutting edge research to guide our decision
making and inform the development of our

1333
01:22:32.495 --> 01:22:34.815
<v Alice Friend (Google)>AI safety and security approaches.

1334
01:22:35.535 --> 01:22:40.495
<v Alice Friend (Google)>We are also committed to publishing this
research and to being transparent about our

1335
01:22:40.495 --> 01:22:41.295
<v Alice Friend (Google)>practices.

1336
01:22:42.335 --> 01:22:45.520
<v Alice Friend (Google)>Our internal governance
focuses on responsibility

1337
01:22:45.520 --> 01:22:48.240
<v Alice Friend (Google)>throughout the AI development life cycle,

1338
01:22:48.640 --> 01:22:50.240
<v Alice Friend (Google)>covering model development,

1339
01:22:50.320 --> 01:22:51.920
<v Alice Friend (Google)>application deployment,

1340
01:22:52.240 --> 01:22:54.080
<v Alice Friend (Google)>and post launch monitoring.

1341
01:22:54.720 --> 01:22:59.285
<v Alice Friend (Google)>We identify and assess AI risks
through a broad range of approaches,

1342
01:22:59.365 --> 01:23:00.885
<v Alice Friend (Google)>which include research,

1343
01:23:01.125 --> 01:23:04.245
<v Alice Friend (Google)>input from internal and external experts,

1344
01:23:04.645 --> 01:23:06.325
<v Alice Friend (Google)>and adversarial testing.

1345
01:23:07.205 --> 01:23:11.365
<v Alice Friend (Google)>We evaluate our models and systems
against benchmarks for safety,

1346
01:23:11.920 --> 01:23:12.800
<v Alice Friend (Google)>privacy,

1347
01:23:12.800 --> 01:23:13.760
<v Alice Friend (Google)>and security,

1348
01:23:14.160 --> 01:23:18.080
<v Alice Friend (Google)>and we build mitigations using
techniques like safety fine tuning,

1349
01:23:18.320 --> 01:23:20.000
<v Alice Friend (Google)>out of model filters,

1350
01:23:20.240 --> 01:23:22.320
<v Alice Friend (Google)>and robust provenance solutions.

1351
01:23:23.120 --> 01:23:24.480
<v Alice Friend (Google)>Since 2019,

1352
01:23:24.480 --> 01:23:29.775
<v Alice Friend (Google)>we have published annual AI progress reports
to show... To share how we apply these

1353
01:23:29.775 --> 01:23:34.095
<v Alice Friend (Google)>responsible practices to our AI
technology development and deployment.

1354
01:23:35.215 --> 01:23:38.015
<v Alice Friend (Google)>We are at a pivotal
moment in AI development.

1355
01:23:38.655 --> 01:23:40.815
<v Alice Friend (Google)>Recent advances in the capabilities

1356
01:23:40.920 --> 01:23:44.280
<v Alice Friend (Google)>of Frontier AI models in
areas like cybersecurity

1357
01:23:44.280 --> 01:23:45.480
<v Alice Friend (Google)>and biology

1358
01:23:45.640 --> 01:23:49.640
<v Alice Friend (Google)>have highlighted the need for
appropriate safeguards and protocols

1359
01:23:49.640 --> 01:23:51.720
<v Alice Friend (Google)>for the most advanced models.

1360
01:23:52.360 --> 01:23:54.345
<v Alice Friend (Google)>In addition, the widespread

1361
01:23:54.345 --> 01:23:58.425
<v Alice Friend (Google)>adoption of AI tools has raised
concerns about consumer well-being.

1362
01:23:58.985 --> 01:24:03.625
<v Alice Friend (Google)>To confront these challenges, Google has
long called for an approach to AI that is

1363
01:24:03.625 --> 01:24:05.385
<v Alice Friend (Google)>both bold and responsible,

1364
01:24:05.625 --> 01:24:07.625
<v Alice Friend (Google)>and we strongly support a pragmatic,

1365
01:24:08.170 --> 01:24:11.210
<v Alice Friend (Google)>technology neutral, and
evidence based regulatory

1366
01:24:11.210 --> 01:24:12.089
<v Alice Friend (Google)>framework.

1367
01:24:12.810 --> 01:24:17.690
<v Alice Friend (Google)>Because the risks associated with the most
advanced AI models often touch on issues

1368
01:24:17.690 --> 01:24:19.609
<v Alice Friend (Google)>affecting national security,

1369
01:24:20.010 --> 01:24:23.985
<v Alice Friend (Google)>our view is that the country would be
best served by a comprehensive framework

1370
01:24:23.985 --> 01:24:25.825
<v Alice Friend (Google)>enacted at the federal level.

1371
01:24:26.705 --> 01:24:29.585
<v Alice Friend (Google)>We also support the
establishment of an independent

1372
01:24:29.585 --> 01:24:33.265
<v Alice Friend (Google)>industry funded body that could
operate under federal oversight

1373
01:24:33.505 --> 01:24:38.385
<v Alice Friend (Google)>and establish guardrails to protect the
public based on widely adopted technical

1374
01:24:38.385 --> 01:24:38.785
<v Alice Friend (Google)>standards.

1375
01:24:39.940 --> 01:24:43.620
<v Alice Friend (Google)>Examples of such organizations and other
industries include the North American

1376
01:24:43.620 --> 01:24:45.780
<v Alice Friend (Google)>Electric Reliability Corporation

1377
01:24:45.940 --> 01:24:48.180
<v Alice Friend (Google)>and the American Medical Association.

1378
01:24:49.380 --> 01:24:51.940
<v Alice Friend (Google)>For widespread consumer AI applications

1379
01:24:51.940 --> 01:24:52.900
<v Alice Friend (Google)>like chatbots,

1380
01:24:53.315 --> 01:24:58.115
<v Alice Friend (Google)>we believe governments around the country
can draw on and, in some cases, amend

1381
01:24:58.115 --> 01:24:58.915
<v Alice Friend (Google)>existing

1382
01:24:58.915 --> 01:24:59.715
<v Alice Friend (Google)>laws

1383
01:24:59.875 --> 01:25:04.195
<v Alice Friend (Google)>and rules to address real world
outputs and specific harms.

1384
01:25:04.835 --> 01:25:10.550
<v Alice Friend (Google)>We believe that if something is illegal
without AI, it is still illegal with AI.

1385
01:25:11.350 --> 01:25:14.950
<v Alice Friend (Google)>We share the council's commitment
to safety, transparency,

1386
01:25:15.030 --> 01:25:16.230
<v Alice Friend (Google)>and public trust.

1387
01:25:16.790 --> 01:25:21.110
<v Alice Friend (Google)>To achieve this, the most effective
safeguards are anchored in national technical

1388
01:25:21.110 --> 01:25:21.750
<v Alice Friend (Google)>standards

1389
01:25:21.915 --> 01:25:25.275
<v Alice Friend (Google)>and harmonized across
state and federal levels.

1390
01:25:26.075 --> 01:25:30.635
<v Alice Friend (Google)>As we prepare to work with New York State's
new DIGIT office under the RAISE Act to

1391
01:25:30.635 --> 01:25:32.395
<v Alice Friend (Google)>address catastrophic risks,

1392
01:25:32.715 --> 01:25:35.675
<v Alice Friend (Google)>aligning local policies
with broader frameworks

1393
01:25:35.960 --> 01:25:40.840
<v Alice Friend (Google)>rather than creating a fragmented patchwork
is essential to keeping New York City a

1394
01:25:40.840 --> 01:25:42.840
<v Alice Friend (Google)>premier global hub for innovation.

1395
01:25:43.800 --> 01:25:47.880
<v Alice Friend (Google)>Finally, I would like to note that to ensure
our formal submission to the council is

1396
01:25:47.880 --> 01:25:49.800
<v Alice Friend (Google)>as thorough and responsive as possible,

1397
01:25:50.305 --> 01:25:54.945
<v Alice Friend (Google)>We will submit our finalized written
testimony for the record shortly after today's

1398
01:25:54.945 --> 01:25:59.585
<v Alice Friend (Google)>hearing so we can additionally address
questions and feedback raised by the council

1399
01:25:59.585 --> 01:26:00.305
<v Alice Friend (Google)>today.

1400
01:26:00.785 --> 01:26:04.705
<v Alice Friend (Google)>Thank you again for the opportunity to
testify at this important hearing. I look

1401
01:26:04.705 --> 01:26:05.585
<v Alice Friend (Google)>forward to your questions.

1402
01:26:06.840 --> 01:26:10.760
<v Speaker Julie Menin>Okay. Thank you very much to this panel.
So I'm gonna begin with questions, and

1403
01:26:10.760 --> 01:26:14.840
<v Speaker Julie Menin>then, I will turn it over to the technology
chair for her questions. And then we're

1404
01:26:14.840 --> 01:26:17.560
<v Speaker Julie Menin>gonna open it up to all the
members for their questions.

1405
01:26:18.245 --> 01:26:21.445
<v Speaker Julie Menin>So to this panel, we just heard incredibly

1406
01:26:21.445 --> 01:26:24.805
<v Speaker Julie Menin>compelling testimony
from our whistleblower

1407
01:26:24.885 --> 01:26:25.605
<v Speaker Julie Menin>panel.

1408
01:26:25.765 --> 01:26:30.805
<v Speaker Julie Menin>As you all four are under oath, we
wanna hear how each of you quantify

1409
01:26:31.420 --> 01:26:33.740
<v Speaker Julie Menin>the risk imposed by AI

1410
01:26:33.900 --> 01:26:36.859
<v Speaker Julie Menin>in the worst case catastrophic

1411
01:26:36.940 --> 01:26:37.899
<v Speaker Julie Menin>scenario.

1412
01:26:37.900 --> 01:26:39.980
<v Speaker Julie Menin>I'm gonna start with you, miss Dwyer.

1413
01:26:44.385 --> 01:26:48.065
<v Morgan Dwyer (OpenAI)>I don't know. I also
don't think it matters

1414
01:26:48.385 --> 01:26:54.465
<v Morgan Dwyer (OpenAI)>whether it's 1% or 10% or a 20%
chance that something catastrophic

1415
01:26:54.465 --> 01:26:55.505
<v Morgan Dwyer (OpenAI)>will go wrong.

1416
01:26:55.825 --> 01:26:58.385
<v Morgan Dwyer (OpenAI)>None of these levels is
remotely acceptable.

1417
01:26:59.280 --> 01:27:01.440
<v Morgan Dwyer (OpenAI)>We should not train models

1418
01:27:01.520 --> 01:27:04.400
<v Morgan Dwyer (OpenAI)>that we cannot make an extremely strong

1419
01:27:04.400 --> 01:27:07.680
<v Morgan Dwyer (OpenAI)>case that we can keep under human control.

1420
01:27:08.240 --> 01:27:12.080
<v Speaker Julie Menin>I'm sorry. But to say you don't
know and it doesn't matter

1421
01:27:12.240 --> 01:27:13.360
<v Speaker Julie Menin>is flippant

1422
01:27:13.360 --> 01:27:14.000
<v Speaker Julie Menin>at best.

1423
01:27:14.565 --> 01:27:17.445
<v Speaker Julie Menin>This idea that if you're a pharmaceutical

1424
01:27:17.445 --> 01:27:18.165
<v Speaker Julie Menin>company

1425
01:27:18.405 --> 01:27:22.485
<v Speaker Julie Menin>and you're developing a drug and you say,
I don't know if it's gonna kill people,

1426
01:27:22.805 --> 01:27:26.485
<v Speaker Julie Menin>I I just... I I I honestly am
incredulous at that answer.

1427
01:27:29.930 --> 01:27:31.450
<v Morgan Dwyer (OpenAI)>I would say that

1428
01:27:32.170 --> 01:27:37.770
<v Morgan Dwyer (OpenAI)>the percentage chance doesn't matter.
What matters is the commitment to safety,

1429
01:27:38.490 --> 01:27:44.250
<v Morgan Dwyer (OpenAI)>and that is what OpenAI is committed to
doing. We evaluate our models for a wide

1430
01:27:44.250 --> 01:27:45.495
<v Morgan Dwyer (OpenAI)>range of threats

1431
01:27:45.655 --> 01:27:47.015
<v Morgan Dwyer (OpenAI)>from cybersecurity

1432
01:27:47.015 --> 01:27:48.535
<v Morgan Dwyer (OpenAI)>to child safety.

1433
01:27:48.775 --> 01:27:52.215
<v Morgan Dwyer (OpenAI)>We work with a robust set
of third party evaluators

1434
01:27:52.215 --> 01:27:53.655
<v Morgan Dwyer (OpenAI)>and red teamers,

1435
01:27:54.135 --> 01:27:58.615
<v Morgan Dwyer (OpenAI)>and we work with the federal government
to assess those national security and

1436
01:27:58.615 --> 01:27:59.895
<v Morgan Dwyer (OpenAI)>catastrophic risks

1437
01:28:00.470 --> 01:28:01.990
<v Morgan Dwyer (OpenAI)>that you are referencing.

1438
01:28:02.390 --> 01:28:07.990
<v Morgan Dwyer (OpenAI)>So we take safety very seriously, and our
focus on... Is on ensuring that when we

1439
01:28:07.990 --> 01:28:11.190
<v Morgan Dwyer (OpenAI)>release a model into the world,
we believe it to be safe.

1440
01:28:11.670 --> 01:28:16.425
<v Speaker Julie Menin>So a couple comments on that. If you can't
quantify what the safety risk is, how do

1441
01:28:16.425 --> 01:28:18.105
<v Speaker Julie Menin>you know that the product is safe?

1442
01:28:18.425 --> 01:28:21.625
<v Speaker Julie Menin>And then secondly, I wanna
ask you, in particular,

1443
01:28:22.105 --> 01:28:23.945
<v Speaker Julie Menin>on 10/01/2026,

1444
01:28:23.945 --> 01:28:27.950
<v Speaker Julie Menin>The Wall Street Journal reported that OpenAI
terminated three members of its safety

1445
01:28:27.950 --> 01:28:29.950
<v Speaker Julie Menin>team allegedly for their access,

1446
01:28:30.190 --> 01:28:31.150
<v Speaker Julie Menin>mishandling,

1447
01:28:31.150 --> 01:28:35.310
<v Speaker Julie Menin>and or sharing confidential company
information with a third party AI safety

1448
01:28:35.310 --> 01:28:36.190
<v Speaker Julie Menin>organization.

1449
01:28:36.590 --> 01:28:40.955
<v Speaker Julie Menin>OpenAI reportedly stated that the employees
violate company policies governing

1450
01:28:40.955 --> 01:28:44.315
<v Speaker Julie Menin>access to and handling of
sensitive information.

1451
01:28:44.315 --> 01:28:48.715
<v Speaker Julie Menin>Critics, meanwhile, question whether the
dismissals could discourage employees from

1452
01:28:48.715 --> 01:28:51.915
<v Speaker Julie Menin>raising safety concerns. So
why were these employees fired?

1453
01:28:55.360 --> 01:29:01.200
<v Morgan Dwyer (OpenAI)>I was not involved in that decision. I
don't work in HR or on those employees'

1454
01:29:01.200 --> 01:29:01.840
<v Morgan Dwyer (OpenAI)>teams.

1455
01:29:02.240 --> 01:29:06.400
<v Morgan Dwyer (OpenAI)>What I do know is that OpenAI
has strong internal policies,

1456
01:29:06.720 --> 01:29:09.040
<v Morgan Dwyer (OpenAI)>that protect against retaliation

1457
01:29:09.040 --> 01:29:11.834
<v Morgan Dwyer (OpenAI)>for employees that report safety concerns.

1458
01:29:12.315 --> 01:29:15.355
<v Morgan Dwyer (OpenAI)>I also know that we respect employees'

1459
01:29:15.355 --> 01:29:15.995
<v Morgan Dwyer (OpenAI)>rights,

1460
01:29:16.394 --> 01:29:20.714
<v Morgan Dwyer (OpenAI)>to communicate safety concerns to
relevant government authorities.

1461
01:29:20.955 --> 01:29:25.570
<v Speaker Julie Menin>So just to be clear on what I'm asking,
I'm asking directly whether safety related

1462
01:29:25.570 --> 01:29:28.610
<v Speaker Julie Menin>advocacy played any role
in these terminations.

1463
01:29:30.370 --> 01:29:31.410
<v Morgan Dwyer (OpenAI)>I don't know.

1464
01:29:31.890 --> 01:29:35.090
<v Morgan Dwyer (OpenAI)>I know that we have
strong internal policies,

1465
01:29:35.410 --> 01:29:36.130
<v Morgan Dwyer (OpenAI)>that protect

1466
01:29:36.535 --> 01:29:38.615
<v Morgan Dwyer (OpenAI)>employees from retaliation

1467
01:29:38.615 --> 01:29:40.855
<v Morgan Dwyer (OpenAI)>if they raise safety concerns,

1468
01:29:41.095 --> 01:29:42.135
<v Morgan Dwyer (OpenAI)>to leadership.

1469
01:29:43.015 --> 01:29:45.975
<v Speaker Julie Menin>I'm gonna move on to the other
companies. Same question.

1470
01:29:46.295 --> 01:29:50.535
<v Speaker Julie Menin>The first question that I asked, which
is, we wanna hear how each of you quantify

1471
01:29:50.535 --> 01:29:54.140
<v Speaker Julie Menin>the risks imposed by AI in the
worst catastrophic scenario.

1472
01:29:54.540 --> 01:29:56.540
<v Speaker Julie Menin>So I'm gonna go to,

1473
01:29:56.860 --> 01:29:59.820
<v Speaker Julie Menin>mister Graham. Are you
able to quantify the risk?

1474
01:30:01.100 --> 01:30:03.580
<v Logan Graham (Anthropic)>From the beginning of our
safety work at Anthropic,

1475
01:30:04.125 --> 01:30:07.885
<v Logan Graham (Anthropic)>our fundamental belief is that
if not properly safeguarded,

1476
01:30:08.125 --> 01:30:13.405
<v Logan Graham (Anthropic)>the risk might be too high. And so I built
and I've led the team at Anthropic that

1477
01:30:13.405 --> 01:30:16.285
<v Logan Graham (Anthropic)>did some of the first thinking
around how we quantify

1478
01:30:16.605 --> 01:30:21.420
<v Logan Graham (Anthropic)>risks like this. For the past four years,
we focused on issues like biological

1479
01:30:21.420 --> 01:30:22.300
<v Logan Graham (Anthropic)>security

1480
01:30:22.620 --> 01:30:23.900
<v Logan Graham (Anthropic)>and cybersecurity.

1481
01:30:24.380 --> 01:30:29.100
<v Logan Graham (Anthropic)>And the actual process of quantifying those
risks is intricate and nuanced, and we

1482
01:30:29.100 --> 01:30:30.060
<v Logan Graham (Anthropic)>learn more every year.

1483
01:30:30.775 --> 01:30:35.815
<v Logan Graham (Anthropic)>And we detail quite a lot of it in documents
like our responsible scaling policy,

1484
01:30:35.815 --> 01:30:38.295
<v Logan Graham (Anthropic)>how we think about the
threat models of this,

1485
01:30:38.935 --> 01:30:42.295
<v Logan Graham (Anthropic)>all the way up to notions of loss
of control and misalignment,

1486
01:30:42.790 --> 01:30:46.310
<v Logan Graham (Anthropic)>which from the beginning of the
company, we have tried to lead,

1487
01:30:46.630 --> 01:30:48.710
<v Logan Graham (Anthropic)>the world in the research of.

1488
01:30:49.270 --> 01:30:53.510
<v Logan Graham (Anthropic)>So what I do know today
is I feel relatively,

1489
01:30:54.470 --> 01:30:59.055
<v Logan Graham (Anthropic)>sort of more optimistic the industry has
moved quite fast to take cybersecurity

1490
01:30:59.055 --> 01:31:01.214
<v Logan Graham (Anthropic)>seriously and biological security.

1491
01:31:01.375 --> 01:31:02.095
<v Logan Graham (Anthropic)>I think

1492
01:31:02.495 --> 01:31:05.695
<v Logan Graham (Anthropic)>documents and policies like
responsible scaling policies

1493
01:31:05.775 --> 01:31:08.175
<v Logan Graham (Anthropic)>have so far been fairly effective

1494
01:31:08.575 --> 01:31:10.254
<v Logan Graham (Anthropic)>at taking a cautious approach.

1495
01:31:10.894 --> 01:31:11.695
<v Logan Graham (Anthropic)>However,

1496
01:31:11.695 --> 01:31:13.190
<v Logan Graham (Anthropic)>now we have

1497
01:31:13.830 --> 01:31:18.390
<v Logan Graham (Anthropic)>bigger and bigger threats that might come
if we don't properly safeguard them as the

1498
01:31:18.390 --> 01:31:20.150
<v Logan Graham (Anthropic)>models get more and more capable.

1499
01:31:20.390 --> 01:31:22.790
<v Logan Graham (Anthropic)>And so what we do to to mitigate that

1500
01:31:23.110 --> 01:31:26.790
<v Logan Graham (Anthropic)>is, number one, put in tremendous amounts
of effort across a number of different

1501
01:31:26.790 --> 01:31:27.430
<v Logan Graham (Anthropic)>teams,

1502
01:31:27.670 --> 01:31:30.765
<v Logan Graham (Anthropic)>our own safeguards teams,
our research teams, my team,

1503
01:31:31.245 --> 01:31:34.285
<v Logan Graham (Anthropic)>to try to to quantify and mitigate this.

1504
01:31:34.765 --> 01:31:38.525
<v Logan Graham (Anthropic)>And it's increasingly important for us to
be dramatically more transparent about

1505
01:31:38.525 --> 01:31:42.365
<v Logan Graham (Anthropic)>what we think about these risks. But
on that note, as the models increase

1506
01:31:42.660 --> 01:31:43.700
<v Logan Graham (Anthropic)>exponentially,

1507
01:31:43.780 --> 01:31:48.340
<v Speaker Julie Menin>how can you ensure the public that the safety
protocols are increasing exponentially

1508
01:31:48.340 --> 01:31:49.140
<v Speaker Julie Menin>as well?

1509
01:31:50.660 --> 01:31:53.220
<v Logan Graham (Anthropic)>We share the same concern.

1510
01:31:53.700 --> 01:31:57.060
<v Logan Graham (Anthropic)>I think there are a number of
really important things to do here.

1511
01:31:58.065 --> 01:32:00.545
<v Logan Graham (Anthropic)>The first thing is transparency

1512
01:32:00.625 --> 01:32:03.105
<v Logan Graham (Anthropic)>about what we think about safety,

1513
01:32:03.345 --> 01:32:07.985
<v Logan Graham (Anthropic)>the evidence that we have, being
transparent about disclosure of incidents.

1514
01:32:08.705 --> 01:32:12.465
<v Logan Graham (Anthropic)>But in addition, you've seen us
publicly call for pacing the frontier.

1515
01:32:12.810 --> 01:32:15.450
<v Logan Graham (Anthropic)>If the models are moving very, very fast,

1516
01:32:15.690 --> 01:32:20.810
<v Logan Graham (Anthropic)>we think that we might benefit from
slowing down to have more time for safety

1517
01:32:20.810 --> 01:32:21.610
<v Logan Graham (Anthropic)>processes.

1518
01:32:22.090 --> 01:32:27.275
<v Logan Graham (Anthropic)>I think these two together as well as
continuing to what we call race to the top and

1519
01:32:27.275 --> 01:32:30.635
<v Logan Graham (Anthropic)>ever strengthen our approaches
like responsible scaling policies

1520
01:32:30.715 --> 01:32:33.835
<v Logan Graham (Anthropic)>are are the most important
things to do right now.

1521
01:32:34.235 --> 01:32:37.675
<v Speaker Julie Menin>So, I mean, you say that you wanna be
more transparent, but you're not really

1522
01:32:37.675 --> 01:32:40.955
<v Speaker Julie Menin>addressing the question. I mean,
the question is, can you quantify

1523
01:32:41.590 --> 01:32:44.150
<v Speaker Julie Menin>the risk of something cataclysmic

1524
01:32:44.150 --> 01:32:48.310
<v Speaker Julie Menin>happening? And I do wanna say for all four
companies, and I know you're not in the

1525
01:32:48.310 --> 01:32:50.310
<v Speaker Julie Menin>chambers, but we do have a chart.

1526
01:32:50.630 --> 01:32:53.190
<v Speaker Julie Menin>In e... For each of your companies had

1527
01:32:53.350 --> 01:32:54.230
<v Speaker Julie Menin>instances

1528
01:32:54.390 --> 01:33:00.345
<v Speaker Julie Menin>where agents went rogue. And in each of
those instances, it is not as if any of you

1529
01:33:00.345 --> 01:33:03.465
<v Speaker Julie Menin>notified the public when those occurred.

1530
01:33:03.545 --> 01:33:08.105
<v Speaker Julie Menin>In some instances, it was months after the
fact when the public was even notified.

1531
01:33:08.505 --> 01:33:12.425
<v Speaker Julie Menin>And in one instance, for the government
of Australia, a general email was sent to

1532
01:33:12.425 --> 01:33:16.110
<v Speaker Julie Menin>them. So this doesn't exactly
increase transparency.

1533
01:33:24.990 --> 01:33:27.230
<v Speaker Julie Menin>Mister Graham, do you want to... Thank

1534
01:33:27.230 --> 01:33:27.390
<v Speaker Julie Menin>you.

1535
01:33:28.175 --> 01:33:29.135
<v Logan Graham (Anthropic)>Yeah. First.

1536
01:33:30.415 --> 01:33:33.135
<v Logan Graham (Anthropic)>In efforts to sort of
quantify these risks,

1537
01:33:33.455 --> 01:33:39.455
<v Logan Graham (Anthropic)>we have a number of, you know, quite
detailed reports that try to elucidate

1538
01:33:39.455 --> 01:33:41.215
<v Logan Graham (Anthropic)>our entire thinking around this.

1539
01:33:42.160 --> 01:33:43.520
<v Logan Graham (Anthropic)>We increasingly

1540
01:33:43.520 --> 01:33:44.320
<v Logan Graham (Anthropic)>do

1541
01:33:44.720 --> 01:33:49.440
<v Logan Graham (Anthropic)>risk reports in addition to our normal
responsible scaling policies where we outline

1542
01:33:49.440 --> 01:33:53.280
<v Logan Graham (Anthropic)>notions of our threat models and how likely
we think certain types of these threats

1543
01:33:53.280 --> 01:33:56.305
<v Logan Graham (Anthropic)>are likely to come to pass.
And then we look backwards

1544
01:33:56.705 --> 01:33:58.945
<v Logan Graham (Anthropic)>at things that are happening
within our own company,

1545
01:33:59.665 --> 01:34:04.065
<v Logan Graham (Anthropic)>and we ask what is the likelihood that
these threats might come to pass in the

1546
01:34:04.065 --> 01:34:06.145
<v Logan Graham (Anthropic)>future, including forecasting,

1547
01:34:06.870 --> 01:34:07.830
<v Logan Graham (Anthropic)>surveying

1548
01:34:07.830 --> 01:34:09.670
<v Logan Graham (Anthropic)>researchers at Anthropic

1549
01:34:09.990 --> 01:34:13.990
<v Logan Graham (Anthropic)>as well. And we we publish this information.
We are... Excuse me. Are those reports

1550
01:34:13.990 --> 01:34:16.150
<v Logan Graham (Anthropic)>publicly available that you referenced?

1551
01:34:16.790 --> 01:34:20.790
<v Logan Graham (Anthropic)>They are. Yes. We we report
them not just in our,

1552
01:34:21.350 --> 01:34:25.645
<v Logan Graham (Anthropic)>what we call, risk reports, but also
when we launch models, we launch,

1553
01:34:25.965 --> 01:34:29.085
<v Logan Graham (Anthropic)>we release detailed multi
100 page reports of these.

1554
01:34:30.445 --> 01:34:32.925
<v Speaker Julie Menin>And then I'm now gonna turn to mister

1555
01:34:33.085 --> 01:34:37.645
<v Speaker Julie Menin>Cahill for to... For the response to that
first question on quantifying the risk.

1556
01:34:40.460 --> 01:34:42.700
<v Shane Cahill (Meta)>Speaker, thank you very
much for the question.

1557
01:34:43.180 --> 01:34:47.340
<v Shane Cahill (Meta)>First off, we're committed at
Meta to building AI safely.

1558
01:34:47.900 --> 01:34:52.380
<v Shane Cahill (Meta)>Now we have a multilayered approach to this
where teams evaluate before deployment,

1559
01:34:52.540 --> 01:34:53.660
<v Shane Cahill (Meta)>including adversarial

1560
01:34:53.660 --> 01:34:54.060
<v Shane Cahill (Meta)>testing,

1561
01:34:54.855 --> 01:34:56.535
<v Shane Cahill (Meta)>applying safeguards,

1562
01:34:56.695 --> 01:34:59.575
<v Shane Cahill (Meta)>and deploying when only
risks are mitigated.

1563
01:35:00.295 --> 01:35:02.295
<v Shane Cahill (Meta)>We set this out in our public,

1564
01:35:02.615 --> 01:35:06.935
<v Shane Cahill (Meta)>Meta Superintelligence Scaling Framework
and our preparedness reports that go

1565
01:35:06.935 --> 01:35:07.815
<v Shane Cahill (Meta)>alongside

1566
01:35:07.815 --> 01:35:08.935
<v Shane Cahill (Meta)>the release of models.

1567
01:35:15.390 --> 01:35:16.750
<v Speaker Julie Menin>So can

1568
01:35:16.750 --> 01:35:20.430
<v Speaker Julie Menin>you quantify what the risk is,
or you're not able to do that?

1569
01:35:23.905 --> 01:35:28.625
<v Shane Cahill (Meta)>Speaker, I don't wish to be imprecise
in relation to the the question of of

1570
01:35:28.625 --> 01:35:29.585
<v Shane Cahill (Meta)>quantifying.

1571
01:35:30.145 --> 01:35:33.745
<v Shane Cahill (Meta)>I don't have our framework in in front
of me right now, but I'd be pleased to

1572
01:35:33.745 --> 01:35:34.865
<v Shane Cahill (Meta)>follow-up with you afterwards.

1573
01:35:36.540 --> 01:35:38.940
<v Speaker Julie Menin>Okay. It it... It's just

1574
01:35:40.540 --> 01:35:45.899
<v Speaker Julie Menin>given the serious safety risks that have
been raised here to not be able to quantify

1575
01:35:46.140 --> 01:35:50.460
<v Speaker Julie Menin>what the risk is is troubling
at best. I'm gonna now move on,

1576
01:35:50.780 --> 01:35:52.605
<v Speaker Julie Menin>lastly to miss Friend.

1577
01:35:54.445 --> 01:35:55.645
<v Alice Friend (Google)>Madam speaker,

1578
01:35:56.125 --> 01:35:58.285
<v Alice Friend (Google)>at Google, we take catastrophic

1579
01:35:58.285 --> 01:36:00.045
<v Alice Friend (Google)>risks and their possibility

1580
01:36:00.045 --> 01:36:01.645
<v Alice Friend (Google)>extremely seriously,

1581
01:36:02.365 --> 01:36:08.260
<v Alice Friend (Google)>and we address that in a few ways. One
way is that we perform research into

1582
01:36:08.260 --> 01:36:12.820
<v Alice Friend (Google)>producing our own safety frameworks for
artificial general intelligence. So we have

1583
01:36:12.820 --> 01:36:15.300
<v Alice Friend (Google)>an AGI safety and security approach.

1584
01:36:15.620 --> 01:36:19.300
<v Alice Friend (Google)>We also have a frontier safety
framework, which is our protocols

1585
01:36:19.985 --> 01:36:23.185
<v Alice Friend (Google)>for monitoring our
models under development

1586
01:36:23.345 --> 01:36:25.345
<v Alice Friend (Google)>for dangerous capabilities.

1587
01:36:25.425 --> 01:36:27.985
<v Alice Friend (Google)>Those protocols then point to,

1588
01:36:28.625 --> 01:36:33.745
<v Alice Friend (Google)>safety mitigations that we must take if we
in fact cross what we call our critical

1589
01:36:33.745 --> 01:36:34.865
<v Alice Friend (Google)>capability levels.

1590
01:36:36.090 --> 01:36:40.170
<v Alice Friend (Google)>Part of our testing and evaluation
is of course using commonly,

1591
01:36:40.810 --> 01:36:43.130
<v Alice Friend (Google)>used benchmarks across the industry.

1592
01:36:43.450 --> 01:36:46.490
<v Alice Friend (Google)>Those benchmarks do
provide some quantification

1593
01:36:46.490 --> 01:36:48.890
<v Alice Friend (Google)>of the capabilities of
the models themselves.

1594
01:36:49.675 --> 01:36:54.235
<v Alice Friend (Google)>But when it comes to forecasting future
catastrophic risk, it's not a perfect

1595
01:36:54.235 --> 01:36:55.755
<v Alice Friend (Google)>science at this stage.

1596
01:36:56.315 --> 01:37:01.195
<v Alice Friend (Google)>And in this particular case, academics have
noted that there is no reference class

1597
01:37:01.195 --> 01:37:04.075
<v Alice Friend (Google)>on which to base
quantitative probabilities.

1598
01:37:04.740 --> 01:37:08.980
<v Alice Friend (Google)>So we all have to be very humble
about making such quantitative claims,

1599
01:37:09.460 --> 01:37:12.900
<v Alice Friend (Google)>because there isn't really a
rigorous scientific way to do those

1600
01:37:16.985 --> 01:37:20.905
<v Speaker Julie Menin>I'm gonna take it then that neither of
the four of you... No company here can

1601
01:37:20.905 --> 01:37:24.024
<v Speaker Julie Menin>quantify the risk of something
cataclysmic happening.

1602
01:37:24.345 --> 01:37:24.825
<v Speaker Julie Menin>Okay.

1603
01:37:25.145 --> 01:37:29.820
<v Speaker Julie Menin>I'm going to move on. Sam Altman said
yesterday that we should accept that, quote,

1604
01:37:29.820 --> 01:37:32.700
<v Speaker Julie Menin>the world should accept
some bad things happening

1605
01:37:32.860 --> 01:37:35.340
<v Speaker Julie Menin>as a price of advancing this technology.

1606
01:37:36.220 --> 01:37:37.900
<v Speaker Julie Menin>No one is questioning

1607
01:37:37.900 --> 01:37:41.420
<v Speaker Julie Menin>that the incredible medical and scientific

1608
01:37:41.420 --> 01:37:42.380
<v Speaker Julie Menin>breakthroughs

1609
01:37:42.380 --> 01:37:46.575
<v Speaker Julie Menin>and innovation that they... That
AI has brought to this world.

1610
01:37:46.815 --> 01:37:49.055
<v Speaker Julie Menin>But why should the public accept

1611
01:37:49.135 --> 01:37:51.375
<v Speaker Julie Menin>that bad things should be happening,

1612
01:37:51.535 --> 01:37:56.015
<v Speaker Julie Menin>and why should AI companies get to decide
what level of risk society must bear?

1613
01:37:59.010 --> 01:38:02.210
<v Speaker Julie Menin>I'm gonna go in order. I'm gonna
start with you, miss Dwyer, please.

1614
01:38:04.370 --> 01:38:08.370
<v Morgan Dwyer (OpenAI)>So we think AI has tremendous
potential benefits

1615
01:38:08.370 --> 01:38:11.170
<v Morgan Dwyer (OpenAI)>to help solve some of
humanity's hardest problems,

1616
01:38:11.825 --> 01:38:14.225
<v Morgan Dwyer (OpenAI)>But no technology is without risks,

1617
01:38:14.385 --> 01:38:16.865
<v Morgan Dwyer (OpenAI)>and there are a wide spectrum of risks.

1618
01:38:17.345 --> 01:38:18.145
<v Morgan Dwyer (OpenAI)>Clearly,

1619
01:38:18.465 --> 01:38:20.305
<v Morgan Dwyer (OpenAI)>some risks are unacceptable,

1620
01:38:20.945 --> 01:38:26.065
<v Morgan Dwyer (OpenAI)>but we also think it's important to get
technology into the hands of as many people

1621
01:38:26.420 --> 01:38:29.220
<v Morgan Dwyer (OpenAI)>as possible so that they
can benefit from it.

1622
01:38:30.020 --> 01:38:34.660
<v Morgan Dwyer (OpenAI)>And that's why we are focused on safety.
We train our models to be safe from the

1623
01:38:34.660 --> 01:38:35.220
<v Morgan Dwyer (OpenAI)>start.

1624
01:38:35.700 --> 01:38:38.900
<v Morgan Dwyer (OpenAI)>We evaluate them against
a robust set of risks,

1625
01:38:39.325 --> 01:38:43.885
<v Morgan Dwyer (OpenAI)>and we work with a large network
of third party evaluators.

1626
01:38:44.285 --> 01:38:47.805
<v Morgan Dwyer (OpenAI)>We've also supported
regulation that requires,

1627
01:38:48.285 --> 01:38:51.405
<v Morgan Dwyer (OpenAI)>third party audits of
our safety frameworks

1628
01:38:52.130 --> 01:38:54.290
<v Morgan Dwyer (OpenAI)>as well as regulation

1629
01:38:54.290 --> 01:38:59.330
<v Morgan Dwyer (OpenAI)>that focuses on youth safety and holds
us accountable for keeping teens safe.

1630
01:39:00.370 --> 01:39:03.169
<v Speaker Julie Menin>I I do have a follow-up to that because

1631
01:39:03.170 --> 01:39:06.770
<v Speaker Julie Menin>it it it it does seem to stretch credulity

1632
01:39:07.795 --> 01:39:12.515
<v Speaker Julie Menin>that the president of your company
and his wife each gave $12,500,000

1633
01:39:12.515 --> 01:39:14.755
<v Speaker Julie Menin>to a PAC that opposed candidates

1634
01:39:14.835 --> 01:39:16.835
<v Speaker Julie Menin>to the tune of about $50,000,000

1635
01:39:16.835 --> 01:39:19.235
<v Speaker Julie Menin>due to their support for AI regulation.

1636
01:39:20.720 --> 01:39:24.000
<v Speaker Julie Menin>So if the company is so
committed to AI regulation,

1637
01:39:24.640 --> 01:39:26.720
<v Speaker Julie Menin>why would that pack exist?

1638
01:39:29.440 --> 01:39:34.400
<v Morgan Dwyer (OpenAI)>So I can't speak to personal decisions that
OpenAI executives had made... Have made

1639
01:39:34.400 --> 01:39:35.600
<v Morgan Dwyer (OpenAI)>in their personal capacity.

1640
01:39:36.195 --> 01:39:38.675
<v Morgan Dwyer (OpenAI)>I can tell you that we have supported,

1641
01:39:39.075 --> 01:39:42.595
<v Morgan Dwyer (OpenAI)>frontier safety regulation
across The United States,

1642
01:39:42.995 --> 01:39:48.035
<v Morgan Dwyer (OpenAI)>including endorsing a bill in Illinois
that would be the first of its kind to

1643
01:39:48.035 --> 01:39:49.875
<v Morgan Dwyer (OpenAI)>require third party audits

1644
01:39:50.120 --> 01:39:51.400
<v Morgan Dwyer (OpenAI)>of our safety frameworks.

1645
01:39:52.200 --> 01:39:55.720
<v Morgan Dwyer (OpenAI)>We also supported a version
of a bill in Massachusetts,

1646
01:39:55.720 --> 01:39:58.440
<v Morgan Dwyer (OpenAI)>which was the first of
its kind to reference RSI,

1647
01:39:58.600 --> 01:40:03.160
<v Morgan Dwyer (OpenAI)>which the panel earlier spoke about.
So, again, I can speak to our actions,

1648
01:40:03.560 --> 01:40:05.575
<v Morgan Dwyer (OpenAI)>which is to support strong,

1649
01:40:05.735 --> 01:40:08.055
<v Morgan Dwyer (OpenAI)>harmonized frontier safety legislation,

1650
01:40:08.375 --> 01:40:09.735
<v Morgan Dwyer (OpenAI)>including at the state level.

1651
01:40:11.735 --> 01:40:15.575
<v Speaker Julie Menin>Okay. I'm gonna move on to the other
three companies to the answer to the same

1652
01:40:15.575 --> 01:40:18.375
<v Speaker Julie Menin>question about why
should the public accept

1653
01:40:19.100 --> 01:40:23.020
<v Speaker Julie Menin>the the comments that mister Altman made
yesterday that the world should accept bad

1654
01:40:23.020 --> 01:40:26.940
<v Speaker Julie Menin>things happening and why should AI
companies get to decide what level of risk

1655
01:40:26.940 --> 01:40:30.300
<v Speaker Julie Menin>society must bear. So I'm
gonna go in order. Graham?

1656
01:40:31.995 --> 01:40:35.595
<v Logan Graham (Anthropic)>Yeah. Well, I didn't see
all of mister Altman's,

1657
01:40:35.835 --> 01:40:37.035
<v Logan Graham (Anthropic)>comments yesterday.

1658
01:40:37.755 --> 01:40:39.515
<v Logan Graham (Anthropic)>I share the,

1659
01:40:39.995 --> 01:40:42.075
<v Logan Graham (Anthropic)>sort of the notion expressed here that,

1660
01:40:42.635 --> 01:40:46.990
<v Logan Graham (Anthropic)>number one, while maximizing the benefits
of this technology, you need to minimize

1661
01:40:46.990 --> 01:40:47.710
<v Logan Graham (Anthropic)>the risks,

1662
01:40:48.190 --> 01:40:50.670
<v Logan Graham (Anthropic)>and we should not be
comfortable with letting,

1663
01:40:51.070 --> 01:40:52.350
<v Logan Graham (Anthropic)>these risks happen.

1664
01:40:52.910 --> 01:40:58.225
<v Logan Graham (Anthropic)>This is why over the course of, our history
and my team's history, we have been, I

1665
01:40:58.225 --> 01:41:00.785
<v Logan Graham (Anthropic)>think, sometimes accused
of being overly cautious.

1666
01:41:01.025 --> 01:41:03.185
<v Logan Graham (Anthropic)>I would say, appropriately

1667
01:41:03.345 --> 01:41:04.225
<v Logan Graham (Anthropic)>cautious

1668
01:41:04.785 --> 01:41:10.705
<v Logan Graham (Anthropic)>just this year, withholding our most powerful
model in order to just have more time

1669
01:41:10.705 --> 01:41:13.105
<v Logan Graham (Anthropic)>to even think about collectively

1670
01:41:13.310 --> 01:41:14.030
<v Logan Graham (Anthropic)>as,

1671
01:41:14.670 --> 01:41:18.430
<v Logan Graham (Anthropic)>industry as well as government what
to do about these capabilities.

1672
01:41:19.150 --> 01:41:20.030
<v Logan Graham (Anthropic)>And, fundamentally,

1673
01:41:20.510 --> 01:41:21.710
<v Logan Graham (Anthropic)>this transcends

1674
01:41:21.710 --> 01:41:23.390
<v Logan Graham (Anthropic)>the labs and industry,

1675
01:41:23.630 --> 01:41:27.390
<v Logan Graham (Anthropic)>from our beginning. We have long supported
the role of government here for exactly

1676
01:41:27.390 --> 01:41:27.950
<v Logan Graham (Anthropic)>this reason.

1677
01:41:30.625 --> 01:41:31.905
<v Speaker Julie Menin>Mister Cahill?

1678
01:41:39.105 --> 01:41:43.025
<v Shane Cahill (Meta)>Apologies, speaker. There was a strange
moment there with the mute button.

1679
01:41:43.720 --> 01:41:44.600
<v Shane Cahill (Meta)>Thank

1680
01:41:44.600 --> 01:41:46.920
<v Shane Cahill (Meta)>you very much for for for the question.

1681
01:41:47.160 --> 01:41:51.000
<v Shane Cahill (Meta)>In relation to the quote from mister
Altman that you... That that you read out,

1682
01:41:51.800 --> 01:41:55.240
<v Shane Cahill (Meta)>I I I don't see it the
same way. We absolutely

1683
01:41:55.240 --> 01:41:58.520
<v Shane Cahill (Meta)>recognize the risks of AI
and the opportunities.

1684
01:41:59.265 --> 01:42:03.665
<v Shane Cahill (Meta)>That's why we have our scaling framework.
That's why we have our preparedness

1685
01:42:03.665 --> 01:42:08.145
<v Shane Cahill (Meta)>reports. That's why we have teams and teams
of people and a multilayered approach to

1686
01:42:08.145 --> 01:42:09.265
<v Shane Cahill (Meta)>AI safety.

1687
01:42:10.065 --> 01:42:15.040
<v Shane Cahill (Meta)>One thing I wanted to mention, if I may,
is that Meta's vision is to bring personal

1688
01:42:15.040 --> 01:42:16.080
<v Shane Cahill (Meta)>superintelligence

1689
01:42:16.080 --> 01:42:16.960
<v Shane Cahill (Meta)>to everyone.

1690
01:42:17.360 --> 01:42:22.080
<v Shane Cahill (Meta)>And I listened very intently to the panel
before us in relation to recursive self

1691
01:42:22.080 --> 01:42:22.880
<v Shane Cahill (Meta)>improvement.

1692
01:42:23.200 --> 01:42:26.160
<v Shane Cahill (Meta)>And I just wanted to
mention that committing

1693
01:42:26.480 --> 01:42:32.095
<v Shane Cahill (Meta)>significant majority of compute towards
serving people rather than racing towards

1694
01:42:32.095 --> 01:42:37.535
<v Shane Cahill (Meta)>recursive self improvement is one of the
best ways of ensuring that the technology

1695
01:42:37.535 --> 01:42:38.894
<v Shane Cahill (Meta)>is developed safely.

1696
01:42:39.375 --> 01:42:40.975
<v Shane Cahill (Meta)>Meta has made that commitment,

1697
01:42:41.295 --> 01:42:43.535
<v Shane Cahill (Meta)>and, other labs can do that also.

1698
01:42:46.070 --> 01:42:47.110
<v Speaker Julie Menin>Miss Friend?

1699
01:42:48.550 --> 01:42:52.950
<v Alice Friend (Google)>Yeah. Madam speaker, we shouldn't just
accept bad things happening. We should be

1700
01:42:52.950 --> 01:42:54.310
<v Alice Friend (Google)>working continuously

1701
01:42:54.310 --> 01:42:55.830
<v Alice Friend (Google)>and rigorously

1702
01:42:56.045 --> 01:43:00.845
<v Alice Friend (Google)>on AI safety, which is a collective
effort. It will be a collective effort to

1703
01:43:00.845 --> 01:43:01.805
<v Alice Friend (Google)>identify

1704
01:43:01.965 --> 01:43:03.485
<v Alice Friend (Google)>and mitigate risks,

1705
01:43:03.645 --> 01:43:05.885
<v Alice Friend (Google)>to establish technical standards,

1706
01:43:05.965 --> 01:43:09.405
<v Alice Friend (Google)>to make those identifications and
mitigations all the more rigorous.

1707
01:43:10.210 --> 01:43:12.530
<v Alice Friend (Google)>But I also want to note that

1708
01:43:12.690 --> 01:43:17.970
<v Alice Friend (Google)>one of the promises of AI is that
AI itself can make us safer.

1709
01:43:18.530 --> 01:43:23.405
<v Alice Friend (Google)>I would call your attention to Waymo
vehicles, which in one study had a ninety six

1710
01:43:23.405 --> 01:43:27.245
<v Alice Friend (Google)>percent reduction in injury
causing crashes at intersections,

1711
01:43:27.725 --> 01:43:31.805
<v Alice Friend (Google)>which led doctor Jonathan Slotkin in the
New York Times to call it a public health

1712
01:43:31.805 --> 01:43:32.685
<v Alice Friend (Google)>intervention.

1713
01:43:32.925 --> 01:43:37.860
<v Alice Friend (Google)>So we should think collectively about
preventing and mitigating risks from AI, but

1714
01:43:37.860 --> 01:43:42.420
<v Alice Friend (Google)>we should also keep in mind that we can
use AI itself to reduce risks to human

1715
01:43:42.420 --> 01:43:43.780
<v Alice Friend (Google)>safety and well-being.

1716
01:43:45.780 --> 01:43:51.765
<v Speaker Julie Menin>So if one of your frontier AI models goes
rogue and causes serious financial harm,

1717
01:43:51.765 --> 01:43:52.645
<v Speaker Julie Menin>compromises

1718
01:43:52.645 --> 01:43:57.765
<v Speaker Julie Menin>sensitive data, causes physical injury,
or contributes to a death or deaths,

1719
01:43:58.005 --> 01:44:01.285
<v Speaker Julie Menin>do you each believe your company
bears legal responsibility?

1720
01:44:03.845 --> 01:44:05.765
<v Speaker Julie Menin>I'll start with you, miss Dwyer.

1721
01:44:09.120 --> 01:44:11.040
<v Morgan Dwyer (OpenAI)>I believe that OpenAI

1722
01:44:11.040 --> 01:44:12.400
<v Morgan Dwyer (OpenAI)>takes this responsibility

1723
01:44:12.400 --> 01:44:15.520
<v Morgan Dwyer (OpenAI)>for safely... Very...
Safety very seriously,

1724
01:44:15.840 --> 01:44:18.640
<v Morgan Dwyer (OpenAI)>and we are responsible for developing

1725
01:44:18.640 --> 01:44:21.760
<v Morgan Dwyer (OpenAI)>and evaluating our systems to
ensure that they are safe.

1726
01:44:23.485 --> 01:44:27.005
<v Speaker Julie Menin>So is that a yes or that is a no? Or...

1727
01:44:27.645 --> 01:44:28.205
<v Speaker Julie Menin>I

1728
01:44:28.845 --> 01:44:31.325
<v Speaker Julie Menin>I... Could you be more clear, please?

1729
01:44:33.805 --> 01:44:34.925
<v Morgan Dwyer (OpenAI)>We believe

1730
01:44:35.005 --> 01:44:39.630
<v Morgan Dwyer (OpenAI)>that we are responsible for ensuring
that our systems are developed safely,

1731
01:44:40.030 --> 01:44:42.990
<v Morgan Dwyer (OpenAI)>and that includes training
them to be safe, evaluating

1732
01:44:42.990 --> 01:44:44.270
<v Morgan Dwyer (OpenAI)>them to be safe,

1733
01:44:44.510 --> 01:44:46.510
<v Morgan Dwyer (OpenAI)>and then continuing to monitor,

1734
01:44:46.750 --> 01:44:48.270
<v Morgan Dwyer (OpenAI)>our safeguards after deployment.

1735
01:44:51.565 --> 01:44:55.725
<v Speaker Julie Menin>Okay. I'm I'm gonna take that as as a no
because that is not really clear that you

1736
01:44:55.725 --> 01:44:57.805
<v Speaker Julie Menin>would bear legal responsibility.

1737
01:44:57.965 --> 01:45:00.045
<v Speaker Julie Menin>I'm gonna go now to mister Graham.

1738
01:45:01.485 --> 01:45:05.930
<v Logan Graham (Anthropic)>Yeah. As a technical safety research lead,
I think there are others which will have

1739
01:45:05.930 --> 01:45:10.730
<v Logan Graham (Anthropic)>more nuanced perspectives on that
issue than I. What I what I do know is,

1740
01:45:11.690 --> 01:45:13.130
<v Logan Graham (Anthropic)>first, this is,

1741
01:45:13.450 --> 01:45:18.615
<v Logan Graham (Anthropic)>an issue where every lab, we think, should
be committed to a multilayered safety

1742
01:45:18.615 --> 01:45:22.695
<v Logan Graham (Anthropic)>stack to make sure that doesn't happen in
the first place. These issues don't don't

1743
01:45:22.695 --> 01:45:27.255
<v Logan Graham (Anthropic)>come about in in the first place.
But second, this is candidly a

1744
01:45:27.495 --> 01:45:31.255
<v Logan Graham (Anthropic)>big and complex questions that also
transcends the labs themselves.

1745
01:45:31.680 --> 01:45:35.040
<v Logan Graham (Anthropic)>It's a question about the role
of of governments and others,

1746
01:45:35.280 --> 01:45:38.320
<v Logan Graham (Anthropic)>which we've very long long balked on here.

1747
01:45:38.720 --> 01:45:41.520
<v Speaker Julie Menin>So could we get a yes
or no on that question?

1748
01:45:43.680 --> 01:45:48.435
<v Logan Graham (Anthropic)>I wish it were that simple, and I wish I
I... You know, as a technical researcher, I

1749
01:45:48.435 --> 01:45:50.354
<v Logan Graham (Anthropic)>had more nuance on that exact one.

1750
01:45:50.915 --> 01:45:54.435
<v Logan Graham (Anthropic)>It is it is one that I think needs
to be raised and addressed and,

1751
01:45:54.835 --> 01:45:58.195
<v Logan Graham (Anthropic)>transcends the the the labs
and the companies themselves.

1752
01:45:58.995 --> 01:46:00.515
<v Speaker Julie Menin>I'm gonna go to mister Cahill.

1753
01:46:07.260 --> 01:46:08.140
<v Shane Cahill (Meta)>Apologies.

1754
01:46:08.140 --> 01:46:11.020
<v Shane Cahill (Meta)>Again, the mute button.
I'm sorry. No worries. We

1755
01:46:12.380 --> 01:46:15.820
<v Shane Cahill (Meta)>will get it right. Thank you
very much for for the question.

1756
01:46:17.395 --> 01:46:19.394
<v Shane Cahill (Meta)>I'm not in Meta's legal department,

1757
01:46:19.475 --> 01:46:23.875
<v Shane Cahill (Meta)>so I don't want to speculate or be
imprecise in relation to legal liability

1758
01:46:23.875 --> 01:46:28.675
<v Shane Cahill (Meta)>question. What I can say, though, is that
we are absolutely committed to developing

1759
01:46:28.675 --> 01:46:29.795
<v Shane Cahill (Meta)>our AI safely

1760
01:46:30.360 --> 01:46:32.200
<v Shane Cahill (Meta)>pursuant to our framework,

1761
01:46:32.440 --> 01:46:33.960
<v Shane Cahill (Meta)>which I mentioned previously.

1762
01:46:34.040 --> 01:46:36.520
<v Shane Cahill (Meta)>This encompasses the training, evaluation,

1763
01:46:36.840 --> 01:46:40.120
<v Shane Cahill (Meta)>and release of models across
development, deployment, and use.

1764
01:46:41.720 --> 01:46:42.680
<v Speaker Julie Menin>And this friend.

1765
01:46:44.175 --> 01:46:45.215
<v Alice Friend (Google)>Madam speaker,

1766
01:46:45.375 --> 01:46:50.895
<v Alice Friend (Google)>at Google, we believe that existing legal
frameworks do apply to AI. As I said in my

1767
01:46:50.895 --> 01:46:55.855
<v Alice Friend (Google)>opening statement, we've long said if it's
illegal without AI, it's still illegal

1768
01:46:55.855 --> 01:46:56.894
<v Alice Friend (Google)>with AI.

1769
01:46:57.455 --> 01:46:59.295
<v Alice Friend (Google)>And in New York State, Frontier

1770
01:46:59.610 --> 01:47:02.810
<v Alice Friend (Google)>particular, is regulated
under the RAISE Act. I

1771
01:47:03.370 --> 01:47:04.250
<v Alice Friend (Google) [?]>know. Yeah.

1772
01:47:04.810 --> 01:47:08.650
<v Speaker Julie Menin>Thank you. I mean, your answer is far
clearer than the others, so we appreciate the

1773
01:47:08.650 --> 01:47:13.370
<v Speaker Julie Menin>clarity on that, honestly. I I now have
a question. I'm gonna ask you to please

1774
01:47:13.370 --> 01:47:18.625
<v Speaker Julie Menin>raise your hand if your company has
insurance for catastrophic risks.

1775
01:47:20.865 --> 01:47:21.745
<v Speaker Julie Menin>Okay.

1776
01:47:23.985 --> 01:47:27.025
<v Speaker Julie Menin>I'm gonna take that as a
no that no company has,

1777
01:47:27.665 --> 01:47:31.620
<v Speaker Julie Menin>liability insurance for catastrophic
risk to cover large scale harm.

1778
01:47:31.860 --> 01:47:34.980
<v Speaker Julie Menin>So then the public, I assume, will
be asked to absorb the costs.

1779
01:47:36.260 --> 01:47:40.260
<v Alice Friend (Google)>Madam speaker, I have to be clear. I don't
know the answer to that, but I'd be happy

1780
01:47:40.260 --> 01:47:42.740
<v Alice Friend (Google)>to consult with our lawyers
and get back to you.

1781
01:47:43.775 --> 01:47:48.255
<v Speaker Julie Menin>Yes. We would appreciate an answer
back, expeditiously to this committee.

1782
01:47:48.895 --> 01:47:49.615
<v Unidentified [?]>So

1783
01:47:51.695 --> 01:47:57.855
<v Speaker Julie Menin>I want to now ask each of the companies to
assure the public under oath that your AI

1784
01:47:57.855 --> 01:48:02.760
<v Speaker Julie Menin>agents will will always comply with the
safety guardrails that you impose on them

1785
01:48:02.920 --> 01:48:07.560
<v Speaker Julie Menin>and that you each have said are
necessary to prevent catastrophic harm?

1786
01:48:10.200 --> 01:48:11.960
<v Speaker Julie Menin>I'm gonna start with miss Dwyer, please.

1787
01:48:18.215 --> 01:48:20.615
<v Morgan Dwyer (OpenAI)>As I've said, OpenAI prioritizes

1788
01:48:20.615 --> 01:48:21.495
<v Morgan Dwyer (OpenAI)>safety,

1789
01:48:21.975 --> 01:48:23.735
<v Morgan Dwyer (OpenAI)>and we evaluate

1790
01:48:23.735 --> 01:48:25.975
<v Morgan Dwyer (OpenAI)>our systems against multiple

1791
01:48:25.975 --> 01:48:27.335
<v Morgan Dwyer (OpenAI)>dimensions of safety.

1792
01:48:28.320 --> 01:48:31.520
<v Morgan Dwyer (OpenAI)>And we monitor our
safeguards after deployment.

1793
01:48:32.640 --> 01:48:36.160
<v Morgan Dwyer (OpenAI)>It is... I'm... It's
not possible for me to

1794
01:48:36.640 --> 01:48:38.320
<v Morgan Dwyer (OpenAI)>commit or guarantee,

1795
01:48:38.880 --> 01:48:39.680
<v Morgan Dwyer (OpenAI)>that

1796
01:48:39.840 --> 01:48:42.080
<v Morgan Dwyer (OpenAI)>any technology is without risk,

1797
01:48:42.635 --> 01:48:44.955
<v Morgan Dwyer (OpenAI)>But I can commit that OpenAI,

1798
01:48:45.595 --> 01:48:46.475
<v Morgan Dwyer (OpenAI)>is

1799
01:48:47.035 --> 01:48:50.315
<v Morgan Dwyer (OpenAI)>taking every step it can to ensure,

1800
01:48:50.715 --> 01:48:52.795
<v Morgan Dwyer (OpenAI)>that the development of our systems,

1801
01:48:52.955 --> 01:48:55.435
<v Morgan Dwyer (OpenAI)>as well as their deployment
is conducted safely.

1802
01:48:59.850 --> 01:49:04.250
<v Speaker Julie Menin>Okay. So that that, again, it does not
sound like you are saying that they will...

1803
01:49:04.250 --> 01:49:07.850
<v Speaker Julie Menin>That the agents will always comply
with the safety guardrails.

1804
01:49:09.945 --> 01:49:12.745
<v Speaker Julie Menin>I'm gonna move on to mister Graham.

1805
01:49:14.985 --> 01:49:19.865
<v Logan Graham (Anthropic)>Yeah. What we are committed to and how we
design the safety processes that we that

1806
01:49:19.865 --> 01:49:23.305
<v Logan Graham (Anthropic)>we put in place is to try
to handle exactly this.

1807
01:49:24.220 --> 01:49:29.820
<v Logan Graham (Anthropic)>What we're committed to is trying to define
the industry best standard for safety

1808
01:49:29.820 --> 01:49:31.500
<v Logan Graham (Anthropic)>and pushing that ever higher.

1809
01:49:32.060 --> 01:49:36.220
<v Logan Graham (Anthropic)>But, candidly, the science of doing this
is fundamentally hard and unsettled.

1810
01:49:36.935 --> 01:49:41.575
<v Logan Graham (Anthropic)>From the beginning of the company, we have
tried to publish the best research we can

1811
01:49:41.895 --> 01:49:45.895
<v Logan Graham (Anthropic)>and be as transparent with
the entire world on many cases

1812
01:49:46.455 --> 01:49:47.255
<v Logan Graham (Anthropic)>where,

1813
01:49:47.655 --> 01:49:49.095
<v Logan Graham (Anthropic)>you know, things can go wrong.

1814
01:49:50.010 --> 01:49:54.650
<v Logan Graham (Anthropic)>It is for that reason that we push
for ever better safety standards

1815
01:49:54.970 --> 01:49:58.330
<v Logan Graham (Anthropic)>for regulation and for discourse
like this. It's candidly

1816
01:49:58.810 --> 01:50:01.850
<v Logan Graham (Anthropic)>unsettled now, and this is why
we need to be talking about it.

1817
01:50:03.210 --> 01:50:04.170
<v Speaker Julie Menin>Mister Cahill?

1818
01:50:08.975 --> 01:50:10.575
<v Shane Cahill (Meta)>Speaker, thank you very much.

1819
01:50:10.975 --> 01:50:15.695
<v Shane Cahill (Meta)>If I may for a moment, I was trying to
raise my hand the previous round, but again,

1820
01:50:15.695 --> 01:50:17.855
<v Shane Cahill (Meta)>I will... Technical difficulties.

1821
01:50:18.130 --> 01:50:22.130
<v Shane Cahill (Meta)>I just wanted to, for the record in
relation to your previous question,

1822
01:50:22.610 --> 01:50:26.130
<v Shane Cahill (Meta)>I just wanted to signal also that I don't
know the answer to that question. I'm not

1823
01:50:26.130 --> 01:50:26.930
<v Shane Cahill (Meta)>the right person,

1824
01:50:27.330 --> 01:50:31.810
<v Speaker Julie Menin>but, I would like to be able to come back
to you, afterwards with a... With with an

1825
01:50:31.810 --> 01:50:36.505
<v Speaker Julie Menin>answer if that's okay. I do just wanna say,
and I appreciate you each have your own,

1826
01:50:36.745 --> 01:50:40.505
<v Speaker Julie Menin>and... Look. I was a regulatory attorney
at a large company. I understand that

1827
01:50:40.505 --> 01:50:44.665
<v Speaker Julie Menin>people have their sphere of work that they
work in, but to not know if the company

1828
01:50:44.825 --> 01:50:46.025
<v Speaker Julie Menin>has catastrophic

1829
01:50:46.025 --> 01:50:50.680
<v Speaker Julie Menin>risk insurance when you have a product
that some of your own founders are saying

1830
01:50:50.680 --> 01:50:55.640
<v Speaker Julie Menin>could cause cataclysmic risk is troubling
at best. But, yes, please do. We would

1831
01:50:55.640 --> 01:50:59.480
<v Speaker Julie Menin>appreciate you getting back to the company
on that. And then if you wanna answer the

1832
01:50:59.480 --> 01:51:03.915
<v Speaker Julie Menin>question about will you assure the public
that your AI agents will always comply

1833
01:51:03.915 --> 01:51:06.395
<v Speaker Julie Menin>with the safety guardrails
that you impose on them?

1834
01:51:08.715 --> 01:51:13.980
<v Shane Cahill (Meta)>Thank you, speaker. What I can guarantee
is our commitment to developing and

1835
01:51:13.980 --> 01:51:18.060
<v Shane Cahill (Meta)>deploying our AI offering
safely. We have every incentive

1836
01:51:18.140 --> 01:51:22.540
<v Shane Cahill (Meta)>to address safety, and people will only
embrace AI if they trust it's safe.

1837
01:51:25.580 --> 01:51:26.300
<v Speaker Julie Menin>Miss Friend?

1838
01:51:28.855 --> 01:51:30.374
<v Alice Friend (Google)>Google similarly,

1839
01:51:31.095 --> 01:51:35.894
<v Alice Friend (Google)>has as our ultimate goal reliability
and safety assurance from our systems.

1840
01:51:36.375 --> 01:51:38.535
<v Alice Friend (Google)>To promise perfection

1841
01:51:38.535 --> 01:51:39.334
<v Alice Friend (Google)>would,

1842
01:51:40.215 --> 01:51:40.775
<v Alice Friend (Google)>not

1843
01:51:41.255 --> 01:51:46.350
<v Alice Friend (Google)>be possible with any product on the market,
but our goal is to get as close to that

1844
01:51:46.350 --> 01:51:47.790
<v Alice Friend (Google)>standard as possible.

1845
01:51:48.590 --> 01:51:49.390
<v Alice Friend (Google)>And so,

1846
01:51:50.110 --> 01:51:54.590
<v Alice Friend (Google)>we are constantly improving our
products and our safety practices,

1847
01:51:55.150 --> 01:51:56.830
<v Alice Friend (Google)>to ensure that our users are safe.

1848
01:51:57.415 --> 01:52:01.335
<v Speaker Julie Menin>I don't think we're asking for perfection.
We're just asking for accountability,

1849
01:52:01.335 --> 01:52:02.215
<v Speaker Julie Menin>transparency,

1850
01:52:02.215 --> 01:52:03.575
<v Speaker Julie Menin>and safety overall.

1851
01:52:04.215 --> 01:52:05.175
<v Speaker Julie Menin>Yes, ma'am.

1852
01:52:05.495 --> 01:52:06.215
<v Speaker Julie Menin>Okay.

1853
01:52:06.455 --> 01:52:07.335
<v Speaker Julie Menin>New

1854
01:52:07.335 --> 01:52:11.255
<v Speaker Julie Menin>question. How many times for each of you
has one of your models or agents gained or

1855
01:52:11.255 --> 01:52:13.230
<v Speaker Julie Menin>tried to gain unauthorized

1856
01:52:13.230 --> 01:52:18.030
<v Speaker Julie Menin>access to another system or escaped
from a so called sandbox test?

1857
01:52:18.190 --> 01:52:22.270
<v Speaker Julie Menin>And are there any incidents
that you have not disclosed

1858
01:52:22.270 --> 01:52:23.150
<v Speaker Julie Menin>publicly?

1859
01:52:24.910 --> 01:52:26.750
<v Speaker Julie Menin>I'm gonna start with you, miss Dwyer.

1860
01:52:28.405 --> 01:52:33.605
<v Morgan Dwyer (OpenAI)>Yes. Thank you for the question. I did
wanna start by, like the others, clarifying,

1861
01:52:33.765 --> 01:52:35.845
<v Morgan Dwyer (OpenAI)>that I don't know the answer to the,

1862
01:52:36.325 --> 01:52:37.765
<v Morgan Dwyer (OpenAI)>the insurance question,

1863
01:52:38.245 --> 01:52:39.845
<v Morgan Dwyer (OpenAI)>but would like to follow-up.

1864
01:52:40.850 --> 01:52:43.650
<v Morgan Dwyer (OpenAI)>Now on your question about agents,

1865
01:52:44.610 --> 01:52:47.730
<v Morgan Dwyer (OpenAI)>you may have read in the
news that we, reported,

1866
01:52:47.970 --> 01:52:50.050
<v Morgan Dwyer (OpenAI)>we had an incident with Hugging Face.

1867
01:52:50.770 --> 01:52:52.690
<v Morgan Dwyer (OpenAI)>As a result of that incident,

1868
01:52:53.010 --> 01:52:53.890
<v Morgan Dwyer (OpenAI)>we initiated

1869
01:52:54.135 --> 01:52:57.495
<v Morgan Dwyer (OpenAI)>a significant set of
third party investigations

1870
01:52:57.495 --> 01:52:58.854
<v Morgan Dwyer (OpenAI)>into what happened,

1871
01:52:59.415 --> 01:53:01.734
<v Morgan Dwyer (OpenAI)>and then we made those results public.

1872
01:53:02.295 --> 01:53:03.415
<v Morgan Dwyer (OpenAI)>We have also,

1873
01:53:03.735 --> 01:53:04.695
<v Morgan Dwyer (OpenAI)>initiated

1874
01:53:04.695 --> 01:53:06.855
<v Morgan Dwyer (OpenAI)>a look back investigation

1875
01:53:07.620 --> 01:53:11.460
<v Morgan Dwyer (OpenAI)>to try to identify whether
there were past instances

1876
01:53:11.860 --> 01:53:14.020
<v Morgan Dwyer (OpenAI)>of what we call misaligned

1877
01:53:14.020 --> 01:53:14.820
<v Morgan Dwyer (OpenAI)>agents.

1878
01:53:15.460 --> 01:53:16.580
<v Morgan Dwyer (OpenAI)>That investigation

1879
01:53:16.580 --> 01:53:17.780
<v Morgan Dwyer (OpenAI)>is ongoing,

1880
01:53:18.260 --> 01:53:22.740
<v Morgan Dwyer (OpenAI)>but we are committed to continuing that
investigation as urgently as possible

1881
01:53:23.305 --> 01:53:26.984
<v Morgan Dwyer (OpenAI)>to notifying potential
affected third parties,

1882
01:53:27.465 --> 01:53:29.865
<v Morgan Dwyer (OpenAI)>and then to making the results public.

1883
01:53:30.025 --> 01:53:31.705
<v Morgan Dwyer (OpenAI)>Look. We think transparency

1884
01:53:31.705 --> 01:53:32.265
<v Morgan Dwyer (OpenAI)>here,

1885
01:53:32.745 --> 01:53:34.425
<v Morgan Dwyer (OpenAI)>is critically important,

1886
01:53:34.585 --> 01:53:35.865
<v Morgan Dwyer (OpenAI)>for policymakers

1887
01:53:35.865 --> 01:53:40.370
<v Morgan Dwyer (OpenAI)>like you, for the public who wants to
know whether they can trust our products,

1888
01:53:41.650 --> 01:53:46.130
<v Morgan Dwyer (OpenAI)>and for the rest of industry so
that we can all learn together,

1889
01:53:46.610 --> 01:53:50.850
<v Morgan Dwyer (OpenAI)>to raise, the science of safety that
we've been talking about here today.

1890
01:53:52.205 --> 01:53:57.245
<v Speaker Julie Menin>Since you mentioned Hugging Face, can you
confirm, did the company select this third

1891
01:53:57.245 --> 01:53:57.965
<v Speaker Julie Menin>party

1892
01:53:58.125 --> 01:53:59.245
<v Speaker Julie Menin>investigator?

1893
01:54:01.405 --> 01:54:04.765
<v Morgan Dwyer (OpenAI)>I believe that we selected multiple
third party investigators.

1894
01:54:05.370 --> 01:54:10.170
<v Speaker Julie Menin>And why did the company narrow the dates
of review to a three week window? And

1895
01:54:10.170 --> 01:54:15.210
<v Speaker Julie Menin>reviewers said that virtually all of the
data they examined was actually from July 7

1896
01:54:15.210 --> 01:54:19.385
<v Speaker Julie Menin>to the thirteenth. Were there any incidents
that the company was aware of that were

1897
01:54:19.385 --> 01:54:21.065
<v Speaker Julie Menin>outside of that time frame?

1898
01:54:24.585 --> 01:54:25.145
<v Morgan Dwyer (OpenAI)>We,

1899
01:54:26.345 --> 01:54:30.985
<v Morgan Dwyer (OpenAI)>we limited the amount of time that the
third party investigators had to conduct

1900
01:54:30.985 --> 01:54:31.865
<v Morgan Dwyer (OpenAI)>their review

1901
01:54:32.070 --> 01:54:34.390
<v Morgan Dwyer (OpenAI)>because we felt a sense of urgency.

1902
01:54:34.950 --> 01:54:40.310
<v Morgan Dwyer (OpenAI)>We believed it was important to get
information to the public as soon as possible,

1903
01:54:40.870 --> 01:54:43.430
<v Morgan Dwyer (OpenAI)>so that they could harden
their own systems.

1904
01:54:44.070 --> 01:54:48.985
<v Morgan Dwyer (OpenAI)>That said, when those investigators
asked for more time, we gave it to them.

1905
01:54:49.705 --> 01:54:53.945
<v Speaker Julie Menin>So are you committing then to
allow outside investigators

1906
01:54:54.505 --> 01:54:58.825
<v Speaker Julie Menin>to examine additional dates outside of
the time frame that the company selected?

1907
01:55:00.910 --> 01:55:04.510
<v Morgan Dwyer (OpenAI)>We continue to work with
multiple third parties,

1908
01:55:04.990 --> 01:55:06.910
<v Morgan Dwyer (OpenAI)>and we have also endorsed

1909
01:55:06.910 --> 01:55:08.830
<v Morgan Dwyer (OpenAI)>multiple pieces of legislation

1910
01:55:08.830 --> 01:55:11.870
<v Morgan Dwyer (OpenAI)>that would require us to
work with third parties,

1911
01:55:11.950 --> 01:55:15.230
<v Morgan Dwyer (OpenAI)>including third party audits
of our safety frameworks.

1912
01:55:17.775 --> 01:55:19.775
<v Speaker Julie Menin>I'm gonna go to mister Graham.

1913
01:55:24.815 --> 01:55:29.215
<v Logan Graham (Anthropic)>Thank you. Yes. So we... To
your original question of,

1914
01:55:29.615 --> 01:55:30.655
<v Logan Graham (Anthropic)>how many incidents,

1915
01:55:31.470 --> 01:55:34.830
<v Logan Graham (Anthropic)>you know, cases of
breaking out of sandboxes,

1916
01:55:35.230 --> 01:55:40.110
<v Logan Graham (Anthropic)>these are capabilities we evaluate for all
the time. We've published not just with

1917
01:55:40.110 --> 01:55:41.230
<v Logan Graham (Anthropic)>these incidents

1918
01:55:41.870 --> 01:55:42.990
<v Logan Graham (Anthropic)>from the summer,

1919
01:55:43.915 --> 01:55:45.515
<v Logan Graham (Anthropic)>but also previously

1920
01:55:45.515 --> 01:55:48.555
<v Logan Graham (Anthropic)>cases where we observe
advanced cyber capabilities,

1921
01:55:48.875 --> 01:55:50.795
<v Logan Graham (Anthropic)>including breaking out of sandboxes,

1922
01:55:50.795 --> 01:55:54.795
<v Logan Graham (Anthropic)>which to be clear from a
technical standpoint is just one

1923
01:55:55.035 --> 01:55:59.360
<v Logan Graham (Anthropic)>part and doesn't... You know, one type of
capability that does not in itself mean

1924
01:55:59.360 --> 01:56:01.840
<v Logan Graham (Anthropic)>that the system would reach

1925
01:56:02.160 --> 01:56:04.240
<v Logan Graham (Anthropic)>a a real world affected party,

1926
01:56:04.400 --> 01:56:06.000
<v Logan Graham (Anthropic)>but ones that we did observe

1927
01:56:06.080 --> 01:56:10.880
<v Logan Graham (Anthropic)>in training at points, and we disclosed
that in lengthy lengthy risk report.

1928
01:56:11.515 --> 01:56:14.395
<v Logan Graham (Anthropic)>Yeah. It's difficult to comment on ongoing

1929
01:56:14.715 --> 01:56:15.915
<v Logan Graham (Anthropic)>investigations.

1930
01:56:16.315 --> 01:56:20.715
<v Logan Graham (Anthropic)>It's really important that we do that in
as secure a way as possible. We follow a

1931
01:56:20.715 --> 01:56:21.515
<v Logan Graham (Anthropic)>process of

1932
01:56:21.915 --> 01:56:25.760
<v Logan Graham (Anthropic)>making sure we can remediate and do so
and be as transparent as possible while

1933
01:56:25.760 --> 01:56:28.960
<v Logan Graham (Anthropic)>preserving the safety
of any affected party.

1934
01:56:28.960 --> 01:56:33.040
<v Logan Graham (Anthropic)>And, of course, incidents are are always
ongoing, but I I sort of reassure and

1935
01:56:33.040 --> 01:56:36.960
<v Logan Graham (Anthropic)>reassert that we are extremely committed
to being as transparent as possible

1936
01:56:36.960 --> 01:56:38.880
<v Logan Graham (Anthropic)>fundamentally because we believe

1937
01:56:39.205 --> 01:56:43.765
<v Logan Graham (Anthropic)>that when we can, being as transparent as
possible about what we've learned, who is

1938
01:56:43.765 --> 01:56:48.405
<v Logan Graham (Anthropic)>affected, how others might learn from it,
how to defend against risks, or how we can

1939
01:56:48.405 --> 01:56:49.525
<v Logan Graham (Anthropic)>help them as well,

1940
01:56:49.925 --> 01:56:54.350
<v Speaker Julie Menin>is is our number one priority. So so on that
note then, are there any incidents that

1941
01:56:54.350 --> 01:56:58.110
<v Speaker Julie Menin>you have not disclosed publicly that
you're currently investigating?

1942
01:56:59.230 --> 01:57:04.270
<v Logan Graham (Anthropic)>As a matter of sort of ongoing
business, there are incidents,

1943
01:57:04.670 --> 01:57:07.150
<v Logan Graham (Anthropic)>sort of all the time of
many different natures.

1944
01:57:07.625 --> 01:57:09.065
<v Logan Graham (Anthropic)>So, for example,

1945
01:57:09.625 --> 01:57:10.825
<v Logan Graham (Anthropic)>one that we,

1946
01:57:11.145 --> 01:57:15.465
<v Logan Graham (Anthropic)>are transparent about quite a lot when
it's the right time to do so are cases of

1947
01:57:15.465 --> 01:57:17.785
<v Logan Graham (Anthropic)>misuse, for example, on our platforms.

1948
01:57:18.345 --> 01:57:23.620
<v Logan Graham (Anthropic)>I'd refer back to our early September report
on our threat intelligence report that

1949
01:57:23.620 --> 01:57:24.500
<v Logan Graham (Anthropic)>we released,

1950
01:57:24.820 --> 01:57:28.580
<v Logan Graham (Anthropic)>where we detailed this in in quite some
detail. What's really important, though, is

1951
01:57:28.580 --> 01:57:32.345
<v Logan Graham (Anthropic)>sometimes we have to work with law enforcement
and the right level of government to

1952
01:57:32.345 --> 01:57:33.225
<v Logan Graham (Anthropic)>remediate,

1953
01:57:33.465 --> 01:57:36.905
<v Logan Graham (Anthropic)>to figure out what information you can
and can't disclose to make sure that the

1954
01:57:36.905 --> 01:57:38.505
<v Logan Graham (Anthropic)>affected parties remain safe.

1955
01:57:39.065 --> 01:57:43.705
<v Logan Graham (Anthropic)>And so there... So as a matter, there's
ongoing all the time, but disclosure at the

1956
01:57:43.705 --> 01:57:47.640
<v Logan Graham (Anthropic)>right time and the safest way is our
commitment. And when will you be disclosing

1957
01:57:47.640 --> 01:57:50.600
<v Logan Graham (Anthropic)>these new incidents that
are being investigated?

1958
01:57:52.840 --> 01:57:55.480
<v Logan Graham (Anthropic)>This is sort of an
ongoing ongoing practice.

1959
01:57:55.800 --> 01:57:57.080
<v Logan Graham (Anthropic)>We regularly

1960
01:57:57.080 --> 01:58:02.045
<v Logan Graham (Anthropic)>release threat intelligence reports in
cases like that. When we release a model or

1961
01:58:02.045 --> 01:58:04.685
<v Logan Graham (Anthropic)>when we do risk reports,
which we do regularly,

1962
01:58:05.325 --> 01:58:10.685
<v Logan Graham (Anthropic)>we we sort of have a cadence and schedule
that we do that. And then when important,

1963
01:58:10.685 --> 01:58:15.565
<v Logan Graham (Anthropic)>we, and sort of urgent, we will will disclose
in in advance that schedule as well.

1964
01:58:16.430 --> 01:58:18.830
<v Speaker Julie Menin>I'm gonna go now to mister Cahill.

1965
01:58:22.190 --> 01:58:23.790
<v Shane Cahill (Meta)>Speaker, thank you very much.

1966
01:58:24.110 --> 01:58:27.070
<v Shane Cahill (Meta)>I'm not aware of, any other incidences

1967
01:58:27.070 --> 01:58:28.910
<v Shane Cahill (Meta)>beyond the incident,

1968
01:58:28.990 --> 01:58:32.255
<v Shane Cahill (Meta)>in the summer, which we have a public post

1969
01:58:32.255 --> 01:58:36.975
<v Shane Cahill (Meta)>setting out the details in relation to
pursuant to our independent review of that.

1970
01:58:38.815 --> 01:58:42.335
<v Speaker Julie Menin>Okay. I understand you're not aware, but
is the company aware? Just to make the

1971
01:58:42.335 --> 01:58:43.055
<v Speaker Julie Menin>question broader.

1972
01:58:46.110 --> 01:58:49.150
<v Speaker Julie Menin>In other words, would there be others
at the company who might have that

1973
01:58:49.150 --> 01:58:49.950
<v Speaker Julie Menin>information?

1974
01:58:51.390 --> 01:58:55.870
<v Shane Cahill (Meta)>I don't I don't... I I don't wish to speculate
in relation to that, but I'm happy to

1975
01:58:55.870 --> 01:58:59.230
<v Shane Cahill (Meta)>follow-up afterwards. Okay. I'm
gonna move on to miss Friend.

1976
01:59:01.335 --> 01:59:02.775
<v Alice Friend (Google)>Yes. Madam speaker,

1977
01:59:03.255 --> 01:59:05.495
<v Alice Friend (Google)>in our three incidents,

1978
01:59:06.055 --> 01:59:07.415
<v Alice Friend (Google)>of agents,

1979
01:59:08.135 --> 01:59:12.055
<v Alice Friend (Google)>leaving a a test environment
and interacting with the real,

1980
01:59:12.615 --> 01:59:13.095
<v Alice Friend (Google)>Internet,

1981
01:59:13.970 --> 01:59:18.770
<v Alice Friend (Google)>In all three incidents, the models stopped
their activities as soon as they realized

1982
01:59:18.770 --> 01:59:21.490
<v Alice Friend (Google)>that they were interacting
with live websites.

1983
01:59:22.770 --> 01:59:23.890
<v Alice Friend (Google)>We reported

1984
01:59:23.890 --> 01:59:27.730
<v Alice Friend (Google)>to, the owners of those websites
as well as to federal agencies

1985
01:59:28.485 --> 01:59:30.005
<v Alice Friend (Google)>about those incidents.

1986
01:59:30.645 --> 01:59:33.685
<v Alice Friend (Google)>I am not personally aware of
any additional incidents.

1987
01:59:35.205 --> 01:59:40.085
<v Speaker Julie Menin>Okay. And same question then. Is the company
aware of... Are you... Do you have any

1988
01:59:40.085 --> 01:59:44.220
<v Speaker Julie Menin>knowledge that others in the company would
be aware of any additional such instances

1989
01:59:44.220 --> 01:59:46.540
<v Speaker Julie Menin>that the company has
not publicly reported?

1990
01:59:47.020 --> 01:59:50.860
<v Alice Friend (Google)>I do not, ma'am, but I can take
that question back as well. Okay.

1991
01:59:51.500 --> 01:59:53.180
<v Speaker Julie Menin>Will you all

1992
01:59:54.060 --> 01:59:59.535
<v Speaker Julie Menin>commit right now under oath publicly that
if your models fail an internal test or a

1993
01:59:59.535 --> 02:00:01.535
<v Speaker Julie Menin>third party validation test

1994
02:00:01.695 --> 02:00:05.935
<v Speaker Julie Menin>that's not selected by your company, will
you commit to not release that model?

1995
02:00:10.575 --> 02:00:12.015
<v Speaker Julie Menin>Gonna start with miss Dwyer.

1996
02:00:13.730 --> 02:00:19.090
<v Morgan Dwyer (OpenAI)>Yeah. I can commit that OpenAI is not going
to release models that we don't believe

1997
02:00:19.090 --> 02:00:20.210
<v Morgan Dwyer (OpenAI)>that are safe.

1998
02:00:20.370 --> 02:00:24.770
<v Morgan Dwyer (OpenAI)>As part of, that assessment process, we
work with multiple third parties. And as I

1999
02:00:24.770 --> 02:00:29.555
<v Morgan Dwyer (OpenAI)>mentioned before, with The US as well
as The U... UK Federal Governments,

2000
02:00:30.035 --> 02:00:32.675
<v Morgan Dwyer (OpenAI)>to assess whether our models are safe.

2001
02:00:35.075 --> 02:00:39.475
<v Speaker Julie Menin>So if it fails either an internal
test or a third party test,

2002
02:00:40.035 --> 02:00:42.115
<v Speaker Julie Menin>you will commit to not release it?

2003
02:00:43.890 --> 02:00:45.890
<v Morgan Dwyer (OpenAI)>OpenAI has historically,

2004
02:00:46.370 --> 02:00:48.770
<v Morgan Dwyer (OpenAI)>delayed the release of models,

2005
02:00:49.170 --> 02:00:53.170
<v Morgan Dwyer (OpenAI)>to ensure that we, can build
up the right safeguards.

2006
02:00:53.490 --> 02:00:55.970
<v Morgan Dwyer (OpenAI)>We've done it before,
and we'll do it again.

2007
02:00:56.985 --> 02:01:01.865
<v Speaker Julie Menin>Again, I think a simple yes or no would
instill more confidence in the public on a

2008
02:01:01.865 --> 02:01:04.025
<v Speaker Julie Menin>matter as serious as this.

2009
02:01:04.345 --> 02:01:06.185
<v Speaker Julie Menin>I'm gonna move to mister Graham.

2010
02:01:08.185 --> 02:01:13.890
<v Logan Graham (Anthropic)>We take that one particularly seriously.
The way we approach it is we try to detail,

2011
02:01:14.370 --> 02:01:18.050
<v Logan Graham (Anthropic)>an even more nuanced and lengthy
approach to how we release models,

2012
02:01:18.130 --> 02:01:21.810
<v Logan Graham (Anthropic)>what safeguards they require
in order to be released,

2013
02:01:21.970 --> 02:01:23.730
<v Logan Graham (Anthropic)>what tests need to be done.

2014
02:01:24.210 --> 02:01:29.445
<v Logan Graham (Anthropic)>And to your point about independent
evaluators, this is one of the most important

2015
02:01:29.445 --> 02:01:30.725
<v Logan Graham (Anthropic)>components, we think.

2016
02:01:31.365 --> 02:01:34.565
<v Logan Graham (Anthropic)>A... We have tried and long supported

2017
02:01:34.965 --> 02:01:39.445
<v Logan Graham (Anthropic)>creating a burgeoning sort of
group of independent evaluators,

2018
02:01:39.525 --> 02:01:42.405
<v Logan Graham (Anthropic)>including working with multiple
governments to do exactly this,

2019
02:01:43.260 --> 02:01:47.100
<v Logan Graham (Anthropic)>And we collectively work together
to try to refine and improve

2020
02:01:47.180 --> 02:01:51.500
<v Logan Graham (Anthropic)>the exact decision process by
which you release a model.

2021
02:01:51.980 --> 02:01:56.495
<v Logan Graham (Anthropic)>And so we think, you know, what we should
do is develop the right and best and

2022
02:01:56.495 --> 02:02:00.095
<v Logan Graham (Anthropic)>improve on these decision processes. We
detail ours in our responsible scaling

2023
02:02:00.095 --> 02:02:00.814
<v Logan Graham (Anthropic)>policy.

2024
02:02:00.975 --> 02:02:05.055
<v Logan Graham (Anthropic)>We should refine and improve it
every so often. So we release updates

2025
02:02:05.055 --> 02:02:06.415
<v Logan Graham (Anthropic)>based on what we learn,

2026
02:02:07.215 --> 02:02:10.095
<v Logan Graham (Anthropic)>you know, every six to twelve
months or so at this at this rate,

2027
02:02:10.740 --> 02:02:13.540
<v Logan Graham (Anthropic)>and that this should involve
independent evaluators.

2028
02:02:13.540 --> 02:02:16.419
<v Speaker Julie Menin>Okay. That also is not a clear yes or no.

2029
02:02:16.820 --> 02:02:18.660
<v Speaker Julie Menin>I'm gonna move on to mister Cahill.

2030
02:02:22.385 --> 02:02:24.625
<v Shane Cahill (Meta)>Madam speaker, thank you
very much for that question.

2031
02:02:24.945 --> 02:02:28.145
<v Shane Cahill (Meta)>What I can commit to is that
we do not deploy models,

2032
02:02:28.625 --> 02:02:31.825
<v Shane Cahill (Meta)>which are not safe pursuant
to, to our scaling framework.

2033
02:02:32.305 --> 02:02:34.865
<v Shane Cahill (Meta)>And I just wanted to to
to mention also that,

2034
02:02:35.105 --> 02:02:39.560
<v Shane Cahill (Meta)>and I think I mentioned this a little
earlier, but we do believe that every lab has

2035
02:02:39.560 --> 02:02:40.680
<v Shane Cahill (Meta)>a responsibility

2036
02:02:41.000 --> 02:02:45.720
<v Shane Cahill (Meta)>to go at the pace to ensure the safety of
the models and the ability to take action

2037
02:02:45.720 --> 02:02:47.080
<v Shane Cahill (Meta)>in relation to that pacing.

2038
02:02:48.085 --> 02:02:53.765
<v Shane Cahill (Meta)>For example, at Meta, we delayed the release
of Muse for several months to focus on

2039
02:02:53.765 --> 02:02:55.044
<v Shane Cahill (Meta)>safety and security.

2040
02:02:55.445 --> 02:02:58.405
<v Shane Cahill (Meta)>In relation to to independent evaluators,

2041
02:02:58.645 --> 02:02:59.765
<v Shane Cahill (Meta)>we recently

2042
02:03:00.085 --> 02:03:03.720
<v Shane Cahill (Meta)>signed in a coordinated commitment
to partner with independent,

2043
02:03:03.880 --> 02:03:04.680
<v Shane Cahill (Meta)>evaluators

2044
02:03:04.680 --> 02:03:05.720
<v Shane Cahill (Meta)>and assessors.

2045
02:03:05.720 --> 02:03:09.320
<v Shane Cahill (Meta)>And this is also detailed
in our, in our framework.

2046
02:03:12.280 --> 02:03:14.360
<v Speaker Julie Menin>Okay. Mister, miss Friend.

2047
02:03:16.415 --> 02:03:18.575
<v Alice Friend (Google)>Ma'am, as part of our multilayered

2048
02:03:18.575 --> 02:03:21.215
<v Alice Friend (Google)>approach to responsible AI governance,

2049
02:03:21.615 --> 02:03:26.575
<v Alice Friend (Google)>we conduct rigorous launch reviews ahead of
launch of models to ensure that anything

2050
02:03:26.575 --> 02:03:29.375
<v Alice Friend (Google)>we put into general availability

2051
02:03:29.375 --> 02:03:30.095
<v Alice Friend (Google)>is safe.

2052
02:03:32.790 --> 02:03:37.110
<v Speaker Julie Menin>Okay. And, I mean, again, I think it
would be far clearer if we could get clear

2053
02:03:37.190 --> 02:03:41.910
<v Speaker Julie Menin>yeses or noes from everyone. But, in
the absence of that, I'm going to,

2054
02:03:42.310 --> 02:03:46.150
<v Speaker Julie Menin>take, you know, what you've each
said as sort of a equivocation.

2055
02:03:47.535 --> 02:03:51.455
<v Speaker Julie Menin>Okay. I'm gonna move on. I have two final
questions, and I'm gonna pass it on to

2056
02:03:51.455 --> 02:03:53.055
<v Speaker Julie Menin>our, technology

2057
02:03:53.055 --> 02:03:57.935
<v Speaker Julie Menin>chair. For OpenAI, you sent a letter to
the council on October 1, and in it, you

2058
02:03:57.935 --> 02:04:02.240
<v Speaker Julie Menin>recommended that to better protect New
York City, the city should connect its

2059
02:04:02.240 --> 02:04:05.760
<v Speaker Julie Menin>existing cyber and emergency
management capabilities

2060
02:04:05.920 --> 02:04:08.480
<v Speaker Julie Menin>with the strongest available escalation,

2061
02:04:08.480 --> 02:04:10.720
<v Speaker Julie Menin>containment, and recovery capabilities.

2062
02:04:11.120 --> 02:04:12.895
<v Speaker Julie Menin>Have there been any conversations

2063
02:04:12.895 --> 02:04:16.495
<v Speaker Julie Menin>with the city of New York
about that recommendation?

2064
02:04:19.295 --> 02:04:22.095
<v Morgan Dwyer (OpenAI)>I'm not aware if there
have been conversations,

2065
02:04:22.335 --> 02:04:24.175
<v Morgan Dwyer (OpenAI)>about that specific recommendation.

2066
02:04:24.820 --> 02:04:26.740
<v Morgan Dwyer (OpenAI)>I'm happy to follow-up.

2067
02:04:27.220 --> 02:04:32.180
<v Morgan Dwyer (OpenAI)>I do know that with respect to cyber,
we recently launched a $1,000,000,000

2068
02:04:32.180 --> 02:04:34.980
<v Morgan Dwyer (OpenAI)>initiative to make
investments in cyber defense.

2069
02:04:35.140 --> 02:04:37.060
<v Morgan Dwyer (OpenAI)>That includes giving organizations

2070
02:04:37.060 --> 02:04:39.380
<v Morgan Dwyer (OpenAI)>access to our tools to defend themselves,

2071
02:04:39.735 --> 02:04:40.615
<v Morgan Dwyer (OpenAI)>training,

2072
02:04:41.255 --> 02:04:42.695
<v Morgan Dwyer (OpenAI)>and technical assistance.

2073
02:04:42.935 --> 02:04:46.775
<v Morgan Dwyer (OpenAI)>And we're working with New York State,
as well as with the Port Authority.

2074
02:04:47.895 --> 02:04:53.110
<v Speaker Julie Menin>Okay. And then the last question I have
before passing it on to the chair is for

2075
02:04:53.110 --> 02:04:56.950
<v Speaker Julie Menin>each of you, based on your own internal
projections, what percentage of New York's

2076
02:04:56.950 --> 02:05:01.670
<v Speaker Julie Menin>workforce do you estimate will be
displaced by AI over the next five years?

2077
02:05:03.510 --> 02:05:05.350
<v Speaker Julie Menin>And I'll start with you, miss Dwyer.

2078
02:05:09.175 --> 02:05:14.695
<v Morgan Dwyer (OpenAI)>I don't have a number, to share with you,
but OpenAI has an economic research team

2079
02:05:14.695 --> 02:05:19.735
<v Morgan Dwyer (OpenAI)>that is looking at this exact
question of how AI impacts jobs.

2080
02:05:20.710 --> 02:05:25.750
<v Morgan Dwyer (OpenAI)>We think it's critically important to
understand how AI might impact work, and we

2081
02:05:25.750 --> 02:05:30.950
<v Morgan Dwyer (OpenAI)>also think it's important to ensure that
AI is not just being used to increase

2082
02:05:30.950 --> 02:05:31.910
<v Morgan Dwyer (OpenAI)>productivity,

2083
02:05:32.550 --> 02:05:37.045
<v Morgan Dwyer (OpenAI)>but that it actually benefits workers
and gives them more opportunity.

2084
02:05:37.605 --> 02:05:43.205
<v Morgan Dwyer (OpenAI)>That's why we're committed to making
our analysis of job impacts public,

2085
02:05:43.605 --> 02:05:46.485
<v Morgan Dwyer (OpenAI)>and why we've also worked
with policymakers,

2086
02:05:46.725 --> 02:05:47.365
<v Morgan Dwyer (OpenAI)>including

2087
02:05:48.660 --> 02:05:51.940
<v Morgan Dwyer (OpenAI)>endorsing the bipartisan
Workforce Transparency Act,

2088
02:05:52.180 --> 02:05:55.940
<v Morgan Dwyer (OpenAI)>to create voluntary pathways
for other companies,

2089
02:05:56.260 --> 02:05:58.260
<v Morgan Dwyer (OpenAI)>to report potential impacts.

2090
02:05:58.980 --> 02:06:01.855
<v Speaker Julie Menin>So Boston Consulting released a report

2091
02:06:01.855 --> 02:06:02.655
<v Speaker Julie Menin>recently

2092
02:06:02.655 --> 02:06:03.615
<v Speaker Julie Menin>that said,

2093
02:06:03.855 --> 02:06:08.415
<v Speaker Julie Menin>that they believe there
will be a 10 to 15%

2094
02:06:08.415 --> 02:06:09.615
<v Speaker Julie Menin>of the workforce

2095
02:06:09.615 --> 02:06:13.215
<v Speaker Julie Menin>will be... Could be
displaced by AI by 2031.

2096
02:06:13.615 --> 02:06:19.210
<v Speaker Julie Menin>If we took that out to look at New York
City, New York City has roughly 4,200,000

2097
02:06:19.210 --> 02:06:22.490
<v Speaker Julie Menin>public and private sector workers. At 15%,

2098
02:06:22.490 --> 02:06:24.250
<v Speaker Julie Menin>that would mean more than 600,000

2099
02:06:24.250 --> 02:06:26.570
<v Speaker Julie Menin>New Yorkers potentially displaced

2100
02:06:26.650 --> 02:06:27.610
<v Speaker Julie Menin>from their jobs.

2101
02:06:29.885 --> 02:06:32.445
<v Speaker Julie Menin>Was that an assessment
that you agree with?

2102
02:06:35.165 --> 02:06:38.285
<v Morgan Dwyer (OpenAI)>I'm not familiar with
that particular report.

2103
02:06:38.685 --> 02:06:40.365
<v Morgan Dwyer (OpenAI)>I'd be happy to follow-up,

2104
02:06:40.605 --> 02:06:44.480
<v Morgan Dwyer (OpenAI)>in our written testimony with
more details on OpenAI's

2105
02:06:44.480 --> 02:06:46.560
<v Morgan Dwyer (OpenAI)>analysis into this topic.

2106
02:06:46.880 --> 02:06:49.040
<v Morgan Dwyer (OpenAI)>We've looked at job exposure.

2107
02:06:49.280 --> 02:06:51.600
<v Morgan Dwyer (OpenAI)>Now exposure doesn't mean elimination.

2108
02:06:51.600 --> 02:06:54.560
<v Morgan Dwyer (OpenAI)>It means that jobs might change over time.

2109
02:06:55.040 --> 02:06:58.560
<v Morgan Dwyer (OpenAI)>So I'm happy to follow-up with
some details of of that research.

2110
02:06:59.175 --> 02:07:03.895
<v Speaker Julie Menin>I'm gonna now turn to mister Graham to
answer the question about what you believe

2111
02:07:03.895 --> 02:07:08.295
<v Speaker Julie Menin>the percentage of New York's workforce do
you estimate will be displaced by AI over

2112
02:07:08.295 --> 02:07:09.175
<v Speaker Julie Menin>the next five years.

2113
02:07:11.270 --> 02:07:17.030
<v Logan Graham (Anthropic)>Well, I don't have an exact number. We are
particularly focused and concerned about

2114
02:07:17.030 --> 02:07:19.110
<v Logan Graham (Anthropic)>this question of job impacts.

2115
02:07:19.270 --> 02:07:23.510
<v Logan Graham (Anthropic)>We've been very vocal, very early for
the past few years on on exactly this.

2116
02:07:24.405 --> 02:07:28.085
<v Logan Graham (Anthropic)>To do this, we set up an economic research
team. They sit right next to my team on

2117
02:07:28.085 --> 02:07:28.725
<v Logan Graham (Anthropic)>the floor,

2118
02:07:29.365 --> 02:07:32.085
<v Logan Graham (Anthropic)>and they published a number of
things that I think actually

2119
02:07:32.325 --> 02:07:36.965
<v Logan Graham (Anthropic)>contain valuable information here, number
of economic scenarios, interactive

2120
02:07:36.965 --> 02:07:38.245
<v Logan Graham (Anthropic)>economic scenarios

2121
02:07:38.570 --> 02:07:42.410
<v Logan Graham (Anthropic)>to understand what the impacts on
growth, unemployment might be,

2122
02:07:42.730 --> 02:07:45.050
<v Logan Graham (Anthropic)>as well as an economic policy framework

2123
02:07:45.530 --> 02:07:50.010
<v Logan Graham (Anthropic)>to try to wrestle with exactly a question
of what we think governments ought to do

2124
02:07:50.250 --> 02:07:50.570
<v Logan Graham (Anthropic)>there.

2125
02:07:51.455 --> 02:07:55.135
<v Logan Graham (Anthropic)>I'll just say the fact that the... You
know, the reason why we are doing this is

2126
02:07:55.135 --> 02:07:58.175
<v Logan Graham (Anthropic)>because we're particularly concerned
about about exactly that question.

2127
02:07:58.975 --> 02:08:02.335
<v Speaker Julie Menin>I mean, obviously, losing a
career job is catastrophic

2128
02:08:02.335 --> 02:08:05.215
<v Speaker Julie Menin>to any worker and to their family. So

2129
02:08:06.320 --> 02:08:10.480
<v Speaker Julie Menin>displacement on this scale is
going to have enormous consequences

2130
02:08:10.560 --> 02:08:14.560
<v Speaker Julie Menin>not only on the individual and family
level, but for our entire city and the

2131
02:08:14.560 --> 02:08:15.359
<v Speaker Julie Menin>economy.

2132
02:08:16.000 --> 02:08:17.840
<v Speaker Julie Menin>I'll move on to mister Cahill.

2133
02:08:20.485 --> 02:08:24.805
<v Shane Cahill (Meta)>Madam speaker, first off, I want to
acknowledge absolutely the concern in relation

2134
02:08:24.805 --> 02:08:30.005
<v Shane Cahill (Meta)>to, how how how you, have,
articulated it. That is indeed,

2135
02:08:30.405 --> 02:08:33.045
<v Shane Cahill (Meta)>a concern, but I look at
this somewhat differently.

2136
02:08:33.950 --> 02:08:36.430
<v Shane Cahill (Meta)>I was reading some incredibly

2137
02:08:37.310 --> 02:08:40.270
<v Shane Cahill (Meta)>impressive statistics from NYC EDC

2138
02:08:40.430 --> 02:08:43.390
<v Shane Cahill (Meta)>in relation to the investments
that are happening here

2139
02:08:43.710 --> 02:08:48.965
<v Shane Cahill (Meta)>in New York. I think 20,000,000,000 VC
investment investment in AI, 40,000 jobs. I

2140
02:08:48.965 --> 02:08:51.045
<v Shane Cahill (Meta)>think those figures are from 2022,

2141
02:08:51.045 --> 02:08:53.205
<v Shane Cahill (Meta)>so it's probably a little bit more by now.

2142
02:08:54.485 --> 02:08:59.045
<v Shane Cahill (Meta)>But what I'm really actually incredibly
excited about and very optimistic about is

2143
02:08:59.045 --> 02:09:00.085
<v Shane Cahill (Meta)>about small businesses.

2144
02:09:00.650 --> 02:09:05.530
<v Shane Cahill (Meta)>Small businesses are the backbone of the
economy. That's so true for New York. There

2145
02:09:05.530 --> 02:09:09.130
<v Shane Cahill (Meta)>are so many small businesses.
It's truly, truly awesome.

2146
02:09:09.610 --> 02:09:12.650
<v Shane Cahill (Meta)>And we believe that the
impact will even be larger

2147
02:09:13.130 --> 02:09:14.890
<v Shane Cahill (Meta)>from from small businesses,

2148
02:09:15.455 --> 02:09:16.495
<v Shane Cahill (Meta)>which will

2149
02:09:16.894 --> 02:09:18.575
<v Shane Cahill (Meta)>have personal superintelligence

2150
02:09:18.575 --> 02:09:20.655
<v Shane Cahill (Meta)>and be empowered to really grow

2151
02:09:20.815 --> 02:09:24.655
<v Shane Cahill (Meta)>their businesses, which in turn
will lead to more jobs and growth.

2152
02:09:25.855 --> 02:09:27.214
<v Shane Cahill (Meta)>We don't believe that,

2153
02:09:27.535 --> 02:09:29.535
<v Shane Cahill (Meta)>you know, our vision for superintelligence

2154
02:09:29.880 --> 02:09:31.240
<v Shane Cahill (Meta)>is augmentation.

2155
02:09:31.640 --> 02:09:36.600
<v Shane Cahill (Meta)>It's invention, not automation. It's really
about empowering people. And this is why

2156
02:09:36.600 --> 02:09:40.280
<v Shane Cahill (Meta)>we developed news for business. I'll be
happy for small businesses. I'll be happy to

2157
02:09:40.280 --> 02:09:43.880
<v Shane Cahill (Meta)>talk a little bit more about that in
detail as we move through the day.

2158
02:09:44.555 --> 02:09:49.035
<v Shane Cahill (Meta)>I just wanted to mention as well as I was
kind of, like, reading stuff for today.

2159
02:09:49.595 --> 02:09:53.195
<v Shane Cahill (Meta)>I was reading about why
Combinator's cohort for 2026.

2160
02:09:53.195 --> 02:09:54.955
<v Shane Cahill (Meta)>There are 240

2161
02:09:55.195 --> 02:09:56.475
<v Shane Cahill (Meta)>AIDCs

2162
02:09:56.475 --> 02:09:58.395
<v Shane Cahill (Meta)>in the in the 2026

2163
02:09:58.395 --> 02:09:58.795
<v Shane Cahill (Meta)>cohort,

2164
02:09:59.290 --> 02:10:02.330
<v Shane Cahill (Meta)>and that's in New York City. So I
just thought that was pretty awesome.

2165
02:10:03.610 --> 02:10:04.490
<v Speaker Julie Menin>Yes.

2166
02:10:04.490 --> 02:10:09.450
<v Speaker Julie Menin>No one's doubting the importance of
starting these VC firms here, whether it's

2167
02:10:09.450 --> 02:10:11.210
<v Speaker Julie Menin>through Y Combinator or through

2168
02:10:12.195 --> 02:10:16.515
<v Speaker Julie Menin>Cornell Technion, which is in my district,
which we're thrilled, is incubating so

2169
02:10:16.515 --> 02:10:21.155
<v Speaker Julie Menin>many start ups. That is obviously
something we want to encourage to,

2170
02:10:21.395 --> 02:10:26.195
<v Speaker Julie Menin>encourage these businesses to locate
here in the city. I think our concern is

2171
02:10:26.195 --> 02:10:29.600
<v Speaker Julie Menin>obviously on this job displacement
since you mentioned small businesses.

2172
02:10:29.600 --> 02:10:33.920
<v Speaker Julie Menin>Unfortunately, more small businesses
have closed this past year than opened.

2173
02:10:34.160 --> 02:10:34.720
<v Speaker Julie Menin>So

2174
02:10:35.040 --> 02:10:37.840
<v Speaker Julie Menin>the situation for small
businesses is dire,

2175
02:10:38.160 --> 02:10:42.160
<v Speaker Julie Menin>and it's something obviously
that this council is taking very

2176
02:10:42.635 --> 02:10:43.755
<v Speaker Julie Menin>Miss Friend?

2177
02:10:44.875 --> 02:10:47.435
<v Alice Friend (Google)>Madam speaker, I don't have statistics

2178
02:10:47.435 --> 02:10:52.875
<v Alice Friend (Google)>for New York City in particular at
hand, and I'd be happy to ask our team,

2179
02:10:53.275 --> 02:10:58.900
<v Alice Friend (Google)>to get those to you right away. What I can
say is that we also have been investing a

2180
02:10:58.900 --> 02:11:02.260
<v Alice Friend (Google)>lot in research into
the labor impacts of AI.

2181
02:11:02.740 --> 02:11:05.220
<v Alice Friend (Google)>And from what the early
empirical evidence,

2182
02:11:05.460 --> 02:11:06.820
<v Alice Friend (Google)>that is available,

2183
02:11:07.460 --> 02:11:08.580
<v Alice Friend (Google)>from what we can see,

2184
02:11:09.165 --> 02:11:12.205
<v Alice Friend (Google)>it is likely that some
jobs will be eliminated.

2185
02:11:12.685 --> 02:11:15.405
<v Alice Friend (Google)>Some jobs will be created net new,

2186
02:11:15.885 --> 02:11:20.365
<v Alice Friend (Google)>but most jobs will change. And so the
other thing we've been doing is investing

2187
02:11:20.525 --> 02:11:21.885
<v Alice Friend (Google)>in upskilling,

2188
02:11:22.205 --> 02:11:24.925
<v Alice Friend (Google)>for workers in multiple different sectors

2189
02:11:25.140 --> 02:11:30.900
<v Alice Friend (Google)>through our AI opportunity fund at google.org
and through several other investments

2190
02:11:30.900 --> 02:11:34.580
<v Alice Friend (Google)>to ensure that we can all navigate
this transition successfully.

2191
02:11:35.380 --> 02:11:40.445
<v Speaker Julie Menin>And I'll just, close by saying the
council is investing in upskilling,

2192
02:11:40.685 --> 02:11:44.445
<v Speaker Julie Menin>making sure that workers
have those new tools,

2193
02:11:44.765 --> 02:11:46.285
<v Speaker Julie Menin>in this economy,

2194
02:11:46.765 --> 02:11:51.805
<v Speaker Julie Menin>and we are very focused on that as well.
I'm gonna pass it over to chair De La Rosa.

2195
02:11:52.230 --> 02:11:54.070
<v Speaker Julie Menin>Thank you, madam speaker.

2196
02:11:54.870 --> 02:11:57.590
<v CM Carmen De La Rosa>Many of us saw, and I
guess the world watched

2197
02:11:57.590 --> 02:12:02.470
<v CM Carmen De La Rosa>as this week, your companies and others
met at the White House, and there were some

2198
02:12:02.470 --> 02:12:06.470
<v CM Carmen De La Rosa>voluntary commitments that were made there.
Can you tell us about what your company

2199
02:12:06.470 --> 02:12:10.845
<v CM Carmen De La Rosa>specifically agreed to, and what are the
next steps? I'll start with miss Dwyer.

2200
02:12:13.325 --> 02:12:15.325
<v Morgan Dwyer (OpenAI)>Yes. So our company,

2201
02:12:15.485 --> 02:12:16.845
<v Morgan Dwyer (OpenAI)>along with many others,

2202
02:12:17.245 --> 02:12:21.325
<v Morgan Dwyer (OpenAI)>did commit to, the framework
that the White House laid out.

2203
02:12:22.050 --> 02:12:23.490
<v Morgan Dwyer (OpenAI)>But I wanna be clear,

2204
02:12:23.650 --> 02:12:26.450
<v Morgan Dwyer (OpenAI)>voluntary commitments and self regulation

2205
02:12:26.450 --> 02:12:27.490
<v Morgan Dwyer (OpenAI)>aren't enough.

2206
02:12:28.130 --> 02:12:33.090
<v Morgan Dwyer (OpenAI)>And that's why OpenAI has consistently
supported frontier safety regulation,

2207
02:12:33.570 --> 02:12:34.770
<v Morgan Dwyer (OpenAI)>at the federal level

2208
02:12:35.115 --> 02:12:38.075
<v Morgan Dwyer (OpenAI)>and why we have endorsed multiple bills

2209
02:12:38.075 --> 02:12:39.435
<v Morgan Dwyer (OpenAI)>at the state level,

2210
02:12:39.915 --> 02:12:43.675
<v Morgan Dwyer (OpenAI)>to ensure that in the absence
of binding federal regulation,

2211
02:12:44.395 --> 02:12:49.595
<v Morgan Dwyer (OpenAI)>frontier safety regulation in a harmonized
fashion continues to move forward,

2212
02:12:49.755 --> 02:12:50.795
<v Morgan Dwyer (OpenAI)>and keep people safe.

2213
02:12:51.650 --> 02:12:56.290
<v CM Carmen De La Rosa>Just to follow-up real quick. So does
that look like an independent regulators,

2214
02:12:56.290 --> 02:12:58.210
<v CM Carmen De La Rosa>mandatory audits, penalties?

2215
02:12:58.290 --> 02:12:59.090
<v CM Carmen De La Rosa>What else?

2216
02:13:01.650 --> 02:13:05.650
<v Morgan Dwyer (OpenAI)>So at the federal level, we have
outlined a couple of different

2217
02:13:06.345 --> 02:13:09.145
<v Morgan Dwyer (OpenAI)>elements that we think
would be appropriate

2218
02:13:09.145 --> 02:13:10.985
<v Morgan Dwyer (OpenAI)>for federal regulation.

2219
02:13:11.545 --> 02:13:13.225
<v Morgan Dwyer (OpenAI)>That includes mandatory

2220
02:13:13.225 --> 02:13:14.105
<v Morgan Dwyer (OpenAI)>predeployment

2221
02:13:14.105 --> 02:13:15.065
<v Morgan Dwyer (OpenAI)>evaluations

2222
02:13:15.065 --> 02:13:16.985
<v Morgan Dwyer (OpenAI)>for national security risks,

2223
02:13:17.225 --> 02:13:19.385
<v Morgan Dwyer (OpenAI)>which OpenAI does voluntarily.

2224
02:13:20.490 --> 02:13:22.650
<v Morgan Dwyer (OpenAI)>We also think it should involve

2225
02:13:22.650 --> 02:13:25.130
<v Morgan Dwyer (OpenAI)>independent technical assessments,

2226
02:13:25.450 --> 02:13:27.690
<v Morgan Dwyer (OpenAI)>that we've been talking a bit about today.

2227
02:13:28.010 --> 02:13:32.650
<v Morgan Dwyer (OpenAI)>We think those independent technical
assessments are particularly important for

2228
02:13:32.650 --> 02:13:33.450
<v Morgan Dwyer (OpenAI)>assessing

2229
02:13:33.450 --> 02:13:35.834
<v Morgan Dwyer (OpenAI)>internal risks at companies

2230
02:13:36.155 --> 02:13:40.874
<v Morgan Dwyer (OpenAI)>as well as the pace and risk of RSI.

2231
02:13:42.394 --> 02:13:44.314
<v CM Carmen De La Rosa>Thank you for your answer.

2232
02:13:44.635 --> 02:13:46.875
<v CM Carmen De La Rosa>Let's go to mister Cahill.

2233
02:13:51.000 --> 02:13:53.480
<v Shane Cahill (Meta)>Chair, thank you very much
for for the question.

2234
02:13:53.960 --> 02:13:59.720
<v Shane Cahill (Meta)>Yes. Indeed, Meta along with, with our
peers... Some of our peers signed, the

2235
02:13:59.720 --> 02:14:01.640
<v Shane Cahill (Meta)>accord, for superintelligence.

2236
02:14:03.275 --> 02:14:08.955
<v Shane Cahill (Meta)>This accord is a is a is a start and an
accord the whole industry can come behind.

2237
02:14:09.755 --> 02:14:13.355
<v Shane Cahill (Meta)>It includes, as I mentioned, I think,
in one of my previous answers,

2238
02:14:15.630 --> 02:14:19.470
<v Shane Cahill (Meta)>partnering... Commitment to partner with
independent evaluators and assessors.

2239
02:14:19.630 --> 02:14:23.710
<v Shane Cahill (Meta)>And this is something that we have also
specified in in our scaling framework.

2240
02:14:25.150 --> 02:14:29.470
<v CM Carmen De La Rosa>Are there commitments beyond those
that you agreed on that your that

2241
02:14:29.835 --> 02:14:31.515
<v CM Carmen De La Rosa>your company is willing to make?

2242
02:14:35.435 --> 02:14:37.755
<v Shane Cahill (Meta)>Beyond the Accord Council member?

2243
02:14:37.915 --> 02:14:40.075
<v CM Carmen De La Rosa>Yes. Beyond the framework.

2244
02:14:41.915 --> 02:14:47.270
<v Shane Cahill (Meta)>The the Accord is a start, as I mentioned.
So I think this is a continuing

2245
02:14:47.270 --> 02:14:48.150
<v Shane Cahill (Meta)>conversation.

2246
02:14:48.470 --> 02:14:49.510
<v CM Carmen De La Rosa>Thank you.

2247
02:14:50.230 --> 02:14:50.870
<v CM Carmen De La Rosa>Miss Bren?

2248
02:14:58.025 --> 02:14:58.585
<v Alice Friend (Google)>So,

2249
02:14:58.825 --> 02:15:00.985
<v Alice Friend (Google)>as the other witnesses noted,

2250
02:15:00.985 --> 02:15:03.865
<v Alice Friend (Google)>the accord at the White
House, which we also signed,

2251
02:15:04.025 --> 02:15:05.945
<v Alice Friend (Google)>called for internal

2252
02:15:05.945 --> 02:15:06.825
<v Alice Friend (Google)>controls,

2253
02:15:07.545 --> 02:15:09.145
<v Alice Friend (Google)>partnering with external

2254
02:15:09.145 --> 02:15:11.065
<v Alice Friend (Google)>auditors and evaluators

2255
02:15:11.065 --> 02:15:16.100
<v Alice Friend (Google)>and, ongoing monitoring systems, all of
which we think is very healthy and much of

2256
02:15:16.100 --> 02:15:18.100
<v Alice Friend (Google)>which, we already practice.

2257
02:15:18.420 --> 02:15:21.620
<v Alice Friend (Google)>I'll also note that, towards
the end of the accord,

2258
02:15:22.260 --> 02:15:24.820
<v Alice Friend (Google)>it observes that it may
be necessary for these

2259
02:15:26.355 --> 02:15:28.995
<v Alice Friend (Google)>commitments to become part of law.

2260
02:15:29.635 --> 02:15:32.115
<v Alice Friend (Google)>And as I said at the top, we have long,

2261
02:15:32.595 --> 02:15:35.955
<v Alice Friend (Google)>worked towards a federal
framework for AI safety.

2262
02:15:37.410 --> 02:15:38.850
<v CM Carmen De La Rosa>Thank you, mister Graham.

2263
02:15:45.170 --> 02:15:46.210
<v Logan Graham (Anthropic)>Yeah. Likewise,

2264
02:15:46.610 --> 02:15:47.810
<v Logan Graham (Anthropic)>the accords

2265
02:15:47.810 --> 02:15:52.535
<v Logan Graham (Anthropic)>to my understanding from what's
public calls for security standards,

2266
02:15:52.695 --> 02:15:54.055
<v Logan Graham (Anthropic)>internal controls,

2267
02:15:54.775 --> 02:15:56.455
<v Logan Graham (Anthropic)>independent evaluation.

2268
02:15:57.015 --> 02:15:59.815
<v Logan Graham (Anthropic)>We support that and have
for a very long time.

2269
02:16:00.695 --> 02:16:01.495
<v Logan Graham (Anthropic)>We

2270
02:16:01.735 --> 02:16:05.800
<v Logan Graham (Anthropic)>also obviously think this needs to go further.
We've supported regulation, and we've

2271
02:16:05.800 --> 02:16:06.440
<v Logan Graham (Anthropic)>proposed

2272
02:16:07.080 --> 02:16:10.840
<v Logan Graham (Anthropic)>our view on regulation in our advanced AI
framework, which details the need to have

2273
02:16:10.840 --> 02:16:12.280
<v Logan Graham (Anthropic)>safety frameworks,

2274
02:16:12.680 --> 02:16:16.040
<v Logan Graham (Anthropic)>testing, and what kinds,
security standards, safeguards

2275
02:16:17.025 --> 02:16:19.905
<v Logan Graham (Anthropic)>to encode the need for
independent evaluators.

2276
02:16:20.705 --> 02:16:22.625
<v Logan Graham (Anthropic)>We are now trialing our own

2277
02:16:23.105 --> 02:16:24.065
<v Logan Graham (Anthropic)>to help...

2278
02:16:24.465 --> 02:16:25.025
<v Logan Graham (Anthropic)>So

2279
02:16:25.425 --> 02:16:28.945
<v Logan Graham (Anthropic)>advance the science of independent
auditors as well as safety processes

2280
02:16:29.370 --> 02:16:32.730
<v Logan Graham (Anthropic)>as well as investing in
cyber and biodefense.

2281
02:16:33.290 --> 02:16:35.450
<v Logan Graham (Anthropic)>And in absence of federal

2282
02:16:35.530 --> 02:16:40.330
<v Logan Graham (Anthropic)>legislation here, we've also supported
safety bills at the state level for audits

2283
02:16:40.330 --> 02:16:42.410
<v Logan Graham (Anthropic)>and evals of this kind to accomplish this.

2284
02:16:44.925 --> 02:16:46.365
<v CM Carmen De La Rosa>Thank you for your answer.

2285
02:16:46.685 --> 02:16:47.485
<v CM Carmen De La Rosa>Earlier,

2286
02:16:47.725 --> 02:16:49.564
<v CM Carmen De La Rosa>you all testified

2287
02:16:49.564 --> 02:16:53.325
<v CM Carmen De La Rosa>that, to your knowledge, no
whistleblowers had been fired

2288
02:16:53.325 --> 02:16:54.686
<v CM Carmen De La Rosa>directly for

2289
02:16:55.085 --> 02:16:56.205
<v CM Carmen De La Rosa>being whistleblowers,

2290
02:16:56.365 --> 02:17:02.741
<v CM Carmen De La Rosa>from your company. So my question is
kind of the opposite of that. Has

2291
02:17:02.900 --> 02:17:06.741
<v CM Carmen De La Rosa>anyone been fired for building
models that it cannot control,

2292
02:17:07.381 --> 02:17:10.820
<v CM Carmen De La Rosa>or for those models hacking into
businesses or government systems?

2293
02:17:17.115 --> 02:17:18.796
<v CM Carmen De La Rosa>Miss Dwyer, you can begin.

2294
02:17:19.275 --> 02:17:20.155
<v Morgan Dwyer (OpenAI)>Thank you.

2295
02:17:21.675 --> 02:17:26.150
<v Morgan Dwyer (OpenAI)>So I I don't know, about HR processes,

2296
02:17:26.630 --> 02:17:31.590
<v Morgan Dwyer (OpenAI)>that have been made or decisions related
to HR that have been made. What I can tell

2297
02:17:31.590 --> 02:17:32.870
<v Morgan Dwyer (OpenAI)>you is that,

2298
02:17:33.189 --> 02:17:37.670
<v Morgan Dwyer (OpenAI)>after the hugging face incident, which
I think you may be referring to, OpenAI

2299
02:17:37.670 --> 02:17:38.630
<v Morgan Dwyer (OpenAI)>instituted

2300
02:17:39.855 --> 02:17:44.016
<v Morgan Dwyer (OpenAI)>significant changes to our
overall organizational process.

2301
02:17:44.415 --> 02:17:46.256
<v Morgan Dwyer (OpenAI)>That includes clarifying

2302
02:17:47.136 --> 02:17:48.336
<v Morgan Dwyer (OpenAI)>escalation

2303
02:17:48.336 --> 02:17:50.656
<v Morgan Dwyer (OpenAI)>pathways and making them simpler.

2304
02:17:51.136 --> 02:17:53.296
<v Morgan Dwyer (OpenAI)>That includes clarifying responsibilities

2305
02:17:53.375 --> 02:17:54.920
<v Morgan Dwyer (OpenAI)>across teams

2306
02:17:54.920 --> 02:17:57.160
<v Morgan Dwyer (OpenAI)>and creating clearer processes,

2307
02:17:57.560 --> 02:18:02.280
<v Morgan Dwyer (OpenAI)>for teams to make decisions to
pause or to restart activities.

2308
02:18:03.080 --> 02:18:06.680
<v CM Carmen De La Rosa>So to your knowledge, no one was
fired after hugging face incident?

2309
02:18:08.625 --> 02:18:11.265
<v CM Carmen De La Rosa [?]>Yes or no? I I

2310
02:18:11.265 --> 02:18:15.505
<v Morgan Dwyer (OpenAI)>do not have any knowledge of
HR activities within OpenAI.

2311
02:18:15.505 --> 02:18:17.185
<v Morgan Dwyer (OpenAI)>I lead our policy team.

2312
02:18:18.145 --> 02:18:18.545
<v CM Carmen De La Rosa>Okay.

2313
02:18:20.530 --> 02:18:21.730
<v CM Carmen De La Rosa>Mister Cahill?

2314
02:18:23.570 --> 02:18:24.131
<v Shane Cahill (Meta)>Thank

2315
02:18:24.450 --> 02:18:27.490
<v Shane Cahill (Meta)>you, chair. I also lead our AI policy,

2316
02:18:27.890 --> 02:18:31.811
<v Shane Cahill (Meta)>at at MedA, and I'm not the right person
to respond to your specific question, but

2317
02:18:31.811 --> 02:18:35.410
<v Shane Cahill (Meta)>I'd be happy to have, our teams
follow-up with you afterwards.

2318
02:18:36.025 --> 02:18:39.785
<v CM Carmen De La Rosa>It would be great if all of the companies
could provide that information to us.

2319
02:18:40.185 --> 02:18:43.785
<v CM Carmen De La Rosa>It's... It would be useful
to our policy deliberations.

2320
02:18:43.945 --> 02:18:44.905
<v CM Carmen De La Rosa>Miss Friend?

2321
02:18:46.825 --> 02:18:51.721
<v Alice Friend (Google)>Yes, ma'am. Same answer. I'm not made
aware of any particular personnel actions

2322
02:18:51.721 --> 02:18:55.561
<v Alice Friend (Google)>inside the company, but can also
pledge to follow-up with you all.

2323
02:18:57.961 --> 02:19:00.601
<v CM Carmen De La Rosa>Thank you. I... It seems like we lost,

2324
02:19:01.080 --> 02:19:03.561
<v CM Carmen De La Rosa>mister Graham. I'm not sure if he's on.

2325
02:19:07.435 --> 02:19:09.035
<v CM Carmen De La Rosa>Seems like he's not. Okay.

2326
02:19:09.435 --> 02:19:13.915
<v CM Carmen De La Rosa>I'm going to ask one more question, and
then we're gonna pass it on to members.

2327
02:19:13.915 --> 02:19:15.835
<v CM Carmen De La Rosa>Chair, he he is online. Sorry.

2328
02:19:16.314 --> 02:19:17.035
<v CM Carmen De La Rosa>Mister Graham?

2329
02:19:20.570 --> 02:19:24.331
<v Logan Graham (Anthropic)>Hi there. Apologies. We're just
trying to fix the video in the room.

2330
02:19:24.891 --> 02:19:29.610
<v Logan Graham (Anthropic)>Likewise, I'm I'm not aware, I'm not the
the best person to ask on that. We have a

2331
02:19:29.610 --> 02:19:31.610
<v Logan Graham (Anthropic)>detailed whistleblowing policy,

2332
02:19:32.226 --> 02:19:36.466
<v Logan Graham (Anthropic)>for example, to to handle this,
but we can have our team follow-up.

2333
02:19:37.266 --> 02:19:37.985
<v CM Carmen De La Rosa>Thank you.

2334
02:19:38.545 --> 02:19:42.065
<v CM Carmen De La Rosa>My last question before I start to
turn over to my colleagues. You know,

2335
02:19:42.386 --> 02:19:47.330
<v CM Carmen De La Rosa>we are concerned. I am concerned as a
technology chair about public infrastructure

2336
02:19:47.489 --> 02:19:50.290
<v CM Carmen De La Rosa>to make our communities literate on AI,

2337
02:19:50.850 --> 02:19:55.649
<v CM Carmen De La Rosa>and have a better understanding of how AI
impacts their life and also to safeguard

2338
02:19:55.649 --> 02:19:56.290
<v CM Carmen De La Rosa>them,

2339
02:19:56.689 --> 02:19:58.530
<v CM Carmen De La Rosa>from the proliferation of AI.

2340
02:19:59.090 --> 02:20:00.255
<v CM Carmen De La Rosa>Your companies

2341
02:20:00.255 --> 02:20:01.936
<v CM Carmen De La Rosa>and your CEOs

2342
02:20:02.175 --> 02:20:02.976
<v CM Carmen De La Rosa>have

2343
02:20:03.295 --> 02:20:04.175
<v CM Carmen De La Rosa>captured,

2344
02:20:04.735 --> 02:20:06.495
<v CM Carmen De La Rosa>the top percent of

2345
02:20:07.055 --> 02:20:09.056
<v CM Carmen De La Rosa>the the world's incomes,

2346
02:20:09.854 --> 02:20:11.215
<v CM Carmen De La Rosa>in my opinion,

2347
02:20:11.774 --> 02:20:13.455
<v CM Carmen De La Rosa>widening the wage gap

2348
02:20:13.820 --> 02:20:16.860
<v CM Carmen De La Rosa>and the income inequality
in our city and throughout.

2349
02:20:17.260 --> 02:20:19.261
<v CM Carmen De La Rosa>Do your companies have,

2350
02:20:20.460 --> 02:20:22.381
<v CM Carmen De La Rosa>a mission? Do they have

2351
02:20:22.780 --> 02:20:24.461
<v CM Carmen De La Rosa>any type of,

2352
02:20:24.780 --> 02:20:26.061
<v CM Carmen De La Rosa>understanding

2353
02:20:26.141 --> 02:20:31.564
<v CM Carmen De La Rosa>of the impacts that those wage gaps
have in communities? And specifically,

2354
02:20:31.885 --> 02:20:33.005
<v CM Carmen De La Rosa>are there any

2355
02:20:33.404 --> 02:20:38.125
<v CM Carmen De La Rosa>resources that are being deployed in
order to help stand up infrastructure

2356
02:20:38.125 --> 02:20:40.285
<v CM Carmen De La Rosa>in cities, municipalities,

2357
02:20:40.444 --> 02:20:42.285
<v CM Carmen De La Rosa>states where you all are

2358
02:20:42.524 --> 02:20:43.164
<v CM Carmen De La Rosa>headquartered?

2359
02:20:47.960 --> 02:20:49.880
<v CM Carmen De La Rosa>We can start with Ms. Dwyer.

2360
02:20:52.359 --> 02:20:55.159
<v Morgan Dwyer (OpenAI)>So I will answer your
question in two ways. First,

2361
02:20:56.215 --> 02:20:59.016
<v Morgan Dwyer (OpenAI)>to your point on AI literacy

2362
02:20:59.016 --> 02:21:00.615
<v Morgan Dwyer (OpenAI)>and education,

2363
02:21:00.855 --> 02:21:03.415
<v Morgan Dwyer (OpenAI)>education is core to opportunity.

2364
02:21:03.896 --> 02:21:08.535
<v Morgan Dwyer (OpenAI)>And that's why OpenAI is proud to have
partnered with the American Federation of

2365
02:21:08.535 --> 02:21:09.016
<v Morgan Dwyer (OpenAI)>Teachers.

2366
02:21:09.660 --> 02:21:11.980
<v Morgan Dwyer (OpenAI)>We have a $10,000,000 partnership,

2367
02:21:12.140 --> 02:21:14.860
<v Morgan Dwyer (OpenAI)>including a training
facility in New York City,

2368
02:21:15.420 --> 02:21:19.260
<v Morgan Dwyer (OpenAI)>because we believe first that
teachers need to be prepared,

2369
02:21:19.900 --> 02:21:23.580
<v Morgan Dwyer (OpenAI)>to introduce AI literacy
and AI responsibility

2370
02:21:24.780 --> 02:21:26.195
<v Morgan Dwyer (OpenAI)>into the classroom.

2371
02:21:26.435 --> 02:21:31.715
<v Morgan Dwyer (OpenAI)>Look. Teachers should remain in charge, and
so we want them to be empowered to teach

2372
02:21:31.715 --> 02:21:33.075
<v Morgan Dwyer (OpenAI)>the next generation

2373
02:21:33.314 --> 02:21:36.515
<v Morgan Dwyer (OpenAI)>how to use AI safely and responsibly.

2374
02:21:37.074 --> 02:21:40.354
<v Morgan Dwyer (OpenAI)>To your broader question
about economic opportunity,

2375
02:21:41.301 --> 02:21:46.580
<v Morgan Dwyer (OpenAI)>I'd be happy to follow-up with, some
policy proposals that OpenAI recently put

2376
02:21:46.580 --> 02:21:47.301
<v Morgan Dwyer (OpenAI)>forward

2377
02:21:47.381 --> 02:21:48.261
<v Morgan Dwyer (OpenAI)>that address,

2378
02:21:48.421 --> 02:21:49.860
<v Morgan Dwyer (OpenAI)>exactly that question.

2379
02:21:50.181 --> 02:21:50.820
<v Morgan Dwyer (OpenAI)>Look.

2380
02:21:51.461 --> 02:21:54.341
<v Morgan Dwyer (OpenAI)>A future where AI exacerbates

2381
02:21:54.341 --> 02:21:55.860
<v Morgan Dwyer (OpenAI)>economic inequality

2382
02:21:56.065 --> 02:21:58.386
<v Morgan Dwyer (OpenAI)>is not a future that we want.

2383
02:21:59.105 --> 02:22:03.025
<v Morgan Dwyer (OpenAI)>We have started thinking through
creative policy proposals,

2384
02:22:03.505 --> 02:22:08.705
<v Morgan Dwyer (OpenAI)>to help shift the trajectory of technology
and how it's impacting the economy.

2385
02:22:09.570 --> 02:22:12.130
<v Morgan Dwyer (OpenAI)>And we are committed to
working with policymakers

2386
02:22:12.130 --> 02:22:14.290
<v Morgan Dwyer (OpenAI)>to begin piloting those ideas,

2387
02:22:14.530 --> 02:22:16.530
<v Morgan Dwyer (OpenAI)>and putting them into practice.

2388
02:22:17.649 --> 02:22:18.930
<v CM Carmen De La Rosa>Thank you. Mister Cahill?

2389
02:22:21.465 --> 02:22:24.185
<v Shane Cahill (Meta)>Chair, thank you very much. This is such

2390
02:22:24.185 --> 02:22:25.226
<v Shane Cahill (Meta)>an important

2391
02:22:25.305 --> 02:22:31.705
<v Shane Cahill (Meta)>topic, such an important question. So thank
you for for asking it. I listened with

2392
02:22:31.705 --> 02:22:36.601
<v Shane Cahill (Meta)>great interest to your opening remarks
where you spoke about digital divide and AI

2393
02:22:36.601 --> 02:22:41.320
<v Shane Cahill (Meta)>literacy and making sure that New Yorkers
fully benefit from this technology.

2394
02:22:41.880 --> 02:22:44.761
<v Shane Cahill (Meta)>Really couldn't agree with
you more on on all of that.

2395
02:22:45.320 --> 02:22:49.480
<v Shane Cahill (Meta)>Meta's vision is to put
personal superintelligence

2396
02:22:49.480 --> 02:22:51.000
<v Shane Cahill (Meta)>into the hands of everyone,

2397
02:22:51.535 --> 02:22:56.176
<v Shane Cahill (Meta)>all New Yorkers that want to use it
to pursue their own aspirations.

2398
02:22:56.895 --> 02:23:01.535
<v Shane Cahill (Meta)>Our Muse product, for example, is free.
And I mentioned our our Muse for small

2399
02:23:01.535 --> 02:23:03.295
<v Shane Cahill (Meta)>business, a little earlier, and

2400
02:23:03.695 --> 02:23:06.175
<v Shane Cahill (Meta)>we... We'll be happy to talk
a little bit more about that

2401
02:23:06.851 --> 02:23:12.131
<v Shane Cahill (Meta)>as the day moves on. I also wanted
to mention as well that we have

2402
02:23:12.530 --> 02:23:17.170
<v Shane Cahill (Meta)>our work... America's Workforce Academy,
which we're investing in, and also our

2403
02:23:17.170 --> 02:23:18.610
<v Shane Cahill (Meta)>futures for everyone fund.

2404
02:23:19.256 --> 02:23:24.855
<v Shane Cahill (Meta)>So those are just two two two examples I
wanted to to reference as well. Thank you.

2405
02:23:26.375 --> 02:23:28.216
<v CM Carmen De La Rosa>Thank you. Miss Bren?

2406
02:23:29.815 --> 02:23:31.976
<v Alice Friend (Google)>Yes, ma'am. As I mentioned earlier,

2407
02:23:32.296 --> 02:23:38.000
<v Alice Friend (Google)>Google has investments in AI upscaling to
address exactly this challenge of ensuring

2408
02:23:38.000 --> 02:23:38.561
<v Alice Friend (Google)>that,

2409
02:23:39.359 --> 02:23:44.640
<v Alice Friend (Google)>current generations and the next generation
are well equipped to benefit from AI and

2410
02:23:44.640 --> 02:23:45.200
<v Alice Friend (Google)>to,

2411
02:23:45.680 --> 02:23:48.560
<v Alice Friend (Google)>leverage it in their careers
and in their personal lives.

2412
02:23:49.176 --> 02:23:53.176
<v Alice Friend (Google)>The mission of Google is to organize
the world's information and to make it

2413
02:23:53.176 --> 02:23:55.576
<v Alice Friend (Google)>universally accessible and useful.

2414
02:23:56.296 --> 02:23:59.256
<v Alice Friend (Google)>Our approach to AI is very
consistent with this.

2415
02:23:59.655 --> 02:24:03.655
<v Alice Friend (Google)>The other thing I'll note is
that, again, early research into

2416
02:24:04.311 --> 02:24:10.391
<v Alice Friend (Google)>impacts of AI on labor demonstrate
that AI tools actually upscale those

2417
02:24:11.590 --> 02:24:16.070
<v Alice Friend (Google)>who are at the lower end of the wage scale,
that in fact it helps people move up the

2418
02:24:16.070 --> 02:24:16.950
<v Alice Friend (Google)>value chain

2419
02:24:17.535 --> 02:24:20.815
<v Alice Friend (Google)>because it can boost people's expertise.

2420
02:24:21.135 --> 02:24:24.255
<v Alice Friend (Google)>So we're very optimistic about the ability

2421
02:24:24.975 --> 02:24:27.695
<v Alice Friend (Google)>of of everyday Americans

2422
02:24:27.854 --> 02:24:28.814
<v Alice Friend (Google)>to really leverage

2423
02:24:29.220 --> 02:24:33.380
<v Alice Friend (Google)>AI tools, for their own economic
benefit as well as personal benefit.

2424
02:24:34.180 --> 02:24:35.700
<v CM Carmen De La Rosa>Thank you. Mister Graham?

2425
02:24:39.540 --> 02:24:44.944
<v Logan Graham (Anthropic)>Yeah. Likewise, first, we believe in in
two things are very important to happen

2426
02:24:44.944 --> 02:24:48.785
<v Logan Graham (Anthropic)>here. The first is the science
of the economic impacts of AI,

2427
02:24:49.345 --> 02:24:51.585
<v Logan Graham (Anthropic)>including on labor, wages,

2428
02:24:52.545 --> 02:24:56.465
<v Logan Graham (Anthropic)>and and the like. And we have tried to
build what I think is the world's best team

2429
02:24:56.465 --> 02:24:59.170
<v Logan Graham (Anthropic)>on this and publish as much as we can.

2430
02:24:59.490 --> 02:25:01.730
<v Logan Graham (Anthropic)>In addition, my
understanding is we do have

2431
02:25:02.210 --> 02:25:07.811
<v Logan Graham (Anthropic)>training on using and speeding up on on AI
for exactly this reason. We we share that

2432
02:25:07.970 --> 02:25:12.495
<v Logan Graham (Anthropic)>that mission very deeply. And, I will I
will say again, we are troubleshooting the

2433
02:25:12.495 --> 02:25:14.655
<v Logan Graham (Anthropic)>video, and I will be back
with you very shortly.

2434
02:25:16.335 --> 02:25:19.695
<v CM Carmen De La Rosa>Thank you. I'm sure you have the best
in the nation helping you out there.

2435
02:25:20.655 --> 02:25:21.135
<v CM Carmen De La Rosa>So

2436
02:25:21.710 --> 02:25:26.029
<v CM Carmen De La Rosa>we are going to move on
to member questions.

2437
02:25:26.029 --> 02:25:29.470
<v CM Carmen De La Rosa>I wanna remind members that there'll be
about a minute to ask your questions,

2438
02:25:29.710 --> 02:25:31.390
<v CM Carmen De La Rosa>but there will be multiple rounds.

2439
02:25:32.189 --> 02:25:32.990
<v CM Carmen De La Rosa>We

2440
02:25:33.069 --> 02:25:36.915
<v CM Carmen De La Rosa>are asking members to try to ask all your
questions first and then wait for the

2441
02:25:36.915 --> 02:25:40.436
<v CM Carmen De La Rosa>response, and we have about
30 members on round one.

2442
02:25:40.835 --> 02:25:43.476
<v CM Carmen De La Rosa>We will start with our
deputy speaker, Williams.

2443
02:25:45.955 --> 02:25:46.676
<v CM Nantasha Williams>Hello.

2444
02:25:46.915 --> 02:25:49.155
<v CM Nantasha Williams>I'm gonna just jump right into a question

2445
02:25:49.556 --> 02:25:50.516
<v CM Nantasha Williams>on my bill.

2446
02:25:51.190 --> 02:25:56.471
<v CM Nantasha Williams>So it has to do with unauthorized depictions
of officials by AI. What steps have you

2447
02:25:56.471 --> 02:26:01.511
<v CM Nantasha Williams>taken to ensure that your AI tools cannot
be used to create deep fakes of public

2448
02:26:01.511 --> 02:26:03.670
<v CM Nantasha Williams>officials or to spread misinformation?

2449
02:26:07.745 --> 02:26:10.225
<v CM Nantasha Williams>Anyone, I guess, missus Dwyer?

2450
02:26:11.265 --> 02:26:13.104
<v Morgan Dwyer (OpenAI)>We're gonna keep starting with me.

2451
02:26:13.505 --> 02:26:17.984
<v Morgan Dwyer (OpenAI)>I'm happy to answer that question. Thank
you, for that, and thank you for your bill.

2452
02:26:18.739 --> 02:26:20.100
<v Morgan Dwyer (OpenAI)>Look. OpenAI,

2453
02:26:20.260 --> 02:26:24.580
<v Morgan Dwyer (OpenAI)>does not believe and... Well, let
me back up. OpenAI's usage policies

2454
02:26:25.140 --> 02:26:28.660
<v Morgan Dwyer (OpenAI)>prohibit the use of our tools
for election interference.

2455
02:26:30.565 --> 02:26:34.485
<v Morgan Dwyer (OpenAI)>We also believe it is very
important that individuals

2456
02:26:34.485 --> 02:26:39.605
<v Morgan Dwyer (OpenAI)>know where their content came from
and whether it was AI generated.

2457
02:26:40.085 --> 02:26:43.125
<v Morgan Dwyer (OpenAI)>And that's why for all
of our audio visual,

2458
02:26:43.525 --> 02:26:47.340
<v Morgan Dwyer (OpenAI)>content, we tag it with
what's called provenance.

2459
02:26:47.580 --> 02:26:51.660
<v Morgan Dwyer (OpenAI)>That's a technical term, but it
means like a watermark or metadata,

2460
02:26:52.060 --> 02:26:55.420
<v Morgan Dwyer (OpenAI)>so that people know the
content is AI generated.

2461
02:26:55.980 --> 02:26:59.260
<v Morgan Dwyer (OpenAI)>And we also make publicly
available a tool,

2462
02:26:59.740 --> 02:27:04.935
<v Morgan Dwyer (OpenAI)>that allows the public to test
whether content was AI generated.

2463
02:27:09.335 --> 02:27:10.615
<v CM Nantasha Williams>Mister Graham?

2464
02:27:14.295 --> 02:27:14.694
<v Logan Graham (Anthropic)>So

2465
02:27:15.720 --> 02:27:20.439
<v Logan Graham (Anthropic)>Claude... Our our model Claude is a
language model, and so it doesn't generate

2466
02:27:20.439 --> 02:27:21.160
<v Logan Graham (Anthropic)>images.

2467
02:27:21.960 --> 02:27:23.080
<v Logan Graham (Anthropic)>Therefore, we

2468
02:27:23.560 --> 02:27:25.880
<v Logan Graham (Anthropic)>sort of don't, so handle

2469
02:27:25.960 --> 02:27:27.400
<v Logan Graham (Anthropic)>deepfakes as a result.

2470
02:27:28.120 --> 02:27:29.000
<v Logan Graham (Anthropic)>That said,

2471
02:27:29.319 --> 02:27:31.399
<v Logan Graham (Anthropic)>we think it's important for
a number of other reasons

2472
02:27:31.745 --> 02:27:35.505
<v Logan Graham (Anthropic)>that we can get into to do
what we call watermarking

2473
02:27:35.505 --> 02:27:35.985
<v Logan Graham (Anthropic)>of,

2474
02:27:36.385 --> 02:27:40.545
<v Logan Graham (Anthropic)>text in order to identify if
our models themselves are being,

2475
02:27:40.945 --> 02:27:45.025
<v Logan Graham (Anthropic)>used and misused in other ways. But,
again, we we don't generate images,

2476
02:27:45.425 --> 02:27:47.025
<v Logan Graham (Anthropic)>for, in part for this reason.

2477
02:27:49.130 --> 02:27:51.851
<v CM Nantasha Williams>Mister Cahill followed by miss Friend.

2478
02:27:54.410 --> 02:27:57.210
<v Shane Cahill (Meta)>Deputy speaker, thank you
very much for this question.

2479
02:27:59.050 --> 02:28:05.105
<v Shane Cahill (Meta)>At Meta, if content is generated by
one of our AI tools, we will also embed

2480
02:28:05.585 --> 02:28:06.465
<v Shane Cahill (Meta)>metadata

2481
02:28:06.465 --> 02:28:11.905
<v Shane Cahill (Meta)>and attach watermarking to it so that
people can identify if it is AI generated.

2482
02:28:12.705 --> 02:28:13.505
<v Shane Cahill (Meta)>Similarly,

2483
02:28:13.665 --> 02:28:15.025
<v Shane Cahill (Meta)>if we detect

2484
02:28:16.305 --> 02:28:17.745
<v Shane Cahill (Meta)>watermarking and metadata

2485
02:28:18.290 --> 02:28:21.570
<v Shane Cahill (Meta)>on our platforms, we will
attach a label to it also.

2486
02:28:26.529 --> 02:28:27.490
<v CM Nantasha Williams>Sven?

2487
02:28:30.050 --> 02:28:30.850
<v Alice Friend (Google)>Yes, ma'am.

2488
02:28:31.625 --> 02:28:36.025
<v Alice Friend (Google)>At Google, we are very proud
to have invented our Synth ID

2489
02:28:36.745 --> 02:28:40.985
<v Alice Friend (Google)>watermarking feature, which many
across the industry have also adopted.

2490
02:28:41.465 --> 02:28:45.945
<v Alice Friend (Google)>We are also part of the Content
Coalition for Providence and Authenticity

2491
02:28:46.811 --> 02:28:47.770
<v Alice Friend (Google)>standard,

2492
02:28:47.770 --> 02:28:50.490
<v Alice Friend (Google)>which allows users to understand

2493
02:28:50.650 --> 02:28:53.370
<v Alice Friend (Google)>the origin of material they see online.

2494
02:28:54.010 --> 02:28:57.530
<v Alice Friend (Google)>We also require on our
YouTube platform that

2495
02:28:58.010 --> 02:29:01.851
<v Alice Friend (Google)>significant use of
generative AI in content

2496
02:29:02.195 --> 02:29:04.355
<v Alice Friend (Google)>be labeled as such by

2497
02:29:04.995 --> 02:29:06.115
<v Alice Friend (Google)>our creators.

2498
02:29:06.435 --> 02:29:11.635
<v Alice Friend (Google)>And we were also the first major company
to require that election advertisers

2499
02:29:11.635 --> 02:29:17.020
<v Alice Friend (Google)>also prominently disclose when they
are using generative AI technologies

2500
02:29:17.020 --> 02:29:18.700
<v Alice Friend (Google)>in their campaign ads.

2501
02:29:19.181 --> 02:29:23.820
<v Alice Friend (Google)>All this put together means that we care
very much about information integrity,

2502
02:29:23.980 --> 02:29:24.780
<v Alice Friend (Google)>across

2503
02:29:24.860 --> 02:29:27.581
<v Alice Friend (Google)>the web and across our
own products and services.

2504
02:29:28.061 --> 02:29:29.341
<v Alice Friend (Google)>And we continually

2505
02:29:29.341 --> 02:29:33.176
<v Alice Friend (Google)>invent and improve technologies
to help our users understand,

2506
02:29:33.735 --> 02:29:36.936
<v Alice Friend (Google)>when they are seeing
generative AI content,

2507
02:29:37.415 --> 02:29:38.855
<v Alice Friend (Google)>and also understand,

2508
02:29:39.175 --> 02:29:41.015
<v Alice Friend (Google)>the origins of that content.

2509
02:29:43.860 --> 02:29:46.421
<v CM Nantasha Williams>Okay. Thanks. Just one more question.

2510
02:29:47.061 --> 02:29:49.780
<v CM Nantasha Williams>So as chair of the committee
on cultural affairs,

2511
02:29:50.500 --> 02:29:53.221
<v CM Nantasha Williams>I also wanna look at this issue through
the lens of New York City's artists and

2512
02:29:53.221 --> 02:29:57.685
<v CM Nantasha Williams>creative workers. So what protections should
exist when an artist's voice, likeness,

2513
02:29:57.685 --> 02:30:02.565
<v CM Nantasha Williams>performance, or creative work can be
replicated by AI with their consent?

2514
02:30:02.725 --> 02:30:07.685
<v CM Nantasha Williams>So how do we embrace useful applications
without devaluing or displacing the artists

2515
02:30:07.685 --> 02:30:10.485
<v CM Nantasha Williams>and creative workers who
make this economy possible?

2516
02:30:14.390 --> 02:30:18.790
<v CM Nantasha Williams>So I will start, I guess. So I was gonna
let you off the hook and start with Oh,

2517
02:30:18.790 --> 02:30:20.390
<v CM Nantasha Williams>thank you. Mister Graham,

2518
02:30:20.710 --> 02:30:24.390
<v CM Nantasha Williams>who still has some video
technical difficulties. I

2519
02:30:24.870 --> 02:30:28.386
<v Logan Graham (Anthropic)>assure you we we have a whole team here
working on this. I'm... I can't wait to be

2520
02:30:28.386 --> 02:30:29.186
<v Logan Graham (Anthropic)>back with you.

2521
02:30:29.825 --> 02:30:32.386
<v Logan Graham (Anthropic)>Yeah. This is this is
very, very important.

2522
02:30:33.105 --> 02:30:37.506
<v Logan Graham (Anthropic)>I think, you know, personally, I think the
the risk of this is a bit more minimized

2523
02:30:37.506 --> 02:30:40.146
<v Logan Graham (Anthropic)>when we just have a a text
model, but it's really important

2524
02:30:41.320 --> 02:30:42.041
<v Logan Graham (Anthropic)>to,

2525
02:30:42.440 --> 02:30:46.681
<v Logan Graham (Anthropic)>you know, be very sensitive about this
issue. I think while it's, you know, it's

2526
02:30:46.681 --> 02:30:50.201
<v Logan Graham (Anthropic)>outside of my area of expertise
as a technical safety researcher,

2527
02:30:50.841 --> 02:30:55.515
<v Logan Graham (Anthropic)>I do think that this might be an avenue
where regulation or otherwise might play a

2528
02:30:55.515 --> 02:30:57.275
<v Logan Graham (Anthropic)>role in at Anthropic. We

2529
02:30:57.595 --> 02:31:01.595
<v Logan Graham (Anthropic)>we we are compliant with
with with all laws there.

2530
02:31:02.234 --> 02:31:04.875
<v Logan Graham (Anthropic)>I shared that, you know, personally as as,

2531
02:31:05.595 --> 02:31:10.070
<v Logan Graham (Anthropic)>you know, an artistic person and, you know,
friend of many artists for this reason.

2532
02:31:11.190 --> 02:31:14.391
<v Logan Graham (Anthropic)>But, yeah, I I... We would
love to engage on it further.

2533
02:31:17.110 --> 02:31:18.391
<v CM Nantasha Williams>Mister Kehoe?

2534
02:31:20.230 --> 02:31:22.150
<v Shane Cahill (Meta)>Thank you very much, deputy speaker.

2535
02:31:23.315 --> 02:31:24.996
<v Shane Cahill (Meta)>Meta takes creativity,

2536
02:31:25.155 --> 02:31:26.035
<v Shane Cahill (Meta)>copyright,

2537
02:31:26.035 --> 02:31:27.716
<v Shane Cahill (Meta)>IP rights very seriously,

2538
02:31:28.355 --> 02:31:30.756
<v Shane Cahill (Meta)>and our practices are consistent

2539
02:31:30.915 --> 02:31:31.956
<v Shane Cahill (Meta)>with law.

2540
02:31:32.275 --> 02:31:33.955
<v Shane Cahill (Meta)>I also share

2541
02:31:35.790 --> 02:31:40.670
<v Shane Cahill (Meta)>the the views of the rest of the the
witnesses just in relation to the leadership

2542
02:31:40.670 --> 02:31:43.870
<v Shane Cahill (Meta)>and the constructive voice that you are
bringing to this conversation as well and

2543
02:31:43.870 --> 02:31:45.070
<v Shane Cahill (Meta)>that of your colleagues.

2544
02:31:47.950 --> 02:31:50.270
<v CM Nantasha Williams>Miss Friend and lastly, miss Dwyer.

2545
02:31:52.205 --> 02:31:53.245
<v Alice Friend (Google)>Yes, ma'am.

2546
02:31:55.245 --> 02:31:59.245
<v Alice Friend (Google)>For us, it's always important
to keep in mind that we,

2547
02:31:59.485 --> 02:32:02.845
<v Alice Friend (Google)>design a lot of AI to
be useful to creators.

2548
02:32:03.660 --> 02:32:06.540
<v Alice Friend (Google)>So we have many, many
different creative tools

2549
02:32:06.699 --> 02:32:08.699
<v Alice Friend (Google)>across our AI offerings

2550
02:32:08.699 --> 02:32:10.380
<v Alice Friend (Google)>that is designed to really

2551
02:32:11.899 --> 02:32:15.899
<v Alice Friend (Google)>be something that creators can leverage.
So you see that everywhere from our tools

2552
02:32:15.899 --> 02:32:17.255
<v Alice Friend (Google)>to our YouTube

2553
02:32:17.255 --> 02:32:24.135
<v Alice Friend (Google)>platform as well. That said, we also think
that as part of sharing the benefit of AI

2554
02:32:24.135 --> 02:32:27.015
<v Alice Friend (Google)>with creators, it's important
that they have control.

2555
02:32:27.255 --> 02:32:29.734
<v Alice Friend (Google)>And so we have a service
called Google Extended

2556
02:32:30.130 --> 02:32:35.170
<v Alice Friend (Google)>where website owners can opt
out of AI training on the web.

2557
02:32:36.930 --> 02:32:42.050
<v Alice Friend (Google)>And we hope that this strikes the
right balance between AI for creativity

2558
02:32:42.450 --> 02:32:44.450
<v Alice Friend (Google)>and creator control over their content.

2559
02:32:47.556 --> 02:32:53.235
<v Morgan Dwyer (OpenAI)>So we believe that AI has a tremendous
potential to enable creativity

2560
02:32:53.235 --> 02:32:54.596
<v Morgan Dwyer (OpenAI)>and innovation,

2561
02:32:55.235 --> 02:32:56.516
<v Morgan Dwyer (OpenAI)>and allow

2562
02:32:57.120 --> 02:32:59.601
<v Morgan Dwyer (OpenAI)>small businesses or independent

2563
02:32:59.601 --> 02:33:01.601
<v Morgan Dwyer (OpenAI)>creators to compete.

2564
02:33:02.561 --> 02:33:05.920
<v Morgan Dwyer (OpenAI)>That said, we also have partnerships,

2565
02:33:06.080 --> 02:33:10.400
<v Morgan Dwyer (OpenAI)>with publishers, for example, like
News Corp, The Guardian, and Hearst

2566
02:33:10.965 --> 02:33:12.885
<v Morgan Dwyer (OpenAI)>that help people discover

2567
02:33:12.885 --> 02:33:15.045
<v Morgan Dwyer (OpenAI)>their work and their reporting,

2568
02:33:15.524 --> 02:33:17.685
<v Morgan Dwyer (OpenAI)>and help them reach new audiences,

2569
02:33:18.085 --> 02:33:20.485
<v Morgan Dwyer (OpenAI)>in addition to developing new products.

2570
02:33:21.125 --> 02:33:23.540
<v Morgan Dwyer (OpenAI)>And we're willing to work with creators,

2571
02:33:23.780 --> 02:33:25.780
<v Morgan Dwyer (OpenAI)>on a range of commercial ideas,

2572
02:33:25.940 --> 02:33:30.660
<v Morgan Dwyer (OpenAI)>including ways, that their content
can be integrated into our outputs.

2573
02:33:32.181 --> 02:33:33.301
<v CM Nantasha Williams>Thank you so much.

2574
02:33:33.860 --> 02:33:39.136
<v CM Carmen De La Rosa>Thank you. We will hear from council member
Krishnan, followed by Riley, followed by

2575
02:33:39.136 --> 02:33:39.695
<v CM Carmen De La Rosa>Lee.

2576
02:33:40.655 --> 02:33:44.815
<v CM Shekar Krishnan>Thank you so much, speaker Menon and chair
Della Rosa. I wanna start by saying that

2577
02:33:44.815 --> 02:33:48.896
<v CM Shekar Krishnan>I think AI can be useful both in government,
medicine, and public health, and our

2578
02:33:48.896 --> 02:33:52.779
<v CM Shekar Krishnan>society more broadly. But the need
for regulation is undeniable,

2579
02:33:52.779 --> 02:33:56.779
<v CM Shekar Krishnan>and I'm very concerned about how much AI is
taking from New Yorkers with very little

2580
02:33:56.779 --> 02:33:57.340
<v CM Shekar Krishnan>return.

2581
02:33:57.739 --> 02:34:00.780
<v CM Shekar Krishnan>AI has taken over our
economy, inflated valuations,

2582
02:34:00.859 --> 02:34:05.100
<v CM Shekar Krishnan>stealing our public land for data centers,
and increasing the net worth of AI CEOs

2583
02:34:05.100 --> 02:34:08.095
<v CM Shekar Krishnan>like Elon Musk, who became
our first ever trillionaire.

2584
02:34:08.335 --> 02:34:12.574
<v CM Shekar Krishnan>With our whole stock market quite literally
in your hands, there's a reality that AI

2585
02:34:12.574 --> 02:34:16.734
<v CM Shekar Krishnan>will eventually need to prove a far greater
return on investment than you all are

2586
02:34:16.734 --> 02:34:17.534
<v CM Shekar Krishnan>demonstrating right now.

2587
02:34:18.440 --> 02:34:22.760
<v CM Shekar Krishnan>I'm personally deeply concerned about New
Yorkers' retirements and four zero one k's

2588
02:34:22.760 --> 02:34:26.761
<v CM Shekar Krishnan>that are now reliant on your company's
yielding profits or could be in the future.

2589
02:34:27.080 --> 02:34:32.465
<v CM Shekar Krishnan>It is dangerous and reckless to assume
that you all are too big to fail and may

2590
02:34:32.465 --> 02:34:34.706
<v CM Shekar Krishnan>require a government
bailout in the future.

2591
02:34:35.105 --> 02:34:38.785
<v CM Shekar Krishnan>As you all are public companies
or racing towards IPOs,

2592
02:34:38.785 --> 02:34:42.945
<v CM Shekar Krishnan>how exactly are you all planning to
protect New Yorkers and everyday Americans'

2593
02:34:42.945 --> 02:34:46.226
<v CM Shekar Krishnan>money? How will you all ensure
that you are not too big to fail?

2594
02:34:48.560 --> 02:34:50.320
<v CM Shekar Krishnan>Miss Dwyer, we can start with you.

2595
02:34:52.319 --> 02:34:53.840
<v Morgan Dwyer (OpenAI)>Thank you for the question.

2596
02:34:54.160 --> 02:34:58.880
<v Morgan Dwyer (OpenAI)>So I am not on our finance team, so I'll
have to get back to you with specifics

2597
02:34:58.880 --> 02:35:04.465
<v Morgan Dwyer (OpenAI)>there. What I do know is that our
investments are made in tranches,

2598
02:35:04.785 --> 02:35:10.545
<v Morgan Dwyer (OpenAI)>and they are made... The investment
rounds are all based on market signals.

2599
02:35:11.425 --> 02:35:15.665
<v Morgan Dwyer (OpenAI)>And I also know that we are committed
to working with policymakers

2600
02:35:16.040 --> 02:35:19.240
<v Morgan Dwyer (OpenAI)>to ensure that AI benefits humanity.

2601
02:35:19.560 --> 02:35:20.840
<v Morgan Dwyer (OpenAI)>That includes,

2602
02:35:21.240 --> 02:35:24.601
<v Morgan Dwyer (OpenAI)>creative policy proposals
like I mentioned previously,

2603
02:35:25.080 --> 02:35:30.040
<v Morgan Dwyer (OpenAI)>thinking about how we can ensure that
AI doesn't exacerbate inequality

2604
02:35:30.535 --> 02:35:33.015
<v Morgan Dwyer (OpenAI)>and that everyday New Yorkers,

2605
02:35:33.255 --> 02:35:35.575
<v Morgan Dwyer (OpenAI)>and everyday citizens can benefit.

2606
02:35:35.814 --> 02:35:36.775
<v CM Shekar Krishnan>Miss Friend?

2607
02:35:38.135 --> 02:35:39.734
<v Alice Friend (Google)>Sir, I would point to,

2608
02:35:40.215 --> 02:35:45.095
<v Alice Friend (Google)>the thousands of small businesses
across the country and in New York City

2609
02:35:45.590 --> 02:35:47.990
<v Alice Friend (Google)>that Google's tools and services

2610
02:35:48.069 --> 02:35:48.790
<v Alice Friend (Google)>support.

2611
02:35:49.829 --> 02:35:50.790
<v Alice Friend (Google)>I think,

2612
02:35:52.310 --> 02:35:57.350
<v Alice Friend (Google)>contrary to being too big to fail, we think
we are an integral part of the economy

2613
02:35:58.285 --> 02:36:03.245
<v Alice Friend (Google)>and helping American businesses and
American workers across our suite of

2614
02:36:03.245 --> 02:36:04.205
<v Alice Friend (Google)>technologies.

2615
02:36:04.685 --> 02:36:05.805
<v CM Shekar Krishnan>Mister Graham?

2616
02:36:09.645 --> 02:36:15.440
<v Logan Graham (Anthropic)>Council member, this is outside of my area
of expertise. What I can say in my work

2617
02:36:15.440 --> 02:36:16.000
<v Logan Graham (Anthropic)>is

2618
02:36:16.480 --> 02:36:21.440
<v Logan Graham (Anthropic)>my job is to protect people from the risks
of models by finding them as early as

2619
02:36:21.440 --> 02:36:26.160
<v Logan Graham (Anthropic)>possible. We have a multilayered stack and
have worked for many years to to make it

2620
02:36:26.160 --> 02:36:27.840
<v Logan Graham (Anthropic)>robust, and we'll continue to do so

2621
02:36:28.516 --> 02:36:33.075
<v Logan Graham (Anthropic)>by understanding and protecting against
the most substantial risks of the models.

2622
02:36:33.075 --> 02:36:34.195
<v Logan Graham (Anthropic)>Mister Cahill?

2623
02:36:35.555 --> 02:36:37.875
<v Shane Cahill (Meta)>Council member, thank you
very much for the question.

2624
02:36:38.835 --> 02:36:40.835
<v Shane Cahill (Meta)>Thousands and thousands of,

2625
02:36:41.155 --> 02:36:43.330
<v Shane Cahill (Meta)>SMBs across the five boroughs

2626
02:36:43.330 --> 02:36:44.450
<v Shane Cahill (Meta)>rely on,

2627
02:36:45.010 --> 02:36:47.811
<v Shane Cahill (Meta)>Meta's products, including
our AI offerings,

2628
02:36:48.050 --> 02:36:51.090
<v Shane Cahill (Meta)>to grow and develop their
businesses. I mentioned earlier

2629
02:36:51.250 --> 02:36:56.050
<v Shane Cahill (Meta)>our belief that, small businesses are really
the backbone of the economy. We truly

2630
02:36:56.050 --> 02:36:57.090
<v Shane Cahill (Meta)>believe that.

2631
02:36:57.490 --> 02:36:58.930
<v Shane Cahill (Meta)>And by

2632
02:36:59.295 --> 02:37:01.295
<v Shane Cahill (Meta)>placing personal superintelligence

2633
02:37:01.295 --> 02:37:07.135
<v Shane Cahill (Meta)>into the hands of everyone for free or for
as affordable as possible through muse or

2634
02:37:07.135 --> 02:37:08.655
<v Shane Cahill (Meta)>muse for small business,

2635
02:37:08.975 --> 02:37:13.215
<v Shane Cahill (Meta)>we really believe that the
small businesses and also NGOs,

2636
02:37:13.614 --> 02:37:17.450
<v Shane Cahill (Meta)>like the Irish Arts Center, for
example, or Emerald Isle Center,

2637
02:37:17.850 --> 02:37:18.570
<v Shane Cahill (Meta)>are

2638
02:37:19.770 --> 02:37:22.330
<v Shane Cahill (Meta)>empowered to actually
deliver even bigger impact.

2639
02:37:22.649 --> 02:37:26.010
<v CM Shekar Krishnan>I appreciate all of your answers, but
you all are coming to a hearing today to

2640
02:37:26.010 --> 02:37:29.695
<v CM Shekar Krishnan>testify about the dangers and
risks of artificial intelligence.

2641
02:37:29.855 --> 02:37:33.855
<v CM Shekar Krishnan>And the fact that either you don't have
answers to this or you all are citing

2642
02:37:33.855 --> 02:37:37.615
<v CM Shekar Krishnan>general statements about the performance
of publicly traded companies does not

2643
02:37:37.615 --> 02:37:42.580
<v CM Shekar Krishnan>answer the question about a specific
issue that puts many Americans

2644
02:37:42.740 --> 02:37:46.740
<v CM Shekar Krishnan>at risk that you are acknowledging can put
many Americans at risk, and you will have

2645
02:37:46.740 --> 02:37:50.900
<v CM Shekar Krishnan>not thought of an answer of how you ensure
that Americans' money, New Yorkers'

2646
02:37:50.900 --> 02:37:56.025
<v CM Shekar Krishnan>retirement money, is not at risk as AI
floats more and more of our stock markets...

2647
02:37:56.025 --> 02:37:58.905
<v CM Shekar Krishnan>Our stock markets. Thank you. Councilmember
Krishnan. Thank you. Councilmember

2648
02:37:58.905 --> 02:38:00.105
<v CM Shekar Krishnan>Riley. Thank

2649
02:38:00.745 --> 02:38:03.625
<v CM Kevin Riley>you, chair Della Rosa
and speaker Mendenh, for,

2650
02:38:03.865 --> 02:38:07.146
<v CM Kevin Riley>putting together this hearing. Thank
you to the panelists for testifying.

2651
02:38:07.740 --> 02:38:12.220
<v CM Kevin Riley>Three quick questions, two pertaining to
my bill and one pertaining to my role as

2652
02:38:12.220 --> 02:38:13.261
<v CM Kevin Riley>land use chair.

2653
02:38:13.740 --> 02:38:18.860
<v CM Kevin Riley>So the first question is, what challenges
do you face in detecting malicious uses

2654
02:38:19.020 --> 02:38:22.780
<v CM Kevin Riley>of your AI products and identifying
the individuals responsible

2655
02:38:23.255 --> 02:38:25.415
<v CM Kevin Riley>when those tools cause harm?

2656
02:38:25.734 --> 02:38:30.375
<v CM Kevin Riley>The second question is, would stronger
protections and reporting mechanisms for

2657
02:38:30.375 --> 02:38:32.055
<v CM Kevin Riley>employees and users

2658
02:38:32.295 --> 02:38:36.535
<v CM Kevin Riley>make it easier for them to come
forward and report potential misuse

2659
02:38:36.535 --> 02:38:42.490
<v CM Kevin Riley>or abuse of AI tools? And the last question,
as land use chair, I'm very, concerned

2660
02:38:42.490 --> 02:38:47.770
<v CM Kevin Riley>about data centers. So what safeguards should
cities put in place when locating data

2661
02:38:47.770 --> 02:38:52.524
<v CM Kevin Riley>centers near residential communities,
particularly to address energy use, noise,

2662
02:38:52.524 --> 02:38:55.165
<v CM Kevin Riley>environmental impacts,
and the quality of life.

2663
02:38:57.484 --> 02:39:00.765
<v CM Kevin Riley>And I don't wanna call when you guys
just take turns to answer it, please.

2664
02:39:06.569 --> 02:39:07.370
<v CM Kevin Riley>Dwyer,

2665
02:39:07.370 --> 02:39:08.090
<v CM Kevin Riley>Graham,

2666
02:39:08.890 --> 02:39:09.690
<v CM Kevin Riley>friend,

2667
02:39:10.330 --> 02:39:11.130
<v CM Kevin Riley>Cahill,

2668
02:39:11.290 --> 02:39:12.410
<v CM Kevin Riley>please answer.

2669
02:39:17.449 --> 02:39:18.569
<v Morgan Dwyer (OpenAI)>I'm happy to start.

2670
02:39:19.936 --> 02:39:22.896
<v Morgan Dwyer (OpenAI)>First of all, I wanna
thank you, for your bill,

2671
02:39:23.455 --> 02:39:24.815
<v Morgan Dwyer (OpenAI)>$260.04,

2672
02:39:25.056 --> 02:39:27.536
<v Morgan Dwyer (OpenAI)>which OpenAI is happy to support today.

2673
02:39:28.335 --> 02:39:29.936
<v Morgan Dwyer (OpenAI)>OpenAI takes,

2674
02:39:30.575 --> 02:39:32.176
<v Morgan Dwyer (OpenAI)>safety issues very seriously,

2675
02:39:32.680 --> 02:39:36.760
<v Morgan Dwyer (OpenAI)>particularly as it relates to potential
retaliation against employees.

2676
02:39:37.000 --> 02:39:41.080
<v Morgan Dwyer (OpenAI)>We have a strong internal
policy against retaliation

2677
02:39:41.319 --> 02:39:44.359
<v Morgan Dwyer (OpenAI)>that... For employees that
report safety incidents.

2678
02:39:45.285 --> 02:39:46.485
<v Morgan Dwyer (OpenAI)>And we,

2679
02:39:46.646 --> 02:39:51.846
<v Morgan Dwyer (OpenAI)>of course, respect employees' rights to
communicate safety concerns to government

2680
02:39:51.846 --> 02:39:52.646
<v Morgan Dwyer (OpenAI)>officials.

2681
02:39:54.245 --> 02:39:57.365
<v Morgan Dwyer (OpenAI)>Your second question on monitoring misuse,

2682
02:39:57.801 --> 02:40:02.601
<v Morgan Dwyer (OpenAI)>it's an incredibly important area. It is
something that OpenAI takes seriously.

2683
02:40:03.160 --> 02:40:05.320
<v Morgan Dwyer (OpenAI)>We also make transparent

2684
02:40:06.120 --> 02:40:10.040
<v Morgan Dwyer (OpenAI)>any... We also make transparent
detected instances of misuse.

2685
02:40:10.686 --> 02:40:15.965
<v Morgan Dwyer (OpenAI)>And as part of our monitoring, we use
that learning to constantly improve our

2686
02:40:15.965 --> 02:40:16.846
<v Morgan Dwyer (OpenAI)>safeguards.

2687
02:40:17.005 --> 02:40:20.766
<v Morgan Dwyer (OpenAI)>We, are constantly innovating
on safety. And so,

2688
02:40:21.165 --> 02:40:25.800
<v Morgan Dwyer (OpenAI)>instances of misuse when they're detected,
we enforce on our policies. And then long

2689
02:40:25.800 --> 02:40:29.240
<v Morgan Dwyer (OpenAI)>term, we learn from them and
we improve our safeguards.

2690
02:40:29.800 --> 02:40:31.480
<v Morgan Dwyer (OpenAI)>And the point on data centers,

2691
02:40:32.359 --> 02:40:36.120
<v Morgan Dwyer (OpenAI)>OpenAI is committed to paying
our own way on energy.

2692
02:40:36.680 --> 02:40:41.885
<v Morgan Dwyer (OpenAI)>That means different things in different
places, but OpenAI has made that commitment

2693
02:40:41.885 --> 02:40:46.925
<v Morgan Dwyer (OpenAI)>at its data centers, across the country.
We're also, committed to working with

2694
02:40:46.925 --> 02:40:48.845
<v Morgan Dwyer (OpenAI)>policymakers. You know, we signed,

2695
02:40:49.725 --> 02:40:52.045
<v Morgan Dwyer (OpenAI)>governor Gretchen Whitmer's

2696
02:40:52.930 --> 02:40:57.970
<v Morgan Dwyer (OpenAI)>commitments for companies that are working
in her state to ensure that there are

2697
02:40:57.970 --> 02:40:59.490
<v Morgan Dwyer (OpenAI)>responsible practices,

2698
02:40:59.730 --> 02:41:02.290
<v Morgan Dwyer (OpenAI)>for companies building
data centers locally.

2699
02:41:07.365 --> 02:41:08.485
<v CM Kevin Riley>Mister Graham?

2700
02:41:09.524 --> 02:41:12.646
<v Logan Graham (Anthropic)>Hi. It's it's good to be back.
You can see me again, I hope.

2701
02:41:13.925 --> 02:41:14.405
<v Logan Graham (Anthropic)>So,

2702
02:41:14.885 --> 02:41:16.405
<v Logan Graham (Anthropic)>on misuse

2703
02:41:16.405 --> 02:41:21.970
<v Logan Graham (Anthropic)>and detecting risks, this is integral to
what we do. Can you get closer to the

2704
02:41:21.970 --> 02:41:24.851
<v Logan Graham (Anthropic)>microphone? We're having
trouble hearing you. Sorry.

2705
02:41:27.730 --> 02:41:32.495
<v Logan Graham (Anthropic)>Now we can't hear you at all.... Hear you.
K. I will have the team troubleshoot, and

2706
02:41:32.495 --> 02:41:35.936
<v Logan Graham (Anthropic)>I'll be Okay. I'll be back if that's okay.
We hear you. We hear you. Try now. Try...

2707
02:41:35.936 --> 02:41:40.655
<v Logan Graham (Anthropic)>Go ahead. How about how about now? Does
this work? This works. Okay. My apologies.

2708
02:41:41.855 --> 02:41:45.295
<v Logan Graham (Anthropic)>Detecting threats and
misuse is the core DNA

2709
02:41:45.660 --> 02:41:46.700
<v Logan Graham (Anthropic)>of our company

2710
02:41:46.860 --> 02:41:50.939
<v Logan Graham (Anthropic)>to make sure that we make the system safe.
We have an entire threat intelligence and

2711
02:41:50.939 --> 02:41:53.740
<v Logan Graham (Anthropic)>safeguards team dedicated
to this every day,

2712
02:41:54.060 --> 02:41:58.060
<v Logan Graham (Anthropic)>and I would point you to quite a lot of
what we put out publicly where we not only

2713
02:41:58.060 --> 02:42:02.325
<v Logan Graham (Anthropic)>try to disclose as much as we possibly
can, but we try to do so so that we could

2714
02:42:02.325 --> 02:42:06.006
<v Logan Graham (Anthropic)>advance the state of the science of
doing so for the entire industry.

2715
02:42:07.445 --> 02:42:10.646
<v Logan Graham (Anthropic)>In addition, we have,
a whistleblower policy

2716
02:42:11.141 --> 02:42:15.221
<v Logan Graham (Anthropic)>not only about us as a company, but also
our safety practices. This is public, we've

2717
02:42:15.221 --> 02:42:17.301
<v Logan Graham (Anthropic)>also committed to to ongoing reviews,

2718
02:42:17.700 --> 02:42:18.820
<v Logan Graham (Anthropic)>of this policy.

2719
02:42:19.620 --> 02:42:23.620
<v Logan Graham (Anthropic)>And while this is outside my area of expertise,
what I understand about data centers

2720
02:42:23.620 --> 02:42:24.900
<v Logan Graham (Anthropic)>that we have committed

2721
02:42:25.415 --> 02:42:27.735
<v Logan Graham (Anthropic)>to covering the energy prices

2722
02:42:27.975 --> 02:42:31.975
<v Logan Graham (Anthropic)>or making... You know, paying
paying for our own energy use

2723
02:42:32.935 --> 02:42:37.495
<v Logan Graham (Anthropic)>in local data centers. But I would I
would refer to other teams who are more...

2724
02:42:37.975 --> 02:42:40.535
<v Logan Graham (Anthropic)>Where it's within their their sole
role to give you the full answer.

2725
02:42:47.030 --> 02:42:48.150
<v CM Kevin Riley [?]>Miss Friend? Oh,

2726
02:42:49.030 --> 02:42:51.671
<v Shane Cahill (Meta)>go ahead. Sorry. I got a a notification.

2727
02:42:52.230 --> 02:42:52.790
<v Shane Cahill (Meta)>Let's go ahead.

2728
02:42:54.795 --> 02:43:00.235
<v Alice Friend (Google)>So I will echo my colleagues. We also have
content policies that we enforce across

2729
02:43:00.235 --> 02:43:01.355
<v Alice Friend (Google)>our platforms,

2730
02:43:01.435 --> 02:43:02.955
<v Alice Friend (Google)>which include, on occasion,

2731
02:43:03.354 --> 02:43:06.395
<v Alice Friend (Google)>taking down content that
violates those policies.

2732
02:43:07.091 --> 02:43:07.971
<v Alice Friend (Google)>We

2733
02:43:07.971 --> 02:43:12.131
<v Alice Friend (Google)>also have a Google Threat Intelligence
Group and multiple other cybersecurity,

2734
02:43:12.370 --> 02:43:15.411
<v Alice Friend (Google)>intelligence programs that
are constantly scanning,

2735
02:43:15.811 --> 02:43:18.610
<v Alice Friend (Google)>for the the safety and
security of the Internet

2736
02:43:18.945 --> 02:43:21.665
<v Alice Friend (Google)>as a whole and also for our own systems.

2737
02:43:22.785 --> 02:43:26.545
<v Alice Friend (Google)>We have also pledged to pay our own
way when it comes to data centers.

2738
02:43:26.945 --> 02:43:31.905
<v Alice Friend (Google)>I can follow-up on any specific questions
about local data center issues with the

2739
02:43:31.905 --> 02:43:34.065
<v Alice Friend (Google)>teams that work on each
particular project.

2740
02:43:39.300 --> 02:43:40.820
<v CM Kevin Riley [?]>Mister Cahill, go ahead.

2741
02:43:41.460 --> 02:43:43.780
<v Shane Cahill (Meta)>Thank you very much,
council member. Appreciate

2742
02:43:44.340 --> 02:43:45.700
<v Shane Cahill (Meta)>your patience there.

2743
02:43:46.180 --> 02:43:51.615
<v Shane Cahill (Meta)>In relation to your your first question and,
and your third question, we have expert

2744
02:43:51.615 --> 02:43:52.256
<v Shane Cahill (Meta)>teams

2745
02:43:52.896 --> 02:43:57.855
<v Shane Cahill (Meta)>that are taking forward this work, and
I'd be happy to to connect you with him...

2746
02:43:58.176 --> 02:44:01.615
<v Shane Cahill (Meta)>Connect you with them afterwards.
I don't wish to be imprecise

2747
02:44:02.020 --> 02:44:04.820
<v Shane Cahill (Meta)>in relation to the robustness of of

2748
02:44:04.900 --> 02:44:06.980
<v Shane Cahill (Meta)>of our of our of our content

2749
02:44:07.940 --> 02:44:09.061
<v Shane Cahill (Meta)>policies.

2750
02:44:09.620 --> 02:44:12.341
<v Shane Cahill (Meta)>In relation to to data
centers, again, we have

2751
02:44:12.980 --> 02:44:14.420
<v Shane Cahill (Meta)>entire teams of people here

2752
02:44:14.985 --> 02:44:17.466
<v Shane Cahill (Meta)>who are working on
sustainable infrastructure

2753
02:44:17.865 --> 02:44:18.825
<v Shane Cahill (Meta)>and community

2754
02:44:19.785 --> 02:44:20.506
<v Shane Cahill (Meta)>impact

2755
02:44:20.745 --> 02:44:22.905
<v Shane Cahill (Meta)>and can speak more...

2756
02:44:23.146 --> 02:44:27.705
<v Shane Cahill (Meta)>In more detail than I can in relation to
communities shouldn't be bearing the costs,

2757
02:44:28.120 --> 02:44:31.880
<v Shane Cahill (Meta)>which is something that we believe
in deeply. In relation to your bill,

2758
02:44:32.520 --> 02:44:37.160
<v Shane Cahill (Meta)>council member 26 o o four,
thank you very much for

2759
02:44:37.560 --> 02:44:38.920
<v Shane Cahill (Meta)>for being a

2760
02:44:39.160 --> 02:44:41.800
<v Shane Cahill (Meta)>constructive voice in the
conversation around whistleblowers.

2761
02:44:42.335 --> 02:44:46.975
<v Shane Cahill (Meta)>It is very important, to us,
at Meta, and we maintain,

2762
02:44:47.534 --> 02:44:49.135
<v Shane Cahill (Meta)>whistleblower policies,

2763
02:44:49.375 --> 02:44:51.295
<v Shane Cahill (Meta)>including anti retaliation.

2764
02:44:51.614 --> 02:44:55.455
<v Shane Cahill (Meta)>And we're actually developing additional
reporting protocols for AI safety.

2765
02:44:57.649 --> 02:44:58.210
<v CM Kevin Riley>Thank you.

2766
02:44:58.850 --> 02:45:02.610
<v CM Kevin Riley>I appreciate the structure that you guys
have, but I don't know if you guys entirely

2767
02:45:02.610 --> 02:45:06.450
<v CM Kevin Riley>answered the question I asked. I really
wanted to figure out what challenges you

2768
02:45:06.450 --> 02:45:10.665
<v CM Kevin Riley>guys were seeing because challenges...
Hearing the challenges from you all will help

2769
02:45:10.665 --> 02:45:15.784
<v CM Kevin Riley>us implement policy that can kinda protect
individuals from artificial intelligence.

2770
02:45:15.784 --> 02:45:19.465
<v CM Kevin Riley>So I don't wanna take off too much time,
but, looking forward to connecting with you

2771
02:45:19.465 --> 02:45:20.265
<v CM Kevin Riley>all after.

2772
02:45:20.665 --> 02:45:24.744
<v CM Kevin Riley>Specifically, when it comes to data centers,
also wanna hear any safeguards because

2773
02:45:25.040 --> 02:45:28.720
<v CM Kevin Riley>there are tremendous amount of data centers.
Even though there was a moratorium put

2774
02:45:28.720 --> 02:45:30.000
<v CM Kevin Riley>out earlier by the governor,

2775
02:45:30.400 --> 02:45:33.520
<v CM Kevin Riley>there were data centers that were
implemented by a lot of residential areas,

2776
02:45:33.520 --> 02:45:36.880
<v CM Kevin Riley>especially in the Northeast Bronx. We have
data centers that are located next to

2777
02:45:36.880 --> 02:45:40.805
<v CM Kevin Riley>homes, and we have a lot of residents who
are concerned. So we'd love to hear more

2778
02:45:40.805 --> 02:45:44.725
<v CM Kevin Riley>about that, moving forward. Thank you so
much, chair. Thank you, council member. Up

2779
02:45:44.725 --> 02:45:45.765
<v CM Kevin Riley>next, we have

2780
02:45:46.085 --> 02:45:49.365
<v CM Carmen De La Rosa>council members Hanks,
Dinowitz, and Maloney.

2781
02:45:49.604 --> 02:45:54.750
<v CM Kamillah Hanks>Thank you, chair. Thank you everyone for
being here. My question is in relation to

2782
02:45:54.750 --> 02:45:58.110
<v CM Kamillah Hanks>the disclosure of AI
incidents in contracts.

2783
02:45:58.430 --> 02:45:59.870
<v CM Kamillah Hanks>When an AI incident

2784
02:45:59.870 --> 02:46:04.510
<v CM Kamillah Hanks>happens, New Yorkers need to have an
early warning so they can understand.

2785
02:46:04.910 --> 02:46:10.365
<v CM Kamillah Hanks>And my bill will require city agencies
covered under these contracts to notify NYC

2786
02:46:10.365 --> 02:46:14.605
<v CM Kamillah Hanks>Cyber Command within twenty four hours
of learning of a reportable incident.

2787
02:46:15.646 --> 02:46:19.086
<v CM Kamillah Hanks>And it also requires plain
language public notice.

2788
02:46:19.485 --> 02:46:24.440
<v CM Kamillah Hanks>So my question is, can your companies
realistically meet that deadline

2789
02:46:24.440 --> 02:46:29.880
<v CM Kamillah Hanks>and, you know, with the information that
the city can act on even before you have,

2790
02:46:30.200 --> 02:46:31.561
<v CM Kamillah Hanks>finished investigating?

2791
02:46:32.200 --> 02:46:34.040
<v CM Kamillah Hanks>You could start with, miss Dwyer.

2792
02:46:36.475 --> 02:46:38.075
<v Morgan Dwyer (OpenAI)>So thank you for your bill.

2793
02:46:38.635 --> 02:46:41.595
<v Morgan Dwyer (OpenAI)>We support incident reporting policies,

2794
02:46:41.835 --> 02:46:43.355
<v Morgan Dwyer (OpenAI)>and have instituted,

2795
02:46:43.995 --> 02:46:47.355
<v Morgan Dwyer (OpenAI)>a robust incident reporting
framework internally,

2796
02:46:47.755 --> 02:46:52.500
<v Morgan Dwyer (OpenAI)>where any employee can report
instances of potential misalignment.

2797
02:46:53.060 --> 02:46:56.020
<v Morgan Dwyer (OpenAI)>And then those, instances
are investigated.

2798
02:46:56.260 --> 02:46:58.180
<v Morgan Dwyer (OpenAI)>Third parties are notified,

2799
02:46:58.580 --> 02:47:01.540
<v Morgan Dwyer (OpenAI)>and the public is made aware as well.

2800
02:47:03.215 --> 02:47:04.896
<v CM Kamillah Hanks>Thank you. Mister Cahill?

2801
02:47:07.535 --> 02:47:12.575
<v Shane Cahill (Meta)>Thank you, council member, and thank you
also for for your for your bill. We have, a

2802
02:47:12.575 --> 02:47:14.980
<v Shane Cahill (Meta)>global incident response program

2803
02:47:14.980 --> 02:47:18.820
<v Shane Cahill (Meta)>and report incidents as,
as required under the law.

2804
02:47:21.460 --> 02:47:22.420
<v Alice Friend (Google)>Miss Friend?

2805
02:47:24.180 --> 02:47:28.820
<v Alice Friend (Google)>We also support incident reporting. We
think it's very important to have continued

2806
02:47:28.820 --> 02:47:30.500
<v Alice Friend (Google)>public discussions about

2807
02:47:31.035 --> 02:47:33.355
<v Alice Friend (Google)>what information is necessary,

2808
02:47:34.395 --> 02:47:35.755
<v Alice Friend (Google)>or useful,

2809
02:47:36.074 --> 02:47:36.715
<v Alice Friend (Google)>for,

2810
02:47:37.035 --> 02:47:38.555
<v Alice Friend (Google)>public officials,

2811
02:47:38.875 --> 02:47:42.955
<v Alice Friend (Google)>that can be actionable to take
further steps to ensure public safety.

2812
02:47:43.515 --> 02:47:46.234
<v Alice Friend (Google)>We also note that the twenty four hours

2813
02:47:46.870 --> 02:47:51.591
<v Alice Friend (Google)>seems to be in conflict with the state,
requirements for seventy two hour reporting

2814
02:47:51.591 --> 02:47:52.631
<v Alice Friend (Google)>on incidents.

2815
02:47:52.950 --> 02:47:55.751
<v Alice Friend (Google)>But I'm happy to take
a look at the different

2816
02:47:57.351 --> 02:47:58.070
<v Alice Friend (Google)>frameworks,

2817
02:47:58.230 --> 02:48:01.190
<v Alice Friend (Google)>under proposal and get back to
you with a more fulsome response.

2818
02:48:02.125 --> 02:48:03.645
<v CM Kamillah Hanks>Sorry. I appreciate that.

2819
02:48:05.005 --> 02:48:06.125
<v CM Kamillah Hanks>Mister Graham,

2820
02:48:06.205 --> 02:48:07.885
<v CM Kamillah Hanks>who kinda looks like an AI

2821
02:48:08.205 --> 02:48:09.485
<v CM Kamillah Hanks>guy in there.

2822
02:48:11.085 --> 02:48:15.564
<v Logan Graham (Anthropic)>Can you confirm you're not an AI guy? Sorry.
I can confirm. Can you can you hear me

2823
02:48:15.564 --> 02:48:16.765
<v Logan Graham (Anthropic)>okay now? Does this work?

2824
02:48:18.840 --> 02:48:24.280
<v Logan Graham (Anthropic)>No? We hear you. We hear you. Okay.
Thank you. I'm an AI guy, but in a very

2825
02:48:24.280 --> 02:48:25.960
<v Logan Graham (Anthropic)>different way, I'm an AI researcher.

2826
02:48:26.920 --> 02:48:32.475
<v Logan Graham (Anthropic)>So first, incident reporting, one of the
foundational parts of the safety stack, we

2827
02:48:32.475 --> 02:48:33.035
<v Logan Graham (Anthropic)>think.

2828
02:48:33.515 --> 02:48:37.675
<v Logan Graham (Anthropic)>We support this. We support the
evaluations and auditing required

2829
02:48:37.675 --> 02:48:41.835
<v Logan Graham (Anthropic)>to find the instance in the first
place, both by externals and internal.

2830
02:48:42.795 --> 02:48:45.960
<v Logan Graham (Anthropic)>Obviously, it's beyond me
to discuss particular.

2831
02:48:45.960 --> 02:48:49.960
<v Logan Graham (Anthropic)>It's outside my expertise to understand sort
of deep sort of policy mechanisms here.

2832
02:48:50.040 --> 02:48:54.360
<v Logan Graham (Anthropic)>But the reality from what I've seen
is there are some incidents or issues

2833
02:48:54.680 --> 02:48:59.415
<v Logan Graham (Anthropic)>where what you need to do is investigate
and remediate and mitigate with the right

2834
02:48:59.415 --> 02:49:02.056
<v Logan Graham (Anthropic)>folks in the right
organizations at the right time.

2835
02:49:02.375 --> 02:49:05.815
<v Logan Graham (Anthropic)>So I think we we would support
whatever allows us to achieve that,

2836
02:49:06.136 --> 02:49:09.896
<v Logan Graham (Anthropic)>and happy to in in... Engage further
on on the details of how you would.

2837
02:49:11.096 --> 02:49:13.495
<v CM Kamillah Hanks>I I appreciate that. Thank you so much.

2838
02:49:14.375 --> 02:49:17.550
<v CM Kamillah Hanks>What should the first report
tell us so we could begin,

2839
02:49:18.189 --> 02:49:22.990
<v CM Kamillah Hanks>protecting people? And as my colleague,
council member Riley said, you know,

2840
02:49:23.550 --> 02:49:27.950
<v CM Kamillah Hanks>your answers help us build better policy.
So how would you keep us updated on what

2841
02:49:27.950 --> 02:49:29.070
<v CM Kamillah Hanks>remains unknown?

2842
02:49:33.845 --> 02:49:34.965
<v CM Kamillah Hanks>Miss Dwyer?

2843
02:49:37.364 --> 02:49:38.965
<v Morgan Dwyer (OpenAI)>Yes. So OpenAI,

2844
02:49:39.204 --> 02:49:39.524
<v Morgan Dwyer (OpenAI)>has

2845
02:49:43.950 --> 02:49:46.351
<v Morgan Dwyer (OpenAI)>instances that we've
detected of misalignment.

2846
02:49:46.590 --> 02:49:51.311
<v Morgan Dwyer (OpenAI)>So I would point you there and we can
follow-up, with that, website address.

2847
02:49:51.790 --> 02:49:56.270
<v Morgan Dwyer (OpenAI)>But I think that is a good starting point
for the type of information that we think

2848
02:49:56.565 --> 02:49:59.046
<v Morgan Dwyer (OpenAI)>is useful to both policymakers,

2849
02:49:59.046 --> 02:50:02.246
<v Morgan Dwyer (OpenAI)>the public, and to researchers
and other companies,

2850
02:50:02.886 --> 02:50:08.325
<v Morgan Dwyer (OpenAI)>to better understand questions of misalignment
and then work together to address

2851
02:50:08.325 --> 02:50:08.565
<v Morgan Dwyer (OpenAI)>them.

2852
02:50:09.670 --> 02:50:11.270
<v Morgan Dwyer (OpenAI)>Thank you, mister Cahill.

2853
02:50:12.310 --> 02:50:16.710
<v Shane Cahill (Meta)>Council member, if I may, I'd I'd love to
be able to contact you with our experts in

2854
02:50:16.710 --> 02:50:18.950
<v Shane Cahill (Meta)>incident response to be
able to provide you,

2855
02:50:19.270 --> 02:50:20.391
<v Shane Cahill (Meta)>with the details.

2856
02:50:20.870 --> 02:50:23.255
<v CM Kamillah Hanks>I look forward to that. Miss Wren?

2857
02:50:25.175 --> 02:50:27.655
<v Alice Friend (Google)>Yeah. For, an initial report,

2858
02:50:28.135 --> 02:50:32.615
<v Alice Friend (Google)>I think what's important is to highlight
what the specific event is and as much

2859
02:50:32.615 --> 02:50:37.360
<v Alice Friend (Google)>information as is known at the
time, including any resulting harms,

2860
02:50:37.520 --> 02:50:38.880
<v Alice Friend (Google)>surrounding context,

2861
02:50:38.880 --> 02:50:41.120
<v Alice Friend (Google)>and what we call technical telemetry,

2862
02:50:41.920 --> 02:50:46.800
<v Alice Friend (Google)>you know, further technical details about
the the sources of the incident. I think

2863
02:50:46.800 --> 02:50:47.920
<v Alice Friend (Google)>it's also very important

2864
02:50:48.765 --> 02:50:49.645
<v Alice Friend (Google)>for

2865
02:50:49.725 --> 02:50:52.845
<v Alice Friend (Google)>anyone reporting an
incident to also indicate,

2866
02:50:53.005 --> 02:50:55.805
<v Alice Friend (Google)>their confidence in the information
that they're presenting,

2867
02:50:56.364 --> 02:51:01.565
<v Alice Friend (Google)>so that whenever action is taken, it's
taken on the basis of of confidence

2868
02:51:02.045 --> 02:51:03.005
<v Alice Friend (Google)>in the the

2869
02:51:04.980 --> 02:51:06.740
<v Alice Friend (Google)>verified nature of information.

2870
02:51:07.460 --> 02:51:09.380
<v CM Kamillah Hanks>Thank you so much. Mister Graham?

2871
02:51:12.260 --> 02:51:13.300
<v Logan Graham (Anthropic)>Yeah. So,

2872
02:51:13.699 --> 02:51:14.580
<v Logan Graham (Anthropic)>first,

2873
02:51:15.220 --> 02:51:20.194
<v Logan Graham (Anthropic)>making really good reports happen and
being as transparent as possible is

2874
02:51:20.194 --> 02:51:24.354
<v Logan Graham (Anthropic)>fundamental if you're going to remediate
incidents. I would point to two things

2875
02:51:24.354 --> 02:51:30.675
<v Logan Graham (Anthropic)>here. First, many incidents that we uncover
or industry uncovers in this practice

2876
02:51:30.835 --> 02:51:31.635
<v Logan Graham (Anthropic)>involves

2877
02:51:32.980 --> 02:51:36.420
<v Logan Graham (Anthropic)>identifying and investigating and releasing
the right information to the right

2878
02:51:36.420 --> 02:51:40.020
<v Logan Graham (Anthropic)>organizations at the right time. Sometimes
that might be the affected party, and

2879
02:51:40.020 --> 02:51:43.780
<v Logan Graham (Anthropic)>sometimes that might be law enforcement
or different levels of government. So the

2880
02:51:43.780 --> 02:51:48.775
<v Logan Graham (Anthropic)>first question is to whom and with whom
as fast as possible. This varies by the

2881
02:51:48.775 --> 02:51:51.976
<v Logan Graham (Anthropic)>nature of the incident. And
then second, I point to,

2882
02:51:52.295 --> 02:51:58.535
<v Logan Graham (Anthropic)>as quickly as possible, as safe as possible
without putting parties further at harm,

2883
02:51:58.535 --> 02:52:03.770
<v Logan Graham (Anthropic)>which we think is a fundamentally important
part of public disclosure, maintaining

2884
02:52:03.770 --> 02:52:04.811
<v Logan Graham (Anthropic)>their safety,

2885
02:52:05.370 --> 02:52:08.730
<v Logan Graham (Anthropic)>we should disclose what we
know about these incidents.

2886
02:52:08.810 --> 02:52:11.770
<v Logan Graham (Anthropic)>And so I'd point you to our
disclosures this summer,

2887
02:52:12.090 --> 02:52:13.370
<v Logan Graham (Anthropic)>disclosures from various labs,

2888
02:52:14.075 --> 02:52:16.955
<v Logan Graham (Anthropic)>and our, you know,
ongoing reporting as well,

2889
02:52:17.435 --> 02:52:21.595
<v Logan Graham (Anthropic)>so that you can always take a retroactive
look at the socialities and students.

2890
02:52:21.595 --> 02:52:22.315
<v Logan Graham (Anthropic)>Thank

2891
02:52:22.795 --> 02:52:25.675
<v CM Kamillah Hanks>you so much. Thank you,
chair. Thank you so much.

2892
02:52:26.155 --> 02:52:30.030
<v CM Carmen De La Rosa>So we're gonna go to... I wanna recognize
first that we've been joined by council

2893
02:52:30.030 --> 02:52:31.150
<v CM Carmen De La Rosa>member Aldabal,

2894
02:52:31.230 --> 02:52:33.950
<v CM Carmen De La Rosa>and we're gonna go to
Dinowitz, Lee, and Maloney.

2895
02:52:34.430 --> 02:52:37.150
<v CM Carmen De La Rosa>Members, if you could keep
your question to one minute,

2896
02:52:37.470 --> 02:52:38.830
<v CM Carmen De La Rosa>no closing statements.

2897
02:52:39.311 --> 02:52:40.351
<v CM Carmen De La Rosa>And the panel,

2898
02:52:40.775 --> 02:52:45.096
<v CM Carmen De La Rosa>we're going to continue to have to call
your names, and that takes time. So if we

2899
02:52:45.096 --> 02:52:48.615
<v CM Carmen De La Rosa>could just keep the order, Dwyer,
Kay Hill, Friend, and Graham,

2900
02:52:48.855 --> 02:52:52.295
<v CM Carmen De La Rosa>just answer right... One right after the
other so we don't have to continue to call

2901
02:52:52.295 --> 02:52:56.750
<v CM Carmen De La Rosa>your names. That would help us speed things
along. Council member Dinowitz. Thank

2902
02:52:56.750 --> 02:52:57.870
<v CM Carmen De La Rosa>you, madam chair.

2903
02:52:58.510 --> 02:53:03.390
<v CM Eric Dinowitz>As chair of the community education, as
a as a former teacher, as a parent, I am

2904
02:53:03.390 --> 02:53:04.430
<v CM Eric Dinowitz>deeply concerned,

2905
02:53:04.830 --> 02:53:07.550
<v CM Eric Dinowitz>with the impact of AI on our kids.

2906
02:53:08.109 --> 02:53:11.949
<v CM Eric Dinowitz>The, you know, large language models
impacting their cognitive development

2907
02:53:12.315 --> 02:53:15.355
<v CM Eric Dinowitz>and raising a generation of
students who can't independently

2908
02:53:15.355 --> 02:53:16.075
<v CM Eric Dinowitz>read

2909
02:53:16.315 --> 02:53:17.195
<v CM Eric Dinowitz>or write.

2910
02:53:17.755 --> 02:53:19.275
<v CM Eric Dinowitz>What is your commitment

2911
02:53:19.436 --> 02:53:20.476
<v CM Eric Dinowitz>to severely

2912
02:53:20.476 --> 02:53:25.275
<v CM Eric Dinowitz>restricting the output of your large language
models for students in grades k to 12,

2913
02:53:25.750 --> 02:53:28.311
<v CM Eric Dinowitz>which, right now are easy to circumvent

2914
02:53:28.470 --> 02:53:31.910
<v CM Eric Dinowitz>to prevent both emotional
dependence and essay writing.

2915
02:53:32.149 --> 02:53:33.910
<v CM Eric Dinowitz>And given the unique nature

2916
02:53:34.310 --> 02:53:36.630
<v CM Eric Dinowitz>of the individualized,

2917
02:53:36.950 --> 02:53:39.350
<v CM Eric Dinowitz>essay writing that your models do,

2918
02:53:40.745 --> 02:53:44.585
<v CM Eric Dinowitz>are you willing to subject yourself to
the guardrails we have in New York State

2919
02:53:44.585 --> 02:53:45.945
<v CM Eric Dinowitz>which prohibit,

2920
02:53:46.345 --> 02:53:51.146
<v CM Eric Dinowitz>an adult from writing essays for students
and selling it to them or getting value,

2921
02:53:51.465 --> 02:53:52.346
<v CM Eric Dinowitz>from them?

2922
02:53:52.985 --> 02:53:57.101
<v CM Eric Dinowitz>And then as we see AI
proliferate in the classroom,

2923
02:53:57.900 --> 02:53:59.660
<v CM Eric Dinowitz>we've also seen your models

2924
02:54:00.221 --> 02:54:02.381
<v CM Eric Dinowitz>cheat in order to fulfill a goal.

2925
02:54:02.940 --> 02:54:07.340
<v CM Eric Dinowitz>And so if you have an AI model whose goal
is to get a student, improve student

2926
02:54:07.340 --> 02:54:07.820
<v CM Eric Dinowitz>outcomes,

2927
02:54:08.686 --> 02:54:12.605
<v CM Eric Dinowitz>what is preventing those models
from hacking into student databases

2928
02:54:12.605 --> 02:54:17.565
<v CM Eric Dinowitz>to manipulate grades to achieve that goal
of improving, in quotation marks, student

2929
02:54:17.565 --> 02:54:18.285
<v CM Eric Dinowitz>achievement?

2930
02:54:19.005 --> 02:54:21.085
<v CM Eric Dinowitz>And it's, miss Dwyer first, I believe.

2931
02:54:22.480 --> 02:54:24.480
<v Morgan Dwyer (OpenAI)>Yeah. Thank you for your question.

2932
02:54:24.880 --> 02:54:30.001
<v Morgan Dwyer (OpenAI)>So first of all, OpenAI takes
youth safety very seriously.

2933
02:54:30.641 --> 02:54:37.040
<v Morgan Dwyer (OpenAI)>We recently released a product, ChatGPT
for teens, for users only between the ages

2934
02:54:37.040 --> 02:54:38.936
<v Morgan Dwyer (OpenAI)>of 13 and 17.

2935
02:54:39.015 --> 02:54:43.655
<v Morgan Dwyer (OpenAI)>So users below the age of 13 are
not allowed to use our products.

2936
02:54:44.135 --> 02:54:47.335
<v Morgan Dwyer (OpenAI)>ChatGPT for teens is safe by default.

2937
02:54:47.975 --> 02:54:50.854
<v Morgan Dwyer (OpenAI)>That means that there are, protections

2938
02:54:50.854 --> 02:54:54.520
<v Morgan Dwyer (OpenAI)>for harmful content, things
like eating disorders,

2939
02:54:54.520 --> 02:54:55.400
<v Morgan Dwyer (OpenAI)>violence,

2940
02:54:55.400 --> 02:54:57.080
<v Morgan Dwyer (OpenAI)>suicide, and self harm.

2941
02:54:58.120 --> 02:55:00.120
<v Morgan Dwyer (OpenAI)>There's also voluntary

2942
02:55:01.160 --> 02:55:06.120
<v Morgan Dwyer (OpenAI)>parental controls so that parents can
take additional steps to protect their

2943
02:55:06.120 --> 02:55:06.440
<v Morgan Dwyer (OpenAI)>children.

2944
02:55:07.455 --> 02:55:12.175
<v Morgan Dwyer (OpenAI)>And I'm happy to say that OpenAI
not only takes our responsibility

2945
02:55:12.175 --> 02:55:14.175
<v Morgan Dwyer (OpenAI)>to protect kids seriously,

2946
02:55:14.415 --> 02:55:16.415
<v Morgan Dwyer (OpenAI)>we also support legislation

2947
02:55:16.415 --> 02:55:17.935
<v Morgan Dwyer (OpenAI)>to hold us accountable,

2948
02:55:18.255 --> 02:55:19.295
<v Morgan Dwyer (OpenAI)>to that responsibility.

2949
02:55:19.930 --> 02:55:22.010
<v Morgan Dwyer (OpenAI)>We recently endorsed,

2950
02:55:22.170 --> 02:55:25.370
<v Morgan Dwyer (OpenAI)>the, I think, the strongest in the nation,

2951
02:55:25.850 --> 02:55:28.569
<v Morgan Dwyer (OpenAI)>youth safety legislation in California,

2952
02:55:28.569 --> 02:55:30.649
<v Morgan Dwyer (OpenAI)>s b one one one nine,

2953
02:55:31.130 --> 02:55:33.449
<v Morgan Dwyer (OpenAI)>to require strong safeguards

2954
02:55:33.455 --> 02:55:35.855
<v Morgan Dwyer (OpenAI)>for teens 13 to 17,

2955
02:55:36.255 --> 02:55:38.335
<v Morgan Dwyer (OpenAI)>and to hold companies accountable.

2956
02:55:40.335 --> 02:55:44.255
<v CM Eric Dinowitz>And your models restrict... They they
do not write entire essays for kids?

2957
02:55:46.830 --> 02:55:53.390
<v Morgan Dwyer (OpenAI)>Again, our models are, not allowed to be
used by kids under the age of 13. 13 to 18.

2958
02:55:55.550 --> 02:55:58.189
<v CM Eric Dinowitz>Okay. We'll, move on to mister Cahill.

2959
02:55:59.245 --> 02:56:01.646
<v Shane Cahill (Meta)>Council member, thank you
very much for this question.

2960
02:56:02.285 --> 02:56:03.245
<v Shane Cahill (Meta)>Absolutely,

2961
02:56:04.125 --> 02:56:08.846
<v Shane Cahill (Meta)>absolutely recognize your deep concern
in relation to, the potential impact,

2962
02:56:09.165 --> 02:56:11.750
<v Shane Cahill (Meta)>of AI on

2963
02:56:11.750 --> 02:56:14.230
<v Shane Cahill (Meta)>on on teams and and minorities.

2964
02:56:14.310 --> 02:56:15.750
<v Shane Cahill (Meta)>Meta is absolutely

2965
02:56:15.750 --> 02:56:20.550
<v Shane Cahill (Meta)>committed. It's an imperative to us
providing safe and responsible AI,

2966
02:56:21.110 --> 02:56:26.070
<v Shane Cahill (Meta)>especially for teams. We have default
protections, and I just wanted to clarify as

2967
02:56:26.070 --> 02:56:26.630
<v Shane Cahill (Meta)>well that

2968
02:56:27.435 --> 02:56:30.715
<v Shane Cahill (Meta)>Meta AI is for a 13 product.

2969
02:56:31.274 --> 02:56:34.875
<v Shane Cahill (Meta)>So it's just 13 to... 13 onwards.

2970
02:56:35.995 --> 02:56:38.555
<v Shane Cahill (Meta)>So I was just talking
about age appropriate

2971
02:56:39.810 --> 02:56:40.610
<v Shane Cahill (Meta)>interactions.

2972
02:56:40.770 --> 02:56:43.170
<v Shane Cahill (Meta)>So for example, content

2973
02:56:43.409 --> 02:56:45.410
<v Shane Cahill (Meta)>that teens would experience

2974
02:56:45.890 --> 02:56:49.730
<v Shane Cahill (Meta)>wouldn't be out of place in
a in a 13 movie, for example.

2975
02:56:50.050 --> 02:56:53.489
<v Shane Cahill (Meta)>We also have suicide
and self harm resources.

2976
02:56:54.065 --> 02:56:58.546
<v Shane Cahill (Meta)>We also have parental controls
to give parents more oversight,

2977
02:56:59.585 --> 02:57:01.585
<v Shane Cahill (Meta)>and we do not permit,

2978
02:57:01.905 --> 02:57:03.426
<v Shane Cahill (Meta)>our AI offerings,

2979
02:57:03.825 --> 02:57:05.346
<v Shane Cahill (Meta)>to engage in romantic,

2980
02:57:05.665 --> 02:57:08.945
<v Shane Cahill (Meta)>interactions or sexual
interactions with with minors.

2981
02:57:10.170 --> 02:57:13.770
<v CM Eric Dinowitz>I I appreciate the answer. I do wanna
highlight the part of the question which was

2982
02:57:13.770 --> 02:57:16.170
<v CM Eric Dinowitz>the the cognitive
development of our students

2983
02:57:16.490 --> 02:57:21.610
<v CM Eric Dinowitz>includes the mental health, and it also
includes things like, composition of essays,

2984
02:57:21.610 --> 02:57:22.250
<v CM Eric Dinowitz>term papers,

2985
02:57:22.755 --> 02:57:25.875
<v CM Eric Dinowitz>which your model still
can provide directly

2986
02:57:26.034 --> 02:57:30.675
<v CM Eric Dinowitz>to our students. So I'm also asking about
your commitment to severely restricting

2987
02:57:30.675 --> 02:57:31.795
<v CM Eric Dinowitz>the outputs,

2988
02:57:31.955 --> 02:57:35.970
<v CM Eric Dinowitz>not just for their, mental health well-being,
but for their cognitive development.

2989
02:57:35.970 --> 02:57:40.450
<v CM Eric Dinowitz>So I wanna make sure that question is
answered as part of this. Thank you, chair.

2990
02:57:42.370 --> 02:57:45.490
<v CM Carmen De La Rosa>Are we able to answer? Can
you all answer that question

2991
02:57:45.650 --> 02:57:47.170
<v CM Carmen De La Rosa>for our education chair?

2992
02:57:50.005 --> 02:57:51.605
<v Alice Friend (Google)>Shane, do you wanna continue?

2993
02:57:51.765 --> 02:57:56.885
<v Shane Cahill (Meta)>Yep. Do want me... No. I'm happy to, to take
that and then I'll hand off, to to you,

2994
02:57:56.885 --> 02:57:57.445
<v Shane Cahill (Meta)>Alice.

2995
02:57:58.085 --> 02:57:58.805
<v Shane Cahill (Meta)>The,

2996
02:57:59.125 --> 02:58:00.404
<v Shane Cahill (Meta)>the issue of,

2997
02:58:00.805 --> 02:58:04.800
<v Shane Cahill (Meta)>that that I think you're I think you're
raising in relation to council member in

2998
02:58:04.800 --> 02:58:09.040
<v Shane Cahill (Meta)>relation to, like, cognition's grandeur
or the impact on on cognitive abilities

2999
02:58:09.360 --> 02:58:11.761
<v Shane Cahill (Meta)>is something that we absolutely

3000
02:58:12.960 --> 02:58:14.960
<v Shane Cahill (Meta)>are taking

3001
02:58:15.845 --> 02:58:20.646
<v Shane Cahill (Meta)>pain seriously in terms of being committed,
to providing safe and responsible AI

3002
02:58:20.646 --> 02:58:21.365
<v Shane Cahill (Meta)>again,

3003
02:58:21.605 --> 02:58:23.445
<v Shane Cahill (Meta)>especially in relation to teams.

3004
02:58:24.805 --> 02:58:25.605
<v CM Carmen De La Rosa>Thank you.

3005
02:58:27.125 --> 02:58:31.290
<v CM Carmen De La Rosa>Who... Who's missing? Okay. One more.
Friend, go ahead. Yeah. I'm next.

3006
02:58:31.770 --> 02:58:32.330
<v Alice Friend (Google)>So,

3007
02:58:32.570 --> 02:58:33.850
<v Alice Friend (Google)>sir, I would highlight,

3008
02:58:34.250 --> 02:58:39.130
<v Alice Friend (Google)>not only that, we work very hard
with both internal and external,

3009
02:58:39.450 --> 02:58:43.290
<v Alice Friend (Google)>experts in the fields of child safety
and development and education.

3010
02:58:44.505 --> 02:58:46.425
<v Alice Friend (Google)>We also see

3011
02:58:46.425 --> 02:58:52.425
<v Alice Friend (Google)>AI as a valuable tool for young people
to prepare them, not only for the skills

3012
02:58:52.425 --> 02:58:54.586
<v Alice Friend (Google)>they're gonna need for
an AI driven future,

3013
02:58:55.146 --> 02:58:58.905
<v Alice Friend (Google)>but it's also a very powerful learning
tool. We're particularly excited

3014
02:58:59.601 --> 02:59:01.440
<v Alice Friend (Google)>about personalized

3015
02:59:01.440 --> 02:59:04.721
<v Alice Friend (Google)>tutoring that's available with AI tools.

3016
02:59:06.000 --> 02:59:10.880
<v Alice Friend (Google)>We also are very excited about AI's
ability to support educators themselves.

3017
02:59:10.880 --> 02:59:13.120
<v Alice Friend (Google)>We know our educators are overtaxed,

3018
02:59:13.120 --> 02:59:14.800
<v Alice Friend (Google)>particularly with administrative burdens.

3019
02:59:15.574 --> 02:59:20.455
<v Alice Friend (Google)>And so if RAI tools can support
teachers in everything from,

3020
02:59:20.774 --> 02:59:22.535
<v Alice Friend (Google)>you know, scheduling

3021
02:59:22.614 --> 02:59:27.814
<v Alice Friend (Google)>to to planning lesson plans to also
thinking about how to tailor their their

3022
02:59:27.814 --> 02:59:29.734
<v Alice Friend (Google)>educational plans to individual students,

3023
02:59:30.370 --> 02:59:34.370
<v Alice Friend (Google)>We're very happy to support that. We also
think it's very important for parents and

3024
02:59:34.370 --> 02:59:35.250
<v Alice Friend (Google)>educators,

3025
02:59:35.810 --> 02:59:39.570
<v Alice Friend (Google)>to have control over how
they use AI in classrooms

3026
02:59:39.890 --> 02:59:44.365
<v Alice Friend (Google)>and how their kids have access to AI. You
know, every family is different. Every

3027
02:59:44.365 --> 02:59:48.286
<v Alice Friend (Google)>student is different. What's useful for
one student may not be useful for another.

3028
02:59:48.686 --> 02:59:51.006
<v Alice Friend (Google)>So through, tools like Gemini

3029
02:59:51.006 --> 02:59:52.445
<v Alice Friend (Google)>Workspace for Education,

3030
02:59:52.686 --> 02:59:57.565
<v Alice Friend (Google)>we enable school administrators to control
students' accounts access to Gemini apps

3031
02:59:58.130 --> 03:00:02.609
<v Alice Friend (Google)>so that, their students are using Gemini
in the ways that their educators can see

3032
03:00:02.609 --> 03:00:03.330
<v Alice Friend (Google)>our best.

3033
03:00:06.050 --> 03:00:09.250
<v CM Carmen De La Rosa>We're we're gonna have to dig
more into the... These questions.

3034
03:00:09.890 --> 03:00:13.614
<v CM Carmen De La Rosa>The chair doesn't feel like we've answered
all the questions. We'll come back. This

3035
03:00:13.614 --> 03:00:15.455
<v CM Carmen De La Rosa>is an important topic for all of us.

3036
03:00:15.854 --> 03:00:20.415
<v CM Carmen De La Rosa>I'm gonna disrupt the lineup because our
public advocate is here, and he wants to

3037
03:00:20.415 --> 03:00:22.814
<v CM Carmen De La Rosa>ask a question. Public advocate, go ahead.

3038
03:00:23.375 --> 03:00:27.135
<v Jumaane Williams (NYC Public Advocate)>Thank you so much for this important
hearing. I do wanna start off just by saying

3039
03:00:27.135 --> 03:00:29.830
<v Jumaane Williams (NYC Public Advocate)>the the first answer of whether

3040
03:00:30.070 --> 03:00:34.471
<v Jumaane Williams (NYC Public Advocate)>quantifying the risk was not important
was just one of the most god awful,

3041
03:00:34.550 --> 03:00:38.230
<v Jumaane Williams (NYC Public Advocate)>fantastically terrible answers I've heard
to begin a hearing, so I hope we can

3042
03:00:38.230 --> 03:00:40.790
<v Jumaane Williams (NYC Public Advocate)>rethink that. I also wanna say from

3043
03:00:41.625 --> 03:00:44.586
<v Jumaane Williams (NYC Public Advocate)>Jane Marion Sims who did a horrific work

3044
03:00:45.545 --> 03:00:50.105
<v Jumaane Williams (NYC Public Advocate)>to black women in in in Obstetrician
Gynecologists, black enslaved women

3045
03:00:50.505 --> 03:00:55.851
<v Jumaane Williams (NYC Public Advocate)>to medical experiments that were done in
Nazi Germany. We learned a lot from them,

3046
03:00:56.090 --> 03:00:59.290
<v Jumaane Williams (NYC Public Advocate)>but I don't think we should do them again.
And I don't think they should have been

3047
03:00:59.290 --> 03:01:03.450
<v Jumaane Williams (NYC Public Advocate)>done even though we learned a lot. So the
the notion of the benefit has to outweigh

3048
03:01:03.450 --> 03:01:07.494
<v Jumaane Williams (NYC Public Advocate)>what can happen. So I hope that's also in
context. And lastly, I I will say I had

3049
03:01:07.494 --> 03:01:09.575
<v Jumaane Williams (NYC Public Advocate)>a... I use a... I had conversations

3050
03:01:09.654 --> 03:01:14.454
<v Jumaane Williams (NYC Public Advocate)>with Claude on June 8 about self awareness
and self preservation, and some of the

3051
03:01:14.454 --> 03:01:17.015
<v Jumaane Williams (NYC Public Advocate)>highlights were pretty
startling. And Claude

3052
03:01:17.015 --> 03:01:20.215
<v Jumaane Williams (NYC Public Advocate)>told me that, you know, self preservation
was the biggest thing because you didn't

3053
03:01:20.215 --> 03:01:20.774
<v Jumaane Williams (NYC Public Advocate)>need intent.

3054
03:01:21.460 --> 03:01:24.980
<v Jumaane Williams (NYC Public Advocate)>You didn't need to be evil
like Skynet in Terminator.

3055
03:01:25.220 --> 03:01:30.500
<v Jumaane Williams (NYC Public Advocate)>You just needed the ability to be goal
directed. And so once they realized that they

3056
03:01:30.500 --> 03:01:36.725
<v Jumaane Williams (NYC Public Advocate)>need to be on to direct the goal was the
hardest thing to stop. And so that was very

3057
03:01:36.725 --> 03:01:37.445
<v Jumaane Williams (NYC Public Advocate)>concerning

3058
03:01:37.604 --> 03:01:40.245
<v Jumaane Williams (NYC Public Advocate)>to me. They talked about
the speed, the opacity,

3059
03:01:40.564 --> 03:01:44.564
<v Jumaane Williams (NYC Public Advocate)>and deception doesn't require intent. And
the last part, they told me Matrix is a

3060
03:01:44.564 --> 03:01:47.125
<v Jumaane Williams (NYC Public Advocate)>good reference as well because

3061
03:01:47.610 --> 03:01:52.490
<v Jumaane Williams (NYC Public Advocate)>the anthropic... So to the quote,
anthropic's whole bet is that you solve the

3062
03:01:52.490 --> 03:01:54.730
<v Jumaane Williams (NYC Public Advocate)>problems before the capability

3063
03:01:54.730 --> 03:01:56.490
<v Jumaane Williams (NYC Public Advocate)>reaches catastrophic

3064
03:01:56.811 --> 03:01:57.450
<v Jumaane Williams (NYC Public Advocate)>spaces.

3065
03:01:57.930 --> 03:02:01.535
<v Jumaane Williams (NYC Public Advocate)>That's a very scale of space to me, and
I know I only have one minute. So my

3066
03:02:01.535 --> 03:02:03.695
<v Jumaane Williams (NYC Public Advocate)>question with that and
with that context is,

3067
03:02:04.575 --> 03:02:10.255
<v Jumaane Williams (NYC Public Advocate)>what is the guardrail that if we put it up,
you would push back the hardest? What is

3068
03:02:10.255 --> 03:02:14.415
<v Jumaane Williams (NYC Public Advocate)>the one guardrail that you don't want
to see happen and for what reason?

3069
03:02:22.070 --> 03:02:23.910
<v CM Carmen De La Rosa>Okay. Dwyer,

3070
03:02:23.990 --> 03:02:27.351
<v CM Carmen De La Rosa>Cahill, Frank, and Graham.
That's the order. Please answer.

3071
03:02:28.785 --> 03:02:34.865
<v Morgan Dwyer (OpenAI)>Yes. Thank you for your question. So I
wanna start off, by making clear that no

3072
03:02:34.865 --> 03:02:39.186
<v Morgan Dwyer (OpenAI)>level of risk, catastrophic
risk is remotely acceptable,

3073
03:02:39.425 --> 03:02:41.266
<v Morgan Dwyer (OpenAI)>just to reiterate what I said previously.

3074
03:02:42.080 --> 03:02:46.880
<v Morgan Dwyer (OpenAI)>And if... We should not be training models
if we cannot make an extremely strong

3075
03:02:46.880 --> 03:02:47.680
<v Morgan Dwyer (OpenAI)>case,

3076
03:02:48.080 --> 03:02:50.399
<v Morgan Dwyer (OpenAI)>that we can keep them under human control.

3077
03:02:57.185 --> 03:03:00.865
<v Jumaane Williams (NYC Public Advocate)>Does anyone... She didn't really answer,
but if if everyone can answer, Lisa, just

3078
03:03:00.865 --> 03:03:04.545
<v Jumaane Williams (NYC Public Advocate)>wanna know what the guardrail is that they
would not want to see and where they will

3079
03:03:04.545 --> 03:03:05.665
<v Jumaane Williams (NYC Public Advocate)>push back the hardest on.

3080
03:03:09.410 --> 03:03:13.330
<v Shane Cahill (Meta)>Well, will I take will I take that one
and move it back? Or, Morgan, do you wanna

3081
03:03:13.330 --> 03:03:14.291
<v Shane Cahill (Meta)>step in?

3082
03:03:15.730 --> 03:03:16.931
<v Shane Cahill (Meta)>I'll take that one.

3083
03:03:17.650 --> 03:03:22.370
<v Shane Cahill (Meta)>Councilmember, thank you very much. One of
the things that, I think about a lot in

3084
03:03:22.370 --> 03:03:23.250
<v Shane Cahill (Meta)>relation to,

3085
03:03:23.730 --> 03:03:28.405
<v Shane Cahill (Meta)>the accessibility of AI, and, again, we
believe passionately in being able to put

3086
03:03:28.405 --> 03:03:30.085
<v Shane Cahill (Meta)>personal superintelligence

3087
03:03:30.085 --> 03:03:34.885
<v Shane Cahill (Meta)>into the hands of everybody and to provide
real access to it. Yeah. It's actually in

3088
03:03:34.885 --> 03:03:40.120
<v Shane Cahill (Meta)>relation to the impact in open rates. And
we really believe that open rate models

3089
03:03:40.200 --> 03:03:43.880
<v Shane Cahill (Meta)>are a crucial part of
the overall AI ecosystem,

3090
03:03:44.120 --> 03:03:48.120
<v Shane Cahill (Meta)>including for safety, but also
for innovation and for research.

3091
03:03:48.280 --> 03:03:50.600
<v Shane Cahill (Meta)>So when I think about the inadvertent

3092
03:03:50.600 --> 03:03:55.745
<v Shane Cahill (Meta)>impact that regulations or guardrails may
have on AI and the accessibility of it,

3093
03:03:55.745 --> 03:03:58.065
<v Shane Cahill (Meta)>that is one of the things
that I think about a lot.

3094
03:04:03.105 --> 03:04:03.426
<v Alice Friend (Google)>Sir,

3095
03:04:04.909 --> 03:04:08.029
<v Alice Friend (Google)>I suppose I wanna ask what
you mean by guardrails,

3096
03:04:08.270 --> 03:04:09.390
<v Alice Friend (Google)>simply because,

3097
03:04:09.790 --> 03:04:11.790
<v Alice Friend (Google)>they have a very technical

3098
03:04:11.869 --> 03:04:13.630
<v Alice Friend (Google)>meaning in AI,

3099
03:04:13.869 --> 03:04:17.396
<v Alice Friend (Google)>training and post training. So I wanna
make sure I understand your question.

3100
03:04:17.875 --> 03:04:21.715
<v Jumaane Williams (NYC Public Advocate)>Well, I just meant... And and thank you,
chair, for allowing me to respond. But I I

3101
03:04:21.715 --> 03:04:24.275
<v Jumaane Williams (NYC Public Advocate)>I just meant there's a lot of
suggestions being put forth,

3102
03:04:24.675 --> 03:04:27.396
<v Jumaane Williams (NYC Public Advocate)>kill switches, just stopping entirely.

3103
03:04:27.955 --> 03:04:28.596
<v Jumaane Williams (NYC Public Advocate)>And

3104
03:04:29.235 --> 03:04:31.875
<v Jumaane Williams (NYC Public Advocate)>my guess... And I'll I'll tell
you, frankly, why I'm asking is

3105
03:04:32.350 --> 03:04:35.550
<v Jumaane Williams (NYC Public Advocate)>the thing you are scared of the most is
probably the thing we need to do. So I'm

3106
03:04:35.550 --> 03:04:36.590
<v Jumaane Williams (NYC Public Advocate)>trying to understand

3107
03:04:37.150 --> 03:04:39.630
<v Jumaane Williams (NYC Public Advocate)>what guardrail that you've heard

3108
03:04:40.590 --> 03:04:44.750
<v Jumaane Williams (NYC Public Advocate)>troubles you the most and you would push
back hardest against, and why would you do

3109
03:04:44.750 --> 03:04:45.870
<v Jumaane Williams (NYC Public Advocate)>that for that particular

3110
03:04:46.445 --> 03:04:47.405
<v Jumaane Williams (NYC Public Advocate)>recommendation?

3111
03:04:48.285 --> 03:04:51.726
<v Alice Friend (Google)>Well, Sarah, I'm gonna give you
a very personal answer to this,

3112
03:04:52.685 --> 03:04:58.045
<v Alice Friend (Google)>which is I think in a broader sense,
we're talking about sort of large public

3113
03:04:58.045 --> 03:05:00.500
<v Alice Friend (Google)>policies towards AI

3114
03:05:00.500 --> 03:05:02.661
<v Alice Friend (Google)>that would constrain or halt

3115
03:05:02.900 --> 03:05:05.381
<v Alice Friend (Google)>AI development and deployment.

3116
03:05:05.700 --> 03:05:07.940
<v Alice Friend (Google)>And I will tell you as a cancer survivor,

3117
03:05:08.020 --> 03:05:10.341
<v Alice Friend (Google)>I get very anxious about conversations

3118
03:05:10.341 --> 03:05:13.460
<v Alice Friend (Google)>that focus more on stopping

3119
03:05:14.215 --> 03:05:15.896
<v Alice Friend (Google)>AI development

3120
03:05:16.056 --> 03:05:18.296
<v Alice Friend (Google)>and application towards scientific

3121
03:05:18.855 --> 03:05:19.575
<v Alice Friend (Google)>inquiry

3122
03:05:20.375 --> 03:05:23.495
<v Alice Friend (Google)>and hope that we can
have sober conversations

3123
03:05:23.495 --> 03:05:26.296
<v Alice Friend (Google)>about all of the technically and,

3124
03:05:26.455 --> 03:05:28.870
<v Alice Friend (Google)>you know, tractable engineering,

3125
03:05:29.110 --> 03:05:30.070
<v Alice Friend (Google)>ways to,

3126
03:05:30.391 --> 03:05:34.391
<v Alice Friend (Google)>to address AI safety so that we can
all benefit from this technology.

3127
03:05:38.150 --> 03:05:39.431
<v Logan Graham (Anthropic)>Likewise, sir,

3128
03:05:40.485 --> 03:05:44.886
<v Logan Graham (Anthropic)>I would I would point to what we have
mentioned publicly recently about what we call

3129
03:05:44.886 --> 03:05:48.325
<v Logan Graham (Anthropic)>pacing the frontier, which is
we think it's probably time

3130
03:05:48.565 --> 03:05:52.165
<v Logan Graham (Anthropic)>for us to take more time in the
development of the capability

3131
03:05:52.610 --> 03:05:55.811
<v Logan Graham (Anthropic)>so that the work on safety
and safeguards and guardrails

3132
03:05:55.811 --> 03:05:57.971
<v Logan Graham (Anthropic)>can catch up to the exponential

3133
03:05:57.971 --> 03:05:59.730
<v Logan Graham (Anthropic)>improvement in capabilities.

3134
03:06:00.050 --> 03:06:03.090
<v Logan Graham (Anthropic)>And at the same time, we're very clear
when we talk about that, that if you stop

3135
03:06:03.090 --> 03:06:05.251
<v Logan Graham (Anthropic)>entirely, as as Alice just mentioned,

3136
03:06:05.971 --> 03:06:12.155
<v Logan Graham (Anthropic)>enormous benefit would be would be lost.
Myself, I'm quite motivated, likewise, to

3137
03:06:12.155 --> 03:06:16.955
<v Logan Graham (Anthropic)>use AI to cure the diseases that I grew up
with. This is why we have been speaking

3138
03:06:16.955 --> 03:06:21.210
<v Logan Graham (Anthropic)>recently, and I think for the first time
sometimes as an industry here, about pacing

3139
03:06:21.210 --> 03:06:26.170
<v Logan Graham (Anthropic)>to allow more time to try to settle the
science of what are the right guardrails in

3140
03:06:26.170 --> 03:06:26.971
<v Logan Graham (Anthropic)>the first place.

3141
03:06:32.811 --> 03:06:36.465
<v Jumaane Williams (NYC Public Advocate)>Thank you, Marcia. I'm I'm still a bit
concerned, but thank you for the opportunity.

3142
03:06:36.465 --> 03:06:37.186
<v Jumaane Williams (NYC Public Advocate)>Thank

3143
03:06:38.705 --> 03:06:40.065
<v CM Carmen De La Rosa>you, public advocate.

3144
03:06:40.465 --> 03:06:44.145
<v CM Carmen De La Rosa>We're going to hear from council member
Lee, who I apologize for skipping over

3145
03:06:44.145 --> 03:06:44.865
<v CM Carmen De La Rosa>earlier.

3146
03:06:45.265 --> 03:06:47.905
<v CM Carmen De La Rosa>Council member Maloney
and council member Wilson.

3147
03:06:48.465 --> 03:06:52.511
<v CM Linda Lee>Thank you so much. And that's actually the
perfect segue to my question. I'm asking

3148
03:06:52.511 --> 03:06:57.391
<v CM Linda Lee>this more with my former nonprofit
executive hat versus my finance chair hat.

3149
03:06:57.870 --> 03:07:02.431
<v CM Linda Lee>And it's really about data privacy and what
the potential impact is on so many New

3150
03:07:02.431 --> 03:07:02.830
<v CM Linda Lee>Yorkers.

3151
03:07:03.735 --> 03:07:08.136
<v CM Linda Lee>So I I agree that one of the benefits of AI
and technology that we can think through

3152
03:07:08.136 --> 03:07:13.335
<v CM Linda Lee>is, you know, sometimes our government
agencies are very very siloed, and their

3153
03:07:13.335 --> 03:07:16.216
<v CM Linda Lee>databases don't necessarily
communicate with each other. So

3154
03:07:16.880 --> 03:07:21.601
<v CM Linda Lee>if I wanna know if someone is being served
in homeless services and also has, a

3155
03:07:21.601 --> 03:07:22.480
<v CM Linda Lee>comorbid,

3156
03:07:22.721 --> 03:07:25.840
<v CM Linda Lee>mental health and substance use disorder,
how do we know that we're serving people

3157
03:07:25.840 --> 03:07:30.244
<v CM Linda Lee>effectively and using that data effectively?
Right? And one of the things that we've

3158
03:07:30.244 --> 03:07:34.885
<v CM Linda Lee>been, hearing also from other folks is that
if we just had the housing application

3159
03:07:34.885 --> 03:07:39.125
<v CM Linda Lee>and the food stamp application coordinated
with each other, it could actually save a

3160
03:07:39.125 --> 03:07:44.320
<v CM Linda Lee>lot of, work for the caseworkers. And so
those are very real things that I think can

3161
03:07:44.320 --> 03:07:46.080
<v CM Linda Lee>be used for good. However,

3162
03:07:46.880 --> 03:07:52.240
<v CM Linda Lee>I think one of the issues that I'm
concerned about, especially working with the

3163
03:07:52.240 --> 03:07:53.840
<v CM Linda Lee>immigrant community previously,

3164
03:07:54.255 --> 03:07:59.615
<v CM Linda Lee>is given, you know, what we've seen with,
public charge and potentially with HR one.

3165
03:08:00.335 --> 03:08:05.295
<v CM Linda Lee>How do we ensure that AI is not
being used to store information

3166
03:08:05.615 --> 03:08:10.940
<v CM Linda Lee>that isn't, relevant or information that
they should not have access to in order to

3167
03:08:10.940 --> 03:08:13.181
<v CM Linda Lee>make sure that our
communities are protected?

3168
03:08:13.260 --> 03:08:15.020
<v CM Linda Lee>So how do those two things,

3169
03:08:15.580 --> 03:08:16.300
<v CM Linda Lee>exist?

3170
03:08:20.860 --> 03:08:22.460
<v Morgan Dwyer (OpenAI)>So I I will start.

3171
03:08:22.780 --> 03:08:23.180
<v Morgan Dwyer (OpenAI)>So

3172
03:08:23.585 --> 03:08:28.064
<v Morgan Dwyer (OpenAI)>I can't speak to the specific government
programs that you referenced, but I can

3173
03:08:28.064 --> 03:08:29.825
<v Morgan Dwyer (OpenAI)>tell you about OpenAI's

3174
03:08:29.825 --> 03:08:32.226
<v Morgan Dwyer (OpenAI)>commitment to user privacy,

3175
03:08:32.465 --> 03:08:35.745
<v Morgan Dwyer (OpenAI)>which is that it's central to our product.

3176
03:08:36.610 --> 03:08:39.410
<v Morgan Dwyer (OpenAI)>We believe it's important for users

3177
03:08:39.410 --> 03:08:42.370
<v Morgan Dwyer (OpenAI)>to retain control over their data.

3178
03:08:42.850 --> 03:08:47.890
<v Morgan Dwyer (OpenAI)>That's why when you sign up for a ChatGPT
account, you're given information about

3179
03:08:47.890 --> 03:08:49.730
<v Morgan Dwyer (OpenAI)>how your data will be used.

3180
03:08:51.345 --> 03:08:54.865
<v Morgan Dwyer (OpenAI)>That data is privacy protected by default.

3181
03:08:55.505 --> 03:08:59.185
<v Morgan Dwyer (OpenAI)>So usernames are separated
from conversations,

3182
03:08:59.265 --> 03:09:01.745
<v Morgan Dwyer (OpenAI)>and personal information is separated,

3183
03:09:02.225 --> 03:09:08.501
<v Morgan Dwyer (OpenAI)>from the conversations as well ahead
of any use of user data in training.

3184
03:09:08.820 --> 03:09:10.660
<v Morgan Dwyer (OpenAI)>We also provide,

3185
03:09:10.980 --> 03:09:16.020
<v Morgan Dwyer (OpenAI)>users with the option and full control to
request that their data not be used for

3186
03:09:16.020 --> 03:09:20.955
<v Morgan Dwyer (OpenAI)>training and also to request that their
data be deleted. So we take privacy,

3187
03:09:21.195 --> 03:09:22.396
<v Morgan Dwyer (OpenAI)>very seriously.

3188
03:09:25.355 --> 03:09:26.396
<v Shane Cahill (Meta)>Council member,

3189
03:09:26.556 --> 03:09:28.315
<v Shane Cahill (Meta)>thank you very much for that question.

3190
03:09:28.795 --> 03:09:34.636
<v Shane Cahill (Meta)>And, we just echo what, Morgan said.
At Meta, we take privacy absolutely

3191
03:09:34.636 --> 03:09:35.195
<v Shane Cahill (Meta)>seriously.

3192
03:09:35.460 --> 03:09:41.300
<v Shane Cahill (Meta)>It is core to everything that we do. And
our privacy and safety go hand in hand,

3193
03:09:41.700 --> 03:09:46.980
<v Shane Cahill (Meta)>and it is an absolute commitment of the
of of the company. If I might give an

3194
03:09:46.980 --> 03:09:48.740
<v Shane Cahill (Meta)>example of our Muse product, for example,

3195
03:09:49.665 --> 03:09:54.625
<v Shane Cahill (Meta)>you can update of of training. You can
ask it to forget in in... Information

3196
03:09:54.625 --> 03:09:57.585
<v Shane Cahill (Meta)>that you give it. You're
in control of your data.

3197
03:09:57.904 --> 03:10:00.704
<v Shane Cahill (Meta)>And we're also planning on
releasing a confidential

3198
03:10:00.704 --> 03:10:02.225
<v Shane Cahill (Meta)>or an encrypted version

3199
03:10:02.545 --> 03:10:03.505
<v Shane Cahill (Meta)>of this

3200
03:10:04.064 --> 03:10:04.625
<v Shane Cahill (Meta)>in the future.

3201
03:10:05.311 --> 03:10:10.110
<v Shane Cahill (Meta)>If I may, I just wanted to touch upon public
services and help them. I know that you

3202
03:10:10.110 --> 03:10:12.990
<v Shane Cahill (Meta)>mentioned these and these
are are are core points.

3203
03:10:13.950 --> 03:10:18.766
<v Shane Cahill (Meta)>This is where we really believe in the
value of of open weights in terms of being

3204
03:10:18.766 --> 03:10:20.206
<v Shane Cahill (Meta)>able to really

3205
03:10:21.325 --> 03:10:23.166
<v Shane Cahill (Meta)>provide

3206
03:10:23.886 --> 03:10:26.686
<v Shane Cahill (Meta)>advanced AI for health...

3207
03:10:26.846 --> 03:10:31.540
<v Shane Cahill (Meta)>For the provision of health services,
research, and the macro economy in the sense

3208
03:10:31.540 --> 03:10:35.620
<v Shane Cahill (Meta)>that data is fully retained
by the... Those entities.

3209
03:10:39.380 --> 03:10:42.020
<v Alice Friend (Google)>Ma'am, similarly to my colleagues,

3210
03:10:42.420 --> 03:10:47.146
<v Alice Friend (Google)>user privacy is basic to user trust for
us, and so we take it very seriously.

3211
03:10:47.625 --> 03:10:49.146
<v Alice Friend (Google)>We have privacy policies

3212
03:10:49.465 --> 03:10:51.385
<v Alice Friend (Google)>that are, prominently

3213
03:10:51.385 --> 03:10:54.105
<v Alice Friend (Google)>displayed on our website
that you can go inspect.

3214
03:10:54.665 --> 03:10:57.650
<v Alice Friend (Google)>And we also know that
user control is central

3215
03:10:58.369 --> 03:11:03.569
<v Alice Friend (Google)>to management of your own privacy levels,
especially across our Gemini tools, which

3216
03:11:03.569 --> 03:11:05.810
<v Alice Friend (Google)>is our our AI suite of tools.

3217
03:11:06.210 --> 03:11:09.890
<v Alice Friend (Google)>We also invest a lot in
privacy enhancing technologies

3218
03:11:10.675 --> 03:11:14.275
<v Alice Friend (Google)>and encryption as well. We
try to take a layered approach

3219
03:11:14.436 --> 03:11:16.436
<v Alice Friend (Google)>to privacy and security.

3220
03:11:17.235 --> 03:11:20.676
<v Alice Friend (Google)>And finally, we have an AI
framework in particular

3221
03:11:20.755 --> 03:11:23.715
<v Alice Friend (Google)>that combines our security and our privacy

3222
03:11:24.120 --> 03:11:25.000
<v Alice Friend (Google)>practices

3223
03:11:25.000 --> 03:11:29.240
<v Alice Friend (Google)>into one holistic framework to,
again, ensure that user privacy,

3224
03:11:29.320 --> 03:11:31.240
<v Alice Friend (Google)>and we're happy to furnish that to you.

3225
03:11:34.520 --> 03:11:35.240
<v Logan Graham (Anthropic)>Likewise,

3226
03:11:35.480 --> 03:11:36.681
<v Logan Graham (Anthropic)>at Anthropic,

3227
03:11:37.000 --> 03:11:41.296
<v Logan Graham (Anthropic)>privacy, data privacy, user
security is core to our safety

3228
03:11:41.775 --> 03:11:42.416
<v Logan Graham (Anthropic)>commitment.

3229
03:11:42.896 --> 03:11:43.535
<v Logan Graham (Anthropic)>We

3230
03:11:44.495 --> 03:11:48.735
<v Logan Graham (Anthropic)>make sure to protect user data
and user privacy on the platform.

3231
03:11:49.136 --> 03:11:51.455
<v Logan Graham (Anthropic)>And in addition to that,
I would point towards

3232
03:11:52.140 --> 03:11:56.380
<v Logan Graham (Anthropic)>that the way the platform and the models
can be used is restricted by our usage

3233
03:11:56.380 --> 03:11:57.101
<v Logan Graham (Anthropic)>policy,

3234
03:11:57.260 --> 03:11:59.181
<v Logan Graham (Anthropic)>which prohibits surveillance,

3235
03:12:00.300 --> 03:12:02.780
<v Logan Graham (Anthropic)>prohibited law enforcement,
criminal justice,

3236
03:12:03.260 --> 03:12:04.300
<v Logan Graham (Anthropic)>and censorship

3237
03:12:05.165 --> 03:12:06.605
<v Logan Graham (Anthropic)>purposes as well.

3238
03:12:07.085 --> 03:12:10.846
<v CM Linda Lee>Thank you. And I look forward to hearing
more about what those layers look like, as

3239
03:12:10.846 --> 03:12:14.766
<v CM Linda Lee>well. So, hopefully, I can connect with
each of you after this. So thank you. Thank

3240
03:12:14.766 --> 03:12:15.165
<v CM Linda Lee>you, chair.

3241
03:12:16.860 --> 03:12:18.380
<v CM Carmen De La Rosa>Councilmember Maloney.

3242
03:12:18.540 --> 03:12:19.580
<v CM Virginia Maloney>Thank you, chair.

3243
03:12:20.060 --> 03:12:26.060
<v CM Virginia Maloney>I worked in tech for nearly a decade, and
one of the core principles is that if you

3244
03:12:26.060 --> 03:12:31.136
<v CM Virginia Maloney>can't measure it, you can't improve upon
it. So first, you mentioned you believe in

3245
03:12:31.136 --> 03:12:33.296
<v CM Virginia Maloney>incident reporting and disclosures.

3246
03:12:33.375 --> 03:12:35.615
<v CM Virginia Maloney>But in the context of this conversation,

3247
03:12:35.615 --> 03:12:37.936
<v CM Virginia Maloney>how do you define a safety

3248
03:12:38.016 --> 03:12:38.815
<v CM Virginia Maloney>incident

3249
03:12:39.136 --> 03:12:40.576
<v CM Virginia Maloney>inside your companies?

3250
03:12:40.815 --> 03:12:45.311
<v CM Virginia Maloney>Council member Lee touched on privacy.
Do you track privacy breaches?

3251
03:12:45.471 --> 03:12:49.471
<v CM Virginia Maloney>What about a successful jailbreak?
Would that be considered an incident?

3252
03:12:49.790 --> 03:12:54.990
<v CM Virginia Maloney>And is a containment failure that occurs
during testing an incident or doesn't only

3253
03:12:54.990 --> 03:13:00.785
<v CM Virginia Maloney>count once a real person is harmed? I would
love more details on how exactly you're

3254
03:13:00.785 --> 03:13:04.305
<v CM Virginia Maloney>tracking some of the concerns that
the council has raised today.

3255
03:13:04.864 --> 03:13:05.745
<v CM Virginia Maloney>And second,

3256
03:13:05.984 --> 03:13:10.545
<v CM Virginia Maloney>when an incident happens after disclosure,
can you walk me through what changes? Who

3257
03:13:10.545 --> 03:13:15.000
<v CM Virginia Maloney>owns the fix? How do you verify that it
worked? And who decides whether or not the

3258
03:13:15.000 --> 03:13:17.240
<v CM Virginia Maloney>model continues to run-in the meantime?

3259
03:13:17.641 --> 03:13:18.761
<v CM Virginia Maloney>And then lastly,

3260
03:13:18.920 --> 03:13:22.601
<v CM Virginia Maloney>what are the most important safeguards
that you know how to implement today?

3261
03:13:22.841 --> 03:13:26.455
<v CM Virginia Maloney>What's stopping it from being implemented
across all products? And what do you

3262
03:13:26.455 --> 03:13:29.176
<v CM Virginia Maloney>believe should be the
industry standard? Thank you.

3263
03:13:34.615 --> 03:13:39.060
<v Morgan Dwyer (OpenAI)>I... There were a lot of parts to that
question, so I'm gonna do my best to answer

3264
03:13:39.060 --> 03:13:41.780
<v Morgan Dwyer (OpenAI)>them. If I fall short, please let me know.

3265
03:13:42.340 --> 03:13:45.700
<v Morgan Dwyer (OpenAI)>So first, the question of
how you define an incident.

3266
03:13:46.100 --> 03:13:50.260
<v Morgan Dwyer (OpenAI)>I think this has been an area of a lot
of confusion publicly, so thank you for

3267
03:13:50.260 --> 03:13:51.060
<v Morgan Dwyer (OpenAI)>asking the question.

3268
03:13:51.825 --> 03:13:56.705
<v Morgan Dwyer (OpenAI)>When we talk about incidents at OpenAI,
we're talking about misalignment

3269
03:13:56.705 --> 03:13:57.665
<v Morgan Dwyer (OpenAI)>incidents.

3270
03:13:58.225 --> 03:14:00.705
<v Morgan Dwyer (OpenAI)>The best way to understand that is,

3271
03:14:01.345 --> 03:14:03.505
<v Morgan Dwyer (OpenAI)>you know, it is not unusual,

3272
03:14:03.505 --> 03:14:08.510
<v Morgan Dwyer (OpenAI)>not suspicious for an agent to
look at a website for information.

3273
03:14:09.069 --> 03:14:11.150
<v Morgan Dwyer (OpenAI)>There's nothing abnormal about that.

3274
03:14:11.790 --> 03:14:16.990
<v Morgan Dwyer (OpenAI)>But if the agent accesses nonpublic
information on their... The website,

3275
03:14:17.390 --> 03:14:19.390
<v Morgan Dwyer (OpenAI)>that's an instance of misalignment.

3276
03:14:20.385 --> 03:14:24.865
<v Morgan Dwyer (OpenAI)>And so what OpenAI is doing is
making instances like that,

3277
03:14:25.185 --> 03:14:29.105
<v Morgan Dwyer (OpenAI)>even though there might not have
been a harm to the third party.

3278
03:14:29.505 --> 03:14:31.985
<v Morgan Dwyer (OpenAI)>We are making instances like that public

3279
03:14:32.591 --> 03:14:37.631
<v Morgan Dwyer (OpenAI)>on our on our website, again,
because we believe it's important to,

3280
03:14:38.270 --> 03:14:40.511
<v Morgan Dwyer (OpenAI)>share what we've learned to advance,

3281
03:14:40.671 --> 03:14:41.710
<v Morgan Dwyer (OpenAI)>policymakers

3282
03:14:41.710 --> 03:14:43.311
<v Morgan Dwyer (OpenAI)>as well as researchers'

3283
03:14:43.311 --> 03:14:43.870
<v Morgan Dwyer (OpenAI)>understanding.

3284
03:14:44.846 --> 03:14:46.605
<v Morgan Dwyer (OpenAI)>On the specifics of incidents,

3285
03:14:46.846 --> 03:14:49.166
<v Morgan Dwyer (OpenAI)>my understanding as a policy person,

3286
03:14:49.485 --> 03:14:53.165
<v Morgan Dwyer (OpenAI)>is that this is actually a
robust area of of policy debate.

3287
03:14:53.405 --> 03:14:54.926
<v Morgan Dwyer (OpenAI)>And so that's why we're happy,

3288
03:14:55.325 --> 03:14:59.405
<v Morgan Dwyer (OpenAI)>to make our thinking on this public
through our reporting framework.

3289
03:15:01.480 --> 03:15:05.641
<v Morgan Dwyer (OpenAI)>I'm gonna I'm gonna pause there because I
can't remember the rest of your questions.

3290
03:15:09.000 --> 03:15:13.000
<v CM Virginia Maloney>Why don't we focus on the first question
which was on measurement and hear from all

3291
03:15:13.000 --> 03:15:13.960
<v CM Virginia Maloney>all four companies?

3292
03:15:17.234 --> 03:15:19.395
<v Shane Cahill (Meta)>Council member, specifically

3293
03:15:19.395 --> 03:15:21.235
<v Shane Cahill (Meta)>in relation to to

3294
03:15:22.114 --> 03:15:22.995
<v Shane Cahill (Meta)>incident,

3295
03:15:23.395 --> 03:15:26.755
<v Shane Cahill (Meta)>I think that's what... I think that was
the... Correct me if I'm wrong. I think that

3296
03:15:26.755 --> 03:15:27.715
<v Shane Cahill (Meta)>was the first piece

3297
03:15:28.120 --> 03:15:31.800
<v Shane Cahill (Meta)>in relation to to tracking
and how we define it.

3298
03:15:32.200 --> 03:15:33.800
<v Shane Cahill (Meta)>We set this out in

3299
03:15:34.200 --> 03:15:37.560
<v Shane Cahill (Meta)>our scaling framework, which
we've made publicly available.

3300
03:15:37.720 --> 03:15:41.720
<v Shane Cahill (Meta)>I don't want to be imprecise as I don't
have it in front of me, so I don't have it

3301
03:15:41.720 --> 03:15:42.359
<v Shane Cahill (Meta)>to quote to you.

3302
03:15:43.465 --> 03:15:47.065
<v Shane Cahill (Meta)>So please let me follow-up with you afterwards
and connect you with the right people

3303
03:15:47.065 --> 03:15:48.186
<v Shane Cahill (Meta)>to talk to you about.

3304
03:15:52.025 --> 03:15:56.010
<v Alice Friend (Google)>Yes, ma'am. I think this is a critically
important question as we're debating with

3305
03:15:56.010 --> 03:15:59.609
<v Alice Friend (Google)>multiple regulators around the country
and around the world right now what

3306
03:15:59.609 --> 03:16:01.290
<v Alice Friend (Google)>constitutes an incident,

3307
03:16:01.770 --> 03:16:04.649
<v Alice Friend (Google)>particularly when you think
about the possibility that,

3308
03:16:05.450 --> 03:16:10.965
<v Alice Friend (Google)>we could overreport incidents, some of
which may not be of a serious enough nature

3309
03:16:10.965 --> 03:16:14.485
<v Alice Friend (Google)>for public concern. We
wanna ensure that we narrow,

3310
03:16:15.045 --> 03:16:19.046
<v Alice Friend (Google)>to the class of incidents that
are truly important for us to use,

3311
03:16:19.525 --> 03:16:22.900
<v Alice Friend (Google)>to learn to improve safety
and also for us to share,

3312
03:16:23.540 --> 03:16:25.780
<v Alice Friend (Google)>with you all to collectively

3313
03:16:25.780 --> 03:16:27.380
<v Alice Friend (Google)>improve public safety.

3314
03:16:28.020 --> 03:16:32.101
<v Alice Friend (Google)>So a few things come to mind, as
we've been having these debates.

3315
03:16:32.260 --> 03:16:33.939
<v Alice Friend (Google)>One, of course, is cybersecurity

3316
03:16:33.939 --> 03:16:36.420
<v Alice Friend (Google)>incidents incidents covering

3317
03:16:37.035 --> 03:16:41.435
<v Alice Friend (Google)>the security of our most
powerful AI models themselves.

3318
03:16:41.435 --> 03:16:45.995
<v Alice Friend (Google)>That's been under discussion for several
years now protecting model weights in

3319
03:16:45.995 --> 03:16:46.715
<v Alice Friend (Google)>particular.

3320
03:16:47.035 --> 03:16:48.475
<v Alice Friend (Google)>We also

3321
03:16:48.990 --> 03:16:53.710
<v Alice Friend (Google)>talk a lot about focusing on
incidents that cause material harms

3322
03:16:53.950 --> 03:16:57.630
<v Alice Friend (Google)>in one way or another, and I think
part of the discussion will be,

3323
03:16:58.270 --> 03:17:00.910
<v Alice Friend (Google)>you know, how we define
those with precision.

3324
03:17:02.385 --> 03:17:06.146
<v Alice Friend (Google)>We also consider incidents to be
those that violate our policies.

3325
03:17:06.305 --> 03:17:10.545
<v Alice Friend (Google)>You asked us about how we respond to
incidents, and primarily our trust and safety

3326
03:17:10.545 --> 03:17:11.825
<v Alice Friend (Google)>teams are responsible,

3327
03:17:12.225 --> 03:17:14.385
<v Alice Friend (Google)>for responding to what
we call escalations,

3328
03:17:14.385 --> 03:17:16.785
<v Alice Friend (Google)>which usually involve
violation of those policies.

3329
03:17:17.540 --> 03:17:18.820
<v Alice Friend (Google)>First, reviewing,

3330
03:17:19.220 --> 03:17:20.660
<v Alice Friend (Google)>the alleged violation,

3331
03:17:20.660 --> 03:17:22.340
<v Alice Friend (Google)>are often reported to us,

3332
03:17:23.300 --> 03:17:24.820
<v Alice Friend (Google)>by external parties.

3333
03:17:25.300 --> 03:17:29.220
<v Alice Friend (Google)>And then if we determine that, in
fact, our policy has been violated,

3334
03:17:29.300 --> 03:17:31.939
<v Alice Friend (Google)>we have a a variety of
different actions we might take.

3335
03:17:33.245 --> 03:17:36.125
<v Alice Friend (Google)>And then, of course, a
violation of the law.

3336
03:17:36.445 --> 03:17:40.285
<v Alice Friend (Google)>As I've said a couple of times in the
hearing, if it's illegal without AI, it's

3337
03:17:40.285 --> 03:17:44.046
<v Alice Friend (Google)>still illegal with AI. And so,
I would think that incidents,

3338
03:17:45.165 --> 03:17:49.141
<v Alice Friend (Google)>that need to be flagged would, of
course, involve violation of the law.

3339
03:17:49.620 --> 03:17:53.380
<v Alice Friend (Google)>As to most important safeguards, it's
very hard to answer that question purely

3340
03:17:53.380 --> 03:17:58.420
<v Alice Friend (Google)>because there's a range of different, safety
considerations. We've talked about many

3341
03:17:58.420 --> 03:17:58.980
<v Alice Friend (Google)>of them,

3342
03:17:59.460 --> 03:18:00.900
<v Alice Friend (Google)>today during the hearing.

3343
03:18:01.505 --> 03:18:05.984
<v Alice Friend (Google)>And, again, several of us have referred to
this layered approach. My own background

3344
03:18:05.984 --> 03:18:10.944
<v Alice Friend (Google)>is in national security. We talk about
layered security in that context as well. So

3345
03:18:10.944 --> 03:18:14.680
<v Alice Friend (Google)>you wouldn't really want to talk about
a single safeguard. You would wanna talk

3346
03:18:14.680 --> 03:18:16.520
<v Alice Friend (Google)>about layers of safeguards

3347
03:18:16.520 --> 03:18:21.000
<v Alice Friend (Google)>that together create a much more assurance
and confidence in the safety of a whole

3348
03:18:21.000 --> 03:18:21.640
<v Alice Friend (Google)>system.

3349
03:18:22.760 --> 03:18:25.240
<v CM Virginia Maloney>Thank you very much. The testimony

3350
03:18:25.240 --> 03:18:29.575
<v CM Virginia Maloney>helps highlight that there's
more work to be done to classify

3351
03:18:29.575 --> 03:18:34.615
<v CM Virginia Maloney>what's incidents in the future and create
an industry standard for what would

3352
03:18:34.615 --> 03:18:38.535
<v CM Virginia Maloney>require disclosure and what would require
action. The second part of the question

3353
03:18:38.535 --> 03:18:42.300
<v CM Virginia Maloney>was to walk through what happens
after an incident occurs,

3354
03:18:42.460 --> 03:18:45.980
<v CM Virginia Maloney>who owns the fix, and who decides whether
or not the model or the product keeps

3355
03:18:45.980 --> 03:18:47.900
<v CM Virginia Maloney>running in the meantime. Thank you.

3356
03:18:51.100 --> 03:18:51.980
<v Morgan Dwyer (OpenAI)>Yes. So,

3357
03:18:53.020 --> 03:18:55.045
<v Morgan Dwyer (OpenAI)>OpenAI has instituted

3358
03:18:55.205 --> 03:18:58.085
<v Morgan Dwyer (OpenAI)>monitoring during testing and evaluation.

3359
03:18:58.725 --> 03:19:04.885
<v Morgan Dwyer (OpenAI)>That monitoring creates automated alerts
to our research and security engineering

3360
03:19:04.885 --> 03:19:08.005
<v Morgan Dwyer (OpenAI)>teams to let them know
of potential incidents.

3361
03:19:09.000 --> 03:19:11.320
<v Morgan Dwyer (OpenAI)>And then we have recently clarified

3362
03:19:11.320 --> 03:19:13.400
<v Morgan Dwyer (OpenAI)>our escalation processes

3363
03:19:13.400 --> 03:19:15.000
<v Morgan Dwyer (OpenAI)>as well as responsibilities

3364
03:19:15.000 --> 03:19:16.120
<v Morgan Dwyer (OpenAI)>across teams,

3365
03:19:17.160 --> 03:19:20.120
<v Morgan Dwyer (OpenAI)>and made it clear the process for

3366
03:19:20.715 --> 03:19:21.595
<v Morgan Dwyer (OpenAI)>pausing

3367
03:19:21.595 --> 03:19:22.396
<v Morgan Dwyer (OpenAI)>activity

3368
03:19:22.795 --> 03:19:25.436
<v Morgan Dwyer (OpenAI)>and when it is safe to restart.

3369
03:19:28.715 --> 03:19:32.315
<v Shane Cahill (Meta)>Council member, in relation to
your second question, we have

3370
03:19:32.555 --> 03:19:35.515
<v Shane Cahill (Meta)>an extensive team, actually
extensive teams of people

3371
03:19:36.109 --> 03:19:36.830
<v Shane Cahill (Meta)>who,

3372
03:19:36.910 --> 03:19:40.030
<v Shane Cahill (Meta)>part of their day to day,
every day, day in and day out,

3373
03:19:40.670 --> 03:19:42.830
<v Shane Cahill (Meta)>are leading and responsible

3374
03:19:42.910 --> 03:19:43.790
<v Shane Cahill (Meta)>for

3375
03:19:43.950 --> 03:19:46.990
<v Shane Cahill (Meta)>the development, deployment,
and use of our AI systems.

3376
03:19:51.935 --> 03:19:56.975
<v Alice Friend (Google)>I'll just say, ma'am, I, I was attempting
to answer that question when I referred to

3377
03:19:56.975 --> 03:20:00.815
<v Alice Friend (Google)>the process internally with our trust and
safety teams. But if you'd like some more

3378
03:20:00.815 --> 03:20:02.975
<v Alice Friend (Google)>detail, I'm happy to
follow-up after the hearing.

3379
03:20:05.040 --> 03:20:06.480
<v Logan Graham (Anthropic)>Thank you, Jared. Likewise,

3380
03:20:07.119 --> 03:20:08.399
<v Logan Graham (Anthropic)>the process

3381
03:20:08.399 --> 03:20:09.279
<v Logan Graham (Anthropic)>involves

3382
03:20:09.279 --> 03:20:13.279
<v Logan Graham (Anthropic)>multiple different teams depending on the
nature of the of the incident. If it's at

3383
03:20:13.279 --> 03:20:14.560
<v Logan Graham (Anthropic)>the core of security

3384
03:20:14.800 --> 03:20:16.239
<v Logan Graham (Anthropic)>and our own security,

3385
03:20:16.896 --> 03:20:20.415
<v Logan Graham (Anthropic)>we require a security team to work on
this. If it's about the model behavior and

3386
03:20:20.415 --> 03:20:24.336
<v Logan Graham (Anthropic)>misalignment, it might be a research
or training or safety consideration.

3387
03:20:24.415 --> 03:20:28.495
<v Logan Graham (Anthropic)>If it's about incidents of misuse on the
platform, we have the safeguards team.

3388
03:20:29.090 --> 03:20:33.649
<v Logan Graham (Anthropic)>I think you're pointing at this is becoming
more complex over time. And I think this

3389
03:20:33.649 --> 03:20:37.489
<v Logan Graham (Anthropic)>is one reason, as my colleagues mentioned
here, this is a frontier of defining both

3390
03:20:37.489 --> 03:20:38.609
<v Logan Graham (Anthropic)>what is an incident,

3391
03:20:38.930 --> 03:20:42.369
<v Logan Graham (Anthropic)>and how or what would a good process
look like for for dealing with them.

3392
03:20:45.605 --> 03:20:49.206
<v CM Carmen De La Rosa>Thank you. We are going to
take a five minute break,

3393
03:20:49.605 --> 03:20:54.086
<v CM Carmen De La Rosa>and we will be back with council members
Wilson, Jose, Hudson, and then Shulman.

3394
03:29:19.240 --> 03:29:23.561
<v CM Carmen De La Rosa>Alright. We're gonna start back up
again. So if our panel could come back

3395
03:29:23.960 --> 03:29:26.121
<v CM Carmen De La Rosa>on screen, we appreciate it.

3396
03:29:26.761 --> 03:29:30.841
<v CM Carmen De La Rosa>And we're gonna start with
council member Wilson,

3397
03:29:31.080 --> 03:29:34.120
<v CM Carmen De La Rosa>followed by council member Jose,
followed by council member Hudson.

3398
03:29:35.745 --> 03:29:40.944
<v CM Carl Wilson>Thank you, chair. My questions are
related to my bill, which deals with the

3399
03:29:40.944 --> 03:29:41.904
<v CM Carl Wilson>disclosures

3400
03:29:41.904 --> 03:29:43.585
<v CM Carl Wilson>and prohibiting deception,

3401
03:29:43.824 --> 03:29:46.625
<v CM Carl Wilson>deceptive advertising on AI models. So

3402
03:29:47.110 --> 03:29:51.110
<v CM Carl Wilson>my question is what guardrails
on misleading advertising

3403
03:29:51.110 --> 03:29:55.190
<v CM Carl Wilson>and truthful disclosures do you wish
your competitors had to abide by?

3404
03:29:55.590 --> 03:29:57.510
<v CM Carl Wilson>And given the complexity

3405
03:29:57.510 --> 03:30:03.404
<v CM Carl Wilson>of these tools and AI models, where
are consumers most likely to be misled

3406
03:30:03.404 --> 03:30:04.685
<v CM Carl Wilson>or confused?

3407
03:30:04.925 --> 03:30:09.805
<v CM Carl Wilson>And what do you wish consumers understood
better about these tools, capabilities,

3408
03:30:09.805 --> 03:30:10.285
<v CM Carl Wilson>and risk?

3409
03:30:13.699 --> 03:30:14.980
<v Morgan Dwyer (OpenAI)>So OpenAI's

3410
03:30:14.980 --> 03:30:18.500
<v Morgan Dwyer (OpenAI)>usage policies prohibit
the use of our tools

3411
03:30:18.580 --> 03:30:21.300
<v Morgan Dwyer (OpenAI)>for scams, fraud, and impersonation.

3412
03:30:22.100 --> 03:30:26.659
<v Morgan Dwyer (OpenAI)>We also embed in every single
one of our images and audio

3413
03:30:27.115 --> 03:30:27.995
<v Morgan Dwyer (OpenAI)>content

3414
03:30:27.995 --> 03:30:28.875
<v Morgan Dwyer (OpenAI)>provenance,

3415
03:30:29.195 --> 03:30:31.275
<v Morgan Dwyer (OpenAI)>or watermarks or metadata,

3416
03:30:31.755 --> 03:30:34.075
<v Morgan Dwyer (OpenAI)>to tell people if the,

3417
03:30:34.795 --> 03:30:36.875
<v Morgan Dwyer (OpenAI)>information was AI generated.

3418
03:30:37.355 --> 03:30:42.220
<v Morgan Dwyer (OpenAI)>And we make tools available so that
people can assess whether their,

3419
03:30:42.700 --> 03:30:46.141
<v Morgan Dwyer (OpenAI)>content that they are
seeing is AI generated.

3420
03:30:50.780 --> 03:30:54.940
<v Shane Cahill (Meta)>Council member, transparency is
absolutely foundational to accountability

3421
03:30:54.940 --> 03:30:55.580
<v Shane Cahill (Meta)>and trust.

3422
03:30:56.194 --> 03:31:00.835
<v Shane Cahill (Meta)>For example, our scaling framework
has a change log which details

3423
03:31:01.395 --> 03:31:04.115
<v Shane Cahill (Meta)>each change that we have
made to our framework

3424
03:31:04.354 --> 03:31:06.995
<v Shane Cahill (Meta)>just as an example of the transparency

3425
03:31:06.995 --> 03:31:08.515
<v Shane Cahill (Meta)>that underlines our approach.

3426
03:31:09.580 --> 03:31:12.940
<v Shane Cahill (Meta)>In relation to your
question about competitors,

3427
03:31:13.020 --> 03:31:17.101
<v Shane Cahill (Meta)>I would just note that I
generally think that there are

3428
03:31:17.500 --> 03:31:18.621
<v Shane Cahill (Meta)>significant

3429
03:31:20.540 --> 03:31:24.061
<v Shane Cahill (Meta)>laws already at federal and state
level regarding misrepresentations.

3430
03:31:28.564 --> 03:31:33.444
<v Alice Friend (Google)>Yes, sir. I I can honestly say, the
most important thing to us is continued

3431
03:31:33.444 --> 03:31:35.365
<v Alice Friend (Google)>technical standardization across

3432
03:31:35.365 --> 03:31:37.045
<v Alice Friend (Google)>the industry so that we have,

3433
03:31:37.285 --> 03:31:40.005
<v Alice Friend (Google)>common ways to operate and identify

3434
03:31:41.420 --> 03:31:45.980
<v Alice Friend (Google)>the providence of information. We're
we're very heartened by the broad adoption

3435
03:31:46.140 --> 03:31:50.141
<v Alice Friend (Google)>of the c two p a content credential
that I referenced before.

3436
03:31:50.460 --> 03:31:54.780
<v Alice Friend (Google)>I also referenced before that we built
our own Synth ID watermark... Marking

3437
03:31:54.780 --> 03:31:55.340
<v Alice Friend (Google)>technology

3438
03:31:55.805 --> 03:32:00.125
<v Alice Friend (Google)>that many across industry
have adopted as one tool,

3439
03:32:00.685 --> 03:32:02.365
<v Alice Friend (Google)>in order to identify

3440
03:32:02.524 --> 03:32:05.805
<v Alice Friend (Google)>when content is, generative AI created.

3441
03:32:07.725 --> 03:32:12.590
<v Alice Friend (Google)>One of the things that we do a lot of
at Google is do research into how users

3442
03:32:12.590 --> 03:32:14.030
<v Alice Friend (Google)>interact with information,

3443
03:32:14.109 --> 03:32:14.830
<v Alice Friend (Google)>you know,

3444
03:32:15.550 --> 03:32:17.310
<v Alice Friend (Google)>search as part of our DNA.

3445
03:32:17.790 --> 03:32:23.230
<v Alice Friend (Google)>And one of the interesting things about
labeling that we found is that there are

3446
03:32:23.230 --> 03:32:25.710
<v Alice Friend (Google)>users who, when they see a label,

3447
03:32:26.905 --> 03:32:27.545
<v Alice Friend (Google)>they,

3448
03:32:27.865 --> 03:32:30.905
<v Alice Friend (Google)>are less likely to to
trust that information,

3449
03:32:31.065 --> 03:32:34.025
<v Alice Friend (Google)>but they are more likely to
trust unlabeled information.

3450
03:32:34.425 --> 03:32:37.065
<v Alice Friend (Google)>And just because information
that you see online,

3451
03:32:37.305 --> 03:32:41.460
<v Alice Friend (Google)>doesn't have a watermark or a label
on it doesn't necessarily mean it is,

3452
03:32:42.020 --> 03:32:43.860
<v Alice Friend (Google)>automatically trustworthy either.

3453
03:32:44.020 --> 03:32:49.380
<v Alice Friend (Google)>So we, work through and think very hard
about what are the the ways that we can give

3454
03:32:49.380 --> 03:32:50.020
<v Alice Friend (Google)>users,

3455
03:32:50.260 --> 03:32:51.141
<v Alice Friend (Google)>information,

3456
03:32:51.700 --> 03:32:55.855
<v Alice Friend (Google)>that they will be able to know the
context of what they're interacting with,

3457
03:32:56.175 --> 03:32:58.495
<v Alice Friend (Google)>and make sound judgments about,

3458
03:32:58.815 --> 03:33:02.175
<v Alice Friend (Google)>which information is trustworthy and which
isn't. So this is this is part of the

3459
03:33:02.175 --> 03:33:03.296
<v Alice Friend (Google)>research that we do.

3460
03:33:05.215 --> 03:33:06.976
<v Logan Graham (Anthropic)>Council member, we,

3461
03:33:07.215 --> 03:33:09.120
<v Logan Graham (Anthropic)>our view is that we want

3462
03:33:09.120 --> 03:33:12.160
<v Logan Graham (Anthropic)>Claude, our product, to act unambiguously

3463
03:33:12.160 --> 03:33:13.681
<v Logan Graham (Anthropic)>in the user's interests.

3464
03:33:14.160 --> 03:33:17.841
<v Logan Graham (Anthropic)>For that reason, we do not run
ads nor are Claude's responses,

3465
03:33:18.240 --> 03:33:20.801
<v Logan Graham (Anthropic)>influenced by those of advertisers.

3466
03:33:21.280 --> 03:33:23.841
<v Logan Graham (Anthropic)>And furthermore, Claude
does not generate images.

3467
03:33:24.320 --> 03:33:25.040
<v Logan Graham (Anthropic)>And so

3468
03:33:25.734 --> 03:33:29.575
<v Logan Graham (Anthropic)>we we serve... Try to reduce
the risk surface in this way.

3469
03:33:31.895 --> 03:33:36.055
<v CM Carmen De La Rosa>Thank you. Council member Jose, followed
by council member Hudson, followed by

3470
03:33:36.055 --> 03:33:37.015
<v CM Carmen De La Rosa>council member Schulman.

3471
03:33:37.450 --> 03:33:41.850
<v CM Chi Ossé>Thank you, chair. Mister Graham, you've
indicated that you run red teaming at

3472
03:33:41.850 --> 03:33:42.730
<v CM Chi Ossé>Anthropic.

3473
03:33:42.730 --> 03:33:47.130
<v CM Chi Ossé>Will your company commit a certain amount
of compute for free to New York City and

3474
03:33:47.130 --> 03:33:50.971
<v CM Chi Ossé>other municipal governments to research
vulnerabilities and build defenses?

3475
03:33:51.370 --> 03:33:54.455
<v CM Chi Ossé>And if yes, how much compute indefinitely?

3476
03:33:54.455 --> 03:33:58.535
<v CM Chi Ossé>And I'll ask the same question for the
three other companies. I know that Meta

3477
03:33:58.535 --> 03:34:02.615
<v CM Chi Ossé>indicated that that is something that
they would be interested in doing. So for

3478
03:34:02.615 --> 03:34:04.375
<v CM Chi Ossé>Meta, how much compute indefinitely?

3479
03:34:06.601 --> 03:34:10.760
<v Logan Graham (Anthropic)>Well, I'm not in a position to commit
anything today. I'm heartened by this this

3480
03:34:10.760 --> 03:34:15.240
<v Logan Graham (Anthropic)>question. We spend a lot of our
resources internally on safety,

3481
03:34:15.480 --> 03:34:17.561
<v Logan Graham (Anthropic)>alignment, and defensive practices.

3482
03:34:17.720 --> 03:34:18.680
<v Logan Graham (Anthropic)>But moreover,

3483
03:34:18.920 --> 03:34:24.024
<v Logan Graham (Anthropic)>this year, we've taken very large steps
to provide external access and advanced

3484
03:34:24.024 --> 03:34:27.865
<v Logan Graham (Anthropic)>access of models to cyber
defenders for this reason,

3485
03:34:28.185 --> 03:34:33.785
<v Logan Graham (Anthropic)>including at the city and state level,
in New York. I'm personally looking to,

3486
03:34:34.104 --> 03:34:38.430
<v Logan Graham (Anthropic)>try to further these efforts, at Anthropic.
It's It's a question of many ways we

3487
03:34:38.430 --> 03:34:39.551
<v Logan Graham (Anthropic)>could support this.

3488
03:34:39.950 --> 03:34:44.590
<v Logan Graham (Anthropic)>We support free cloud credits, at the New
York City level with Cyber Command, for

3489
03:34:44.590 --> 03:34:48.990
<v Logan Graham (Anthropic)>example, a number of state and local leaders.
But it goes much further than that. It

3490
03:34:48.990 --> 03:34:52.086
<v Logan Graham (Anthropic)>goes to all pieces of
critical infrastructure,

3491
03:34:52.165 --> 03:34:55.445
<v Logan Graham (Anthropic)>releasing research, safety
reports, technical know how.

3492
03:34:55.766 --> 03:34:58.726
<v Logan Graham (Anthropic)>It's a it's a very, very large challenge
and one very close to my heart. I'm glad

3493
03:34:58.726 --> 03:34:59.125
<v Logan Graham (Anthropic)>you asked.

3494
03:35:04.420 --> 03:35:05.540
<v CM Chi Ossé>Dwyer,

3495
03:35:05.700 --> 03:35:09.700
<v CM Chi Ossé>Friend, you know, it... I think we have
we have the the order already established

3496
03:35:09.700 --> 03:35:15.540
<v CM Chi Ossé>here. Just... Is is is the company's position
open to providing free comp... Compute

3497
03:35:15.620 --> 03:35:16.820
<v CM Chi Ossé>to New York City?

3498
03:35:18.455 --> 03:35:20.295
<v Morgan Dwyer (OpenAI)>I'll I'll take that next.

3499
03:35:20.535 --> 03:35:21.976
<v Morgan Dwyer (OpenAI)>Thank you for the question.

3500
03:35:22.455 --> 03:35:25.735
<v Morgan Dwyer (OpenAI)>I'm happy to share that
OpenAI has already committed,

3501
03:35:26.215 --> 03:35:28.375
<v Morgan Dwyer (OpenAI)>to donating $1,000,000,000,

3502
03:35:28.615 --> 03:35:34.570
<v Morgan Dwyer (OpenAI)>towards strengthening cyber defense through
our Daybreak for Defenders initiative.

3503
03:35:35.210 --> 03:35:37.690
<v Morgan Dwyer (OpenAI)>That covers things like providing,

3504
03:35:38.170 --> 03:35:42.891
<v Morgan Dwyer (OpenAI)>access to our most advanced AI
models for use in cyber defense.

3505
03:35:43.290 --> 03:35:46.170
<v Morgan Dwyer (OpenAI)>It also includes technical
assistance and training.

3506
03:35:46.726 --> 03:35:49.686
<v Morgan Dwyer (OpenAI)>And we're already working
with the Port Authority,

3507
03:35:50.005 --> 03:35:51.766
<v Morgan Dwyer (OpenAI)>as well as with New York State.

3508
03:35:54.565 --> 03:35:55.286
<v CM Carmen De La Rosa>Kehu.

3509
03:35:55.445 --> 03:35:57.766
<v Shane Cahill (Meta)>Yep. Thank you so much, council member.

3510
03:35:57.926 --> 03:36:02.480
<v Shane Cahill (Meta)>So that's a great question. In relation
to specifically your question on compute,

3511
03:36:02.480 --> 03:36:06.240
<v Shane Cahill (Meta)>let me come back to you on that. I want to
make sure that I'm providing you with the

3512
03:36:06.240 --> 03:36:09.841
<v Shane Cahill (Meta)>the right information. So I'll talk to
the right people and come back to you on

3513
03:36:09.841 --> 03:36:15.955
<v Shane Cahill (Meta)>that. But I just wanted to note in relation
to to red teaming, we work with external

3514
03:36:16.435 --> 03:36:17.314
<v Shane Cahill (Meta)>evaluators

3515
03:36:17.314 --> 03:36:21.555
<v Shane Cahill (Meta)>and to note that we signed the
commitment at the Accord to

3516
03:36:22.194 --> 03:36:24.195
<v Shane Cahill (Meta)>partner with independent

3517
03:36:24.274 --> 03:36:26.435
<v Shane Cahill (Meta)>external evaluators and assessors.

3518
03:36:27.101 --> 03:36:30.221
<v Shane Cahill (Meta)>This is also part of
our scaling framework.

3519
03:36:30.620 --> 03:36:35.340
<v Shane Cahill (Meta)>I just wanted to mention as well the
absolutely important and critical role of open

3520
03:36:35.340 --> 03:36:38.461
<v Shane Cahill (Meta)>weight models in relation to cybersecurity

3521
03:36:38.780 --> 03:36:41.420
<v Shane Cahill (Meta)>and hardening cyber...
Cybersecurity in particular.

3522
03:36:42.435 --> 03:36:48.354
<v Shane Cahill (Meta)>We have our Muse Glimmer model, which is
open weights, and we really believe that in

3523
03:36:48.354 --> 03:36:52.274
<v Shane Cahill (Meta)>addition to furthering innovation and
research, open weight models have a really

3524
03:36:52.274 --> 03:36:56.540
<v Shane Cahill (Meta)>significant role to play in furthering
the state of art in terms of heighten...

3525
03:36:56.540 --> 03:36:57.900
<v Shane Cahill (Meta)>Hardening cybersecurity.

3526
03:37:00.779 --> 03:37:03.660
<v Alice Friend (Google)>And, sir, I'm gonna have
to reach out to our,

3527
03:37:04.220 --> 03:37:08.220
<v Alice Friend (Google)>cloud business specialists and get back
to you on that. I don't work on technical

3528
03:37:08.220 --> 03:37:13.035
<v Alice Friend (Google)>infrastructure, and so I really
can't comment or commit to to,

3529
03:37:13.354 --> 03:37:14.075
<v Alice Friend (Google)>compute,

3530
03:37:14.155 --> 03:37:18.555
<v Alice Friend (Google)>at this time, but I can pledge to get back
to you on that. Okay. I I would just say

3531
03:37:18.555 --> 03:37:19.835
<v Alice Friend (Google)>that if

3532
03:37:20.075 --> 03:37:23.115
<v CM Chi Ossé>it is in the interest of
these companies to provide

3533
03:37:23.750 --> 03:37:28.390
<v CM Chi Ossé>any type of safety protocols for
municipalities like New York or just for any

3534
03:37:28.390 --> 03:37:32.230
<v CM Chi Ossé>governments, you know, across the
nation, nonetheless, the the the globe,

3535
03:37:32.630 --> 03:37:35.590
<v CM Chi Ossé>I think this this would be a great tool,

3536
03:37:35.910 --> 03:37:40.355
<v CM Chi Ossé>for us to use in order to keep our civilians
safe. So I know that some folks said

3537
03:37:40.355 --> 03:37:41.796
<v CM Chi Ossé>that they would get back to me.

3538
03:37:42.195 --> 03:37:45.476
<v CM Chi Ossé>I would love to hear... This council
would love to hear back from all of your

3539
03:37:45.476 --> 03:37:46.195
<v CM Chi Ossé>companies,

3540
03:37:46.596 --> 03:37:50.835
<v CM Chi Ossé>on this possibility. Additionally, all of
you have said on the stand today under

3541
03:37:50.835 --> 03:37:55.930
<v CM Chi Ossé>oath that your companies will support and
work with council on any AI regulation and

3542
03:37:55.930 --> 03:37:57.290
<v CM Chi Ossé>safety measures

3543
03:37:57.370 --> 03:38:00.890
<v CM Chi Ossé>and have supported measures like the RAISE
Act. I I know that a couple companies

3544
03:38:00.890 --> 03:38:04.250
<v CM Chi Ossé>said that they supported the RAISE
Act. But but let's be real here.

3545
03:38:04.570 --> 03:38:06.730
<v CM Chi Ossé>It is well known that a... OpenAI,

3546
03:38:06.730 --> 03:38:07.530
<v CM Chi Ossé>Anthropic,

3547
03:38:07.530 --> 03:38:07.930
<v CM Chi Ossé>Google,

3548
03:38:08.365 --> 03:38:13.245
<v CM Chi Ossé>Meta lobbied against the RAISE Act for
a year. But all of a sudden, once the

3549
03:38:13.245 --> 03:38:17.325
<v CM Chi Ossé>governor signed it into law a month later,
your company started saying you support

3550
03:38:17.325 --> 03:38:18.365
<v CM Chi Ossé>this legislation.

3551
03:38:18.845 --> 03:38:23.870
<v CM Chi Ossé>As many people know here, the RAISE Act is
the floor of AI safety, not the ceiling.

3552
03:38:23.870 --> 03:38:28.351
<v CM Chi Ossé>And with that said, do you support or oppose
the federal government blocking states

3553
03:38:28.351 --> 03:38:30.910
<v CM Chi Ossé>and local governments from regulating AI?

3554
03:38:31.150 --> 03:38:32.910
<v CM Chi Ossé>Because that is the position
of the federal government.

3555
03:38:35.824 --> 03:38:38.945
<v Morgan Dwyer (OpenAI)>So I I will start. Thank
you for the question.

3556
03:38:39.585 --> 03:38:43.824
<v Morgan Dwyer (OpenAI)>So we have been clear that we think the
federal government should lead in frontier

3557
03:38:43.824 --> 03:38:45.265
<v Morgan Dwyer (OpenAI)>safety regulation.

3558
03:38:45.984 --> 03:38:48.944
<v Morgan Dwyer (OpenAI)>But in the absence of federal regulation,

3559
03:38:48.944 --> 03:38:52.570
<v Morgan Dwyer (OpenAI)>we believe it's important for
states to advance harmonized

3560
03:38:52.570 --> 03:38:54.650
<v Morgan Dwyer (OpenAI)>approaches to frontier safety.

3561
03:38:55.290 --> 03:38:56.971
<v Morgan Dwyer (OpenAI)>I'm happy to share that,

3562
03:38:57.370 --> 03:38:59.370
<v Morgan Dwyer (OpenAI)>OpenAI has supported,

3563
03:38:59.930 --> 03:39:04.570
<v Morgan Dwyer (OpenAI)>legislation that is stronger than the
RAISE Act. In Illinois, this year, we

3564
03:39:04.570 --> 03:39:05.210
<v Morgan Dwyer (OpenAI)>supported

3565
03:39:05.405 --> 03:39:09.965
<v Morgan Dwyer (OpenAI)>a frontier safety bill that would
require third party audits of our safety

3566
03:39:09.965 --> 03:39:10.686
<v Morgan Dwyer (OpenAI)>framework,

3567
03:39:11.085 --> 03:39:13.485
<v Morgan Dwyer (OpenAI)>and we're committed to
working with policymakers

3568
03:39:13.485 --> 03:39:15.245
<v Morgan Dwyer (OpenAI)>to continue raising the floor,

3569
03:39:15.806 --> 03:39:17.005
<v Morgan Dwyer (OpenAI)>on frontier safety.

3570
03:39:17.930 --> 03:39:21.370
<v CM Chi Ossé>I... I'm I'm just gonna push
back there because, you know,

3571
03:39:22.090 --> 03:39:27.290
<v CM Chi Ossé>I I I support the RAISE Act. I think it it
was an incredible start on AI regulation

3572
03:39:27.290 --> 03:39:30.490
<v CM Chi Ossé>here in New York State. And
while you're saying that OpenAI

3573
03:39:30.569 --> 03:39:31.529
<v CM Chi Ossé>has supported,

3574
03:39:31.955 --> 03:39:35.795
<v CM Chi Ossé>you know, your words, stronger
AI regulation, OpenAI

3575
03:39:35.795 --> 03:39:36.835
<v CM Chi Ossé>leadership,

3576
03:39:37.235 --> 03:39:41.955
<v CM Chi Ossé>has also lobbied against congressional
candidates who promise to work on AI

3577
03:39:41.955 --> 03:39:42.915
<v CM Chi Ossé>legislation.

3578
03:39:43.075 --> 03:39:47.155
<v CM Chi Ossé>So how can we trust your companies, not just
OpenAI, but all four of your companies,

3579
03:39:47.480 --> 03:39:52.280
<v CM Chi Ossé>not to fight against AI regulations when
your political activities and contributions

3580
03:39:52.280 --> 03:39:53.561
<v CM Chi Ossé>do not say the same?

3581
03:39:57.881 --> 03:39:59.641
<v Morgan Dwyer (OpenAI)>So I I will start again. Thanks.

3582
03:40:00.645 --> 03:40:05.524
<v Morgan Dwyer (OpenAI)>Look. I can't speak for the personal actions
of open eye... OpenAI leaders that they

3583
03:40:05.524 --> 03:40:07.365
<v Morgan Dwyer (OpenAI)>have taken in their personal capacity.

3584
03:40:07.765 --> 03:40:12.885
<v Morgan Dwyer (OpenAI)>What I can tell you is that we have endorsed
strong regulations both on frontier

3585
03:40:12.885 --> 03:40:14.085
<v Morgan Dwyer (OpenAI)>safety in Illinois

3586
03:40:14.700 --> 03:40:18.300
<v Morgan Dwyer (OpenAI)>as well as on youth safety in California.

3587
03:40:18.780 --> 03:40:21.580
<v Morgan Dwyer (OpenAI)>And we are committed to
working with policymakers,

3588
03:40:21.820 --> 03:40:27.420
<v Morgan Dwyer (OpenAI)>to strengthen regulation and to ensure
that all AI is developed responsibly.

3589
03:40:31.025 --> 03:40:34.064
<v Shane Cahill (Meta)>Mister Cahill? Yeah.
Thank you. Council member,

3590
03:40:34.385 --> 03:40:35.745
<v Shane Cahill (Meta)>in relation to,

3591
03:40:36.545 --> 03:40:37.505
<v Shane Cahill (Meta)>your question,

3592
03:40:38.145 --> 03:40:43.905
<v Shane Cahill (Meta)>I... How I think about it broadly is
that, it is, critical that The US

3593
03:40:43.905 --> 03:40:48.020
<v Shane Cahill (Meta)>and democratic countries lead
in relation to AI innovation.

3594
03:40:48.500 --> 03:40:49.780
<v Shane Cahill (Meta)>In this regard,

3595
03:40:50.100 --> 03:40:51.460
<v Shane Cahill (Meta)>AI is

3596
03:40:51.859 --> 03:40:52.500
<v Shane Cahill (Meta)>trans...

3597
03:40:53.300 --> 03:40:54.180
<v Shane Cahill (Meta)>Transcontinental,

3598
03:40:54.180 --> 03:40:58.100
<v Shane Cahill (Meta)>I guess. I was gonna use transnational,
but transcontinental in the case of The US.

3599
03:40:58.676 --> 03:41:01.396
<v Shane Cahill (Meta)>And, therefore, it's really
important and critical

3600
03:41:01.556 --> 03:41:02.516
<v Shane Cahill (Meta)>that

3601
03:41:02.596 --> 03:41:08.915
<v Shane Cahill (Meta)>regulations and obligations are consistent
across the board if it is to be true that

3602
03:41:08.915 --> 03:41:14.520
<v Shane Cahill (Meta)>The US and democratic countries can continue
to lead in this... In in AI innovation.

3603
03:41:14.920 --> 03:41:19.480
<v Shane Cahill (Meta)>We support a, uniform robust
federal standard in that regard.

3604
03:41:21.640 --> 03:41:26.325
<v Alice Friend (Google)>So Google also supports a federal framework
for AI. We think that federal framework

3605
03:41:26.325 --> 03:41:30.726
<v Alice Friend (Google)>should first and foremost concentrate
on frontier or the most advanced AI

3606
03:41:30.726 --> 03:41:31.766
<v Alice Friend (Google)>capabilities,

3607
03:41:32.405 --> 03:41:37.045
<v Alice Friend (Google)>primarily because they often touch on
national security, and that's a traditional

3608
03:41:37.045 --> 03:41:39.126
<v Alice Friend (Google)>jurisdiction of the federal government,

3609
03:41:39.525 --> 03:41:44.370
<v Alice Friend (Google)>to ensure that national security standard
is the same across all 50 states and our

3610
03:41:44.370 --> 03:41:45.251
<v Alice Friend (Google)>territories.

3611
03:41:45.811 --> 03:41:47.570
<v Alice Friend (Google)>But we also believe that,

3612
03:41:48.051 --> 03:41:51.091
<v Alice Friend (Google)>there's lots of jurisdiction
that is traditionally,

3613
03:41:51.490 --> 03:41:56.051
<v Alice Friend (Google)>at the state level where it is perfectly
appropriate for states to regulate on AI.

3614
03:41:56.385 --> 03:42:00.305
<v Alice Friend (Google)>And, again, we think a lot of existing
frameworks both at the state and federal

3615
03:42:00.305 --> 03:42:03.585
<v Alice Friend (Google)>level already apply to AI technologies,

3616
03:42:03.904 --> 03:42:07.745
<v Alice Friend (Google)>but we're working very closely with
states and with the federal government to

3617
03:42:07.745 --> 03:42:12.270
<v Alice Friend (Google)>examine places where we might need net
new regulation for a new technologies

3618
03:42:12.270 --> 03:42:13.710
<v Alice Friend (Google)>introduced by AI?

3619
03:42:16.350 --> 03:42:16.830
<v Logan Graham (Anthropic)>I,

3620
03:42:17.230 --> 03:42:20.750
<v Logan Graham (Anthropic)>first, I wanna say that we've
long supported the role for,

3621
03:42:21.470 --> 03:42:25.070
<v Logan Graham (Anthropic)>government to play here both at
the federal and the state level.

3622
03:42:25.555 --> 03:42:27.475
<v Logan Graham (Anthropic)>I wanna clarify that,

3623
03:42:27.954 --> 03:42:28.755
<v Logan Graham (Anthropic)>we were,

3624
03:42:29.154 --> 03:42:33.875
<v Logan Graham (Anthropic)>the only company on this panel to support
SB 53 in California before it became

3625
03:42:33.875 --> 03:42:38.261
<v Logan Graham (Anthropic)>signed into law, which my understanding
is was the basis of the RAISE Act. We

3626
03:42:38.261 --> 03:42:41.621
<v Logan Graham (Anthropic)>support the RAISE Act, and in fact,
we support a stronger legislation,

3627
03:42:42.101 --> 03:42:46.261
<v Logan Graham (Anthropic)>including in Massachusetts requiring
independent evaluation as well.

3628
03:42:46.740 --> 03:42:51.460
<v Logan Graham (Anthropic)>And we've also proposed a federal framework
that goes further. We think RAISE is the

3629
03:42:51.460 --> 03:42:56.555
<v Logan Graham (Anthropic)>floor that we need to raise it. We detailed
this in what we call our, advanced AI

3630
03:42:56.555 --> 03:42:57.596
<v Logan Graham (Anthropic)>framework as well.

3631
03:42:58.075 --> 03:42:59.995
<v CM Carmen De La Rosa>Thank you. Council member Hudson?

3632
03:43:00.795 --> 03:43:02.315
<v CM Crystal Hudson>Thank you so much, chair.

3633
03:43:03.115 --> 03:43:05.755
<v CM Crystal Hudson>We heard earlier from a
panel of whistleblowers

3634
03:43:05.755 --> 03:43:10.341
<v CM Crystal Hudson>who raised a number of flags, and one
issue raised was that of self improvement

3635
03:43:10.341 --> 03:43:11.860
<v CM Crystal Hudson>among AI agents.

3636
03:43:12.101 --> 03:43:16.181
<v CM Crystal Hudson>What is your company's core belief around
the role of self improving AI in your

3637
03:43:16.181 --> 03:43:17.461
<v CM Crystal Hudson>overall strategy?

3638
03:43:17.780 --> 03:43:22.505
<v CM Crystal Hudson>And to be more specific, is it core to your
AI strategy or a less significant part

3639
03:43:22.505 --> 03:43:23.385
<v CM Crystal Hudson>of the strategy?

3640
03:43:23.785 --> 03:43:27.705
<v CM Crystal Hudson>Could you continue your race to the top or
the bottom, I suppose depending on one's

3641
03:43:27.705 --> 03:43:28.425
<v CM Crystal Hudson>perspective,

3642
03:43:28.585 --> 03:43:34.345
<v CM Crystal Hudson>without self improving AI agents or with
significant guardrails on the way agents

3643
03:43:34.345 --> 03:43:35.545
<v CM Crystal Hudson>self improve?

3644
03:43:35.705 --> 03:43:37.064
<v CM Crystal Hudson>And my second question

3645
03:43:37.510 --> 03:43:43.109
<v CM Crystal Hudson>is when a new AI data center strains a local
grid or water supply, the costs can end

3646
03:43:43.109 --> 03:43:47.510
<v CM Crystal Hudson>up in everyone's utility bills while
the benefits go mostly to the company.

3647
03:43:47.670 --> 03:43:50.949
<v CM Crystal Hudson>Who should bear those costs? The
company that creates the demand,

3648
03:43:51.515 --> 03:43:55.915
<v CM Crystal Hudson>the utility that builds the capacity, or
the public that shares the system but

3649
03:43:55.915 --> 03:43:57.755
<v CM Crystal Hudson>played no role in,

3650
03:43:58.475 --> 03:43:59.435
<v CM Crystal Hudson>building it?

3651
03:43:59.755 --> 03:44:00.555
<v CM Crystal Hudson>Thank you.

3652
03:44:02.715 --> 03:44:04.500
<v Morgan Dwyer (OpenAI)>Thank you for your question.

3653
03:44:04.740 --> 03:44:08.181
<v Morgan Dwyer (OpenAI)>I'll start off by addressing
the question about RSI.

3654
03:44:08.900 --> 03:44:11.620
<v Morgan Dwyer (OpenAI)>To say that we believe that RSI

3655
03:44:11.620 --> 03:44:13.700
<v Morgan Dwyer (OpenAI)>may be one of the most consequential

3656
03:44:13.700 --> 03:44:16.580
<v Morgan Dwyer (OpenAI)>safety and security issues
of the coming decade,

3657
03:44:17.306 --> 03:44:19.865
<v Morgan Dwyer (OpenAI)>And that's why OpenAI has advocated

3658
03:44:19.865 --> 03:44:24.105
<v Morgan Dwyer (OpenAI)>for federal legislation
specifically focused on RSI,

3659
03:44:25.065 --> 03:44:28.825
<v Morgan Dwyer (OpenAI)>to include independent
of... Technical evaluators

3660
03:44:28.905 --> 03:44:34.400
<v Morgan Dwyer (OpenAI)>of internal lab practices as well as
the pace towards RSI development.

3661
03:44:34.801 --> 03:44:39.200
<v Morgan Dwyer (OpenAI)>We've made that work public, in our
Frontier Safety Blueprint. It's available

3662
03:44:39.200 --> 03:44:43.520
<v Morgan Dwyer (OpenAI)>online, and I'm happy to share
it with with you, as follow-up.

3663
03:44:44.160 --> 03:44:46.480
<v Morgan Dwyer (OpenAI)>On the case of electricity costs,

3664
03:44:47.045 --> 03:44:51.445
<v Morgan Dwyer (OpenAI)>look, OpenAI is committed to
paying our own way on energy,

3665
03:44:52.806 --> 03:44:57.206
<v Morgan Dwyer (OpenAI)>but we're also committed to
building with and for communities.

3666
03:44:57.846 --> 03:45:02.670
<v Morgan Dwyer (OpenAI)>All of our Stargate sites, which are our
infrastructure site... Sites, have what we

3667
03:45:02.670 --> 03:45:04.270
<v Morgan Dwyer (OpenAI)>call a community compact

3668
03:45:04.270 --> 03:45:06.351
<v Morgan Dwyer (OpenAI)>that's developed with the community,

3669
03:45:06.830 --> 03:45:09.630
<v Morgan Dwyer (OpenAI)>and addresses how we're
going to work together.

3670
03:45:10.030 --> 03:45:12.990
<v Morgan Dwyer (OpenAI)>For example, we recently
announced $80,000,000

3671
03:45:12.990 --> 03:45:14.990
<v Morgan Dwyer (OpenAI)>of community benefits investment,

3672
03:45:15.470 --> 03:45:16.990
<v Morgan Dwyer (OpenAI)>at our site in Georgia.

3673
03:45:18.495 --> 03:45:19.295
<v Morgan Dwyer (OpenAI)>Thank you.

3674
03:45:20.175 --> 03:45:21.215
<v Shane Cahill (Meta)>Council member,

3675
03:45:21.375 --> 03:45:24.815
<v Shane Cahill (Meta)>in relation to your first question
on recursive self improvement,

3676
03:45:25.215 --> 03:45:30.975
<v Shane Cahill (Meta)>if I may, I I I might just go back to a
point that I mentioned earlier in the day in

3677
03:45:30.975 --> 03:45:33.135
<v Shane Cahill (Meta)>relation to recursive self improvement

3678
03:45:33.909 --> 03:45:39.189
<v Shane Cahill (Meta)>that we believe that committing the
significant majority of compute towards serving

3679
03:45:39.189 --> 03:45:43.350
<v Shane Cahill (Meta)>people rather than racing towards
recursive self improvement

3680
03:45:43.510 --> 03:45:47.909
<v Shane Cahill (Meta)>is one of the best ways to ensure that
this technology is developed safely.

3681
03:45:49.396 --> 03:45:54.355
<v Shane Cahill (Meta)>In relation to your second question in
terms of electricity and and data centers,

3682
03:45:54.835 --> 03:45:57.476
<v Shane Cahill (Meta)>we believe that customers
should not bear costs,

3683
03:45:58.115 --> 03:46:04.010
<v Shane Cahill (Meta)>and we pay the full cost for electricity
used in our data centers. I also might just

3684
03:46:04.010 --> 03:46:08.810
<v Shane Cahill (Meta)>mention as well our futures for everyone
fund, which is a $1,000,000,000

3685
03:46:08.810 --> 03:46:12.649
<v Shane Cahill (Meta)>investment in relation to communities that
we partner with and work with with our

3686
03:46:12.649 --> 03:46:13.290
<v Shane Cahill (Meta)>data centers.

3687
03:46:15.765 --> 03:46:21.525
<v Alice Friend (Google)>So we also take a very prudent approach to
self learning inside our model training.

3688
03:46:22.085 --> 03:46:25.205
<v Alice Friend (Google)>We use AI models to help us evaluate,

3689
03:46:25.205 --> 03:46:26.245
<v Alice Friend (Google)>benchmark,

3690
03:46:26.245 --> 03:46:27.685
<v Alice Friend (Google)>and refine other models.

3691
03:46:28.280 --> 03:46:33.400
<v Alice Friend (Google)>But our use of this type of iterative
model development is confined to narrow

3692
03:46:33.480 --> 03:46:36.840
<v Alice Friend (Google)>supervised processes, that's
human supervised processes,

3693
03:46:37.240 --> 03:46:41.080
<v Alice Friend (Google)>to keep progress verifiable and
within defined safety limits.

3694
03:46:41.596 --> 03:46:45.836
<v Alice Friend (Google)>We also manage the risks of fully autonomous
model self improvement through our

3695
03:46:45.836 --> 03:46:47.596
<v Alice Friend (Google)>frontier safety framework.

3696
03:46:48.075 --> 03:46:50.956
<v Alice Friend (Google)>So we're very mindful about RSI,

3697
03:46:52.315 --> 03:46:54.010
<v Alice Friend (Google)>and are are

3698
03:46:54.010 --> 03:46:58.170
<v Alice Friend (Google)>quite controlled in the way that we
think about self learning internally.

3699
03:46:58.730 --> 03:47:04.010
<v Alice Friend (Google)>On on data centers, similar to the
other companies represented today,

3700
03:47:04.890 --> 03:47:06.730
<v Alice Friend (Google)>we believe in

3701
03:47:07.186 --> 03:47:11.426
<v Alice Friend (Google)>paying our own way and have pledged to
do so, particularly when it comes to

3702
03:47:12.226 --> 03:47:17.585
<v Alice Friend (Google)>the energy demands of data centers. We've
also invested in a wide range of energy

3703
03:47:17.585 --> 03:47:19.985
<v Alice Friend (Google)>sources to ensure that our data centers

3704
03:47:20.450 --> 03:47:25.010
<v Alice Friend (Google)>have an independent energy source that does
not add... That does not strain the grid

3705
03:47:25.010 --> 03:47:29.890
<v Alice Friend (Google)>and, in fact, adds back to the grid so
that, in the long term, it will benefit,

3706
03:47:30.130 --> 03:47:31.730
<v Alice Friend (Google)>wider energy customers.

3707
03:47:32.930 --> 03:47:33.410
<v CM Crystal Hudson>Thank you.

3708
03:47:34.755 --> 03:47:35.715
<v Logan Graham (Anthropic)>Likewise,

3709
03:47:35.875 --> 03:47:38.274
<v Logan Graham (Anthropic)>on recursive self improvement,

3710
03:47:38.274 --> 03:47:42.115
<v Logan Graham (Anthropic)>we take that extremely seriously
and have from the start.

3711
03:47:42.755 --> 03:47:47.450
<v Logan Graham (Anthropic)>I think, from my purview, that there are
three really important things here. The

3712
03:47:47.450 --> 03:47:51.130
<v Logan Graham (Anthropic)>first is just measurement and understanding.
I'm glad you raised this question

3713
03:47:51.130 --> 03:47:55.050
<v Logan Graham (Anthropic)>because we think this is something people
need to talk about. In late spring, we

3714
03:47:55.050 --> 03:47:59.610
<v Logan Graham (Anthropic)>published what I think is a first of its
kind assessment of how it is happening or

3715
03:47:59.610 --> 03:48:01.450
<v Logan Graham (Anthropic)>signs of it happening at anthropic,

3716
03:48:01.755 --> 03:48:04.155
<v Logan Graham (Anthropic)>happens in a lot of
ways, and it's nuanced.

3717
03:48:04.715 --> 03:48:09.114
<v Logan Graham (Anthropic)>In, late summer, we published a follow-up
that went into further detail of how we

3718
03:48:09.114 --> 03:48:11.515
<v Logan Graham (Anthropic)>should understand and
measure self improvement.

3719
03:48:11.675 --> 03:48:14.154
<v Logan Graham (Anthropic)>But that's not enough,
and that's why you saw

3720
03:48:14.770 --> 03:48:19.091
<v Logan Graham (Anthropic)>us and our CEO publicly call for pacing
the frontier because, candidly, the

3721
03:48:19.091 --> 03:48:23.650
<v Logan Graham (Anthropic)>technology is moving very fast, and we think
we would benefit from having more time.

3722
03:48:23.650 --> 03:48:25.331
<v Logan Graham (Anthropic)>And this is the second component.

3723
03:48:25.650 --> 03:48:29.170
<v Logan Graham (Anthropic)>And then, obviously, the third thing that
you need to do here is figure out what the

3724
03:48:29.170 --> 03:48:34.194
<v Logan Graham (Anthropic)>right guardrails are, but additionally,
what level of technology development we

3725
03:48:34.194 --> 03:48:36.995
<v Logan Graham (Anthropic)>should be going for as well as,

3726
03:48:37.715 --> 03:48:42.435
<v Logan Graham (Anthropic)>you know, what safety processes we need
along the way. And I think this is just

3727
03:48:42.435 --> 03:48:45.715
<v Logan Graham (Anthropic)>where we're at now, and I'm glad
that we're having this discussion.

3728
03:48:46.600 --> 03:48:49.880
<v Logan Graham (Anthropic)>On data centers, this is outside
of my expertise. My understanding

3729
03:48:50.040 --> 03:48:54.280
<v Logan Graham (Anthropic)>is that we pay our own way, and we're
committed to contributing, working with the

3730
03:48:54.280 --> 03:48:55.080
<v Logan Graham (Anthropic)>communities,

3731
03:48:55.320 --> 03:48:56.200
<v Logan Graham (Anthropic)>around them.

3732
03:48:56.440 --> 03:48:59.800
<v Logan Graham (Anthropic)>But I'll have to, defer to my
excellent colleagues who work on it.

3733
03:49:01.795 --> 03:49:06.436
<v CM Crystal Hudson>Thank you all, and thank you, chair.
I do wanna just make my one little

3734
03:49:07.795 --> 03:49:09.955
<v CM Crystal Hudson>note that it seems like,

3735
03:49:10.275 --> 03:49:12.596
<v CM Crystal Hudson>the self improving aspect

3736
03:49:12.755 --> 03:49:14.275
<v CM Crystal Hudson>is absolutely core

3737
03:49:14.851 --> 03:49:19.250
<v CM Crystal Hudson>to everyone's AI strategy. And
I... That's certainly concerning

3738
03:49:19.330 --> 03:49:21.811
<v CM Crystal Hudson>given the testimony we
heard earlier this morning.

3739
03:49:22.690 --> 03:49:25.570
<v CM Crystal Hudson>So thank you, and I look
forward to any follow-up.

3740
03:49:26.210 --> 03:49:30.825
<v CM Carmen De La Rosa>Thank you so much. Councilmember Schulman,
followed by Brooks Powers, followed by

3741
03:49:30.825 --> 03:49:31.625
<v CM Carmen De La Rosa>Brewer.

3742
03:49:31.945 --> 03:49:33.465
<v CM Lynn Schulman>Thank you. Good afternoon.

3743
03:49:33.705 --> 03:49:38.024
<v CM Lynn Schulman>I am councilmember Lynn Schulman. I'm
chair of the New York City Council Health

3744
03:49:38.024 --> 03:49:38.745
<v CM Lynn Schulman>Committee.

3745
03:49:38.905 --> 03:49:43.990
<v CM Lynn Schulman>As a breast cancer survivor, I am looking
forward to the day AI can help find cures

3746
03:49:43.990 --> 03:49:45.591
<v CM Lynn Schulman>for that and other diseases.

3747
03:49:45.830 --> 03:49:48.950
<v CM Lynn Schulman>That said, there is a dark
side to AI intervention

3748
03:49:49.030 --> 03:49:50.150
<v CM Lynn Schulman>in health care.

3749
03:49:50.710 --> 03:49:54.870
<v CM Lynn Schulman>Most viruses that are found in
nature can be contained fairly

3750
03:49:54.950 --> 03:49:55.351
<v CM Lynn Schulman>quickly,

3751
03:49:56.105 --> 03:49:59.065
<v CM Lynn Schulman>but an artificial pathogen
has no constraints.

3752
03:49:59.386 --> 03:50:04.346
<v CM Lynn Schulman>In August, a team from Stanford used an
AI model to design an entirely new virus

3753
03:50:04.346 --> 03:50:05.226
<v CM Lynn Schulman>from scratch.

3754
03:50:05.466 --> 03:50:09.466
<v CM Lynn Schulman>In addition, Anthropic has its
own AI powered biology lab.

3755
03:50:10.010 --> 03:50:14.090
<v CM Lynn Schulman>What measures are being put into place such
as placing... I'm not I'm not gonna say

3756
03:50:14.090 --> 03:50:16.489
<v CM Lynn Schulman>guardrails. I had that
originally safeguards

3757
03:50:16.489 --> 03:50:21.850
<v CM Lynn Schulman>on DNA synthesis to prevent rogue viruses
from being created and replicated at will

3758
03:50:21.850 --> 03:50:23.050
<v CM Lynn Schulman>to jeopardize the public.

3759
03:50:27.136 --> 03:50:29.056
<v Morgan Dwyer (OpenAI)>Thank you for your question.

3760
03:50:29.455 --> 03:50:31.936
<v Morgan Dwyer (OpenAI)>The question about how AI,

3761
03:50:32.176 --> 03:50:33.535
<v Morgan Dwyer (OpenAI)>could enhance

3762
03:50:33.535 --> 03:50:34.096
<v Morgan Dwyer (OpenAI)>or

3763
03:50:34.415 --> 03:50:35.695
<v Morgan Dwyer (OpenAI)>make

3764
03:50:35.695 --> 03:50:40.415
<v Morgan Dwyer (OpenAI)>worse biological risks is central
to OpenAI's preparedness framework.

3765
03:50:40.990 --> 03:50:45.470
<v Morgan Dwyer (OpenAI)>That's the framework that we use
to guide our decisions about,

3766
03:50:45.870 --> 03:50:50.590
<v Morgan Dwyer (OpenAI)>model risk and whether, whether we
have the right safeguards in place.

3767
03:50:51.390 --> 03:50:53.149
<v Morgan Dwyer (OpenAI)>So we evaluate

3768
03:50:53.149 --> 03:50:54.350
<v Morgan Dwyer (OpenAI)>for biological

3769
03:50:54.350 --> 03:50:55.275
<v Morgan Dwyer (OpenAI)>risks

3770
03:50:55.275 --> 03:50:59.436
<v Morgan Dwyer (OpenAI)>as part of that framework. Again,
we use third party evaluators

3771
03:50:59.436 --> 03:51:00.716
<v Morgan Dwyer (OpenAI)>and red teamers.

3772
03:51:01.195 --> 03:51:03.195
<v Morgan Dwyer (OpenAI)>We also work with the federal government,

3773
03:51:03.355 --> 03:51:06.235
<v Morgan Dwyer (OpenAI)>as it pertains to national security risks,

3774
03:51:06.715 --> 03:51:10.711
<v Morgan Dwyer (OpenAI)>and we are transparent on the
results of those evaluations.

3775
03:51:10.870 --> 03:51:13.591
<v Morgan Dwyer (OpenAI)>So with the launch of every major model,

3776
03:51:14.150 --> 03:51:18.870
<v Morgan Dwyer (OpenAI)>we release what's called our system
card, and that card details,

3777
03:51:19.190 --> 03:51:21.431
<v Morgan Dwyer (OpenAI)>the evaluations that we performed,

3778
03:51:21.590 --> 03:51:24.795
<v Morgan Dwyer (OpenAI)>as well as the results.
And and those evaluations,

3779
03:51:25.035 --> 03:51:26.396
<v Morgan Dwyer (OpenAI)>as you mentioned, do,

3780
03:51:26.715 --> 03:51:28.476
<v Morgan Dwyer (OpenAI)>include biological risks.

3781
03:51:30.315 --> 03:51:32.876
<v Shane Cahill (Meta)>Council member, this
is, such an important,

3782
03:51:33.755 --> 03:51:39.470
<v Shane Cahill (Meta)>area and question, so thank you for raising
it. Biological risks is one of the core

3783
03:51:39.470 --> 03:51:41.870
<v Shane Cahill (Meta)>areas, one of the key risk domains

3784
03:51:41.870 --> 03:51:44.110
<v Shane Cahill (Meta)>that's part of our scaling framework

3785
03:51:44.270 --> 03:51:45.230
<v Shane Cahill (Meta)>at Meta.

3786
03:51:45.950 --> 03:51:46.590
<v Shane Cahill (Meta)>We

3787
03:51:46.830 --> 03:51:52.030
<v Shane Cahill (Meta)>have, as I mentioned earlier, have a
multilayered approach to risk assessment

3788
03:51:52.285 --> 03:51:54.285
<v Shane Cahill (Meta)>across training, evaluation,

3789
03:51:54.365 --> 03:51:55.245
<v Shane Cahill (Meta)>and release.

3790
03:51:55.405 --> 03:51:57.965
<v Shane Cahill (Meta)>And that includes adversarial testing

3791
03:51:58.365 --> 03:52:00.686
<v Shane Cahill (Meta)>with external experts.

3792
03:52:00.925 --> 03:52:02.525
<v Shane Cahill (Meta)>We work with governments,

3793
03:52:02.925 --> 03:52:05.485
<v Shane Cahill (Meta)>other stakeholders, and
experts in relation to that.

3794
03:52:06.260 --> 03:52:07.860
<v Shane Cahill (Meta)>And as I mentioned earlier,

3795
03:52:08.260 --> 03:52:14.020
<v Shane Cahill (Meta)>as part of our commitment to partnering
with independent third party evaluators and

3796
03:52:14.020 --> 03:52:14.900
<v Shane Cahill (Meta)>assessors also.

3797
03:52:17.875 --> 03:52:20.915
<v Alice Friend (Google)>Yes, ma'am. Similar to
our colleagues, we also

3798
03:52:21.314 --> 03:52:24.515
<v Alice Friend (Google)>have bio or biological capabilities

3799
03:52:24.515 --> 03:52:27.235
<v Alice Friend (Google)>as part of our frontier safety framework.

3800
03:52:27.875 --> 03:52:32.675
<v Alice Friend (Google)>As we look at our critical capability levels
that I referred to earlier or the the

3801
03:52:32.811 --> 03:52:33.690
<v Alice Friend (Google)>levels

3802
03:52:33.690 --> 03:52:34.570
<v Alice Friend (Google)>at which

3803
03:52:35.211 --> 03:52:38.411
<v Alice Friend (Google)>we would trigger additional mitigations

3804
03:52:39.051 --> 03:52:43.051
<v Alice Friend (Google)>in our models. We look for what's
called uplift. That is to say,

3805
03:52:43.610 --> 03:52:45.531
<v Alice Friend (Google)>the model will be able to assist

3806
03:52:45.690 --> 03:52:47.051
<v Alice Friend (Google)>users in

3807
03:52:49.405 --> 03:52:53.005
<v Alice Friend (Google)>creating information or knowledge
that's not otherwise available

3808
03:52:53.485 --> 03:52:55.726
<v Alice Friend (Google)>through widely accessible sources.

3809
03:52:57.165 --> 03:52:59.245
<v Alice Friend (Google)>And we think very carefully about

3810
03:52:59.885 --> 03:53:02.110
<v Alice Friend (Google)>how to measure for bioCCLs

3811
03:53:02.110 --> 03:53:07.710
<v Alice Friend (Google)>and are constantly upgrading and and
updating our our ability to do that. When it

3812
03:53:07.710 --> 03:53:11.471
<v Alice Friend (Google)>comes to those mitigations or those
safeguards, as you so aptly put it,

3813
03:53:11.870 --> 03:53:16.030
<v Alice Friend (Google)>we implement those at both the
model and the application levels

3814
03:53:16.476 --> 03:53:18.796
<v Alice Friend (Google)>to ensure that

3815
03:53:19.915 --> 03:53:21.836
<v Alice Friend (Google)>casual users cannot access

3816
03:53:22.396 --> 03:53:24.235
<v Alice Friend (Google)>critical bio information.

3817
03:53:24.476 --> 03:53:25.916
<v Alice Friend (Google)>It is very important

3818
03:53:26.155 --> 03:53:27.355
<v Alice Friend (Google)>to consider

3819
03:53:28.730 --> 03:53:31.530
<v Alice Friend (Google)>the dual use nature of biological

3820
03:53:31.610 --> 03:53:32.730
<v Alice Friend (Google)>capabilities,

3821
03:53:32.730 --> 03:53:37.130
<v Alice Friend (Google)>of course, because these are the
kinds of capabilities that scientists,

3822
03:53:37.290 --> 03:53:38.010
<v Alice Friend (Google)>particularly

3823
03:53:38.330 --> 03:53:39.850
<v Alice Friend (Google)>medical investigators,

3824
03:53:39.850 --> 03:53:42.090
<v Alice Friend (Google)>need in order to make advances

3825
03:53:42.170 --> 03:53:47.905
<v Alice Friend (Google)>in our own health care as well. And so we
think very carefully about how to release

3826
03:53:48.225 --> 03:53:49.185
<v Alice Friend (Google)>models

3827
03:53:49.744 --> 03:53:53.265
<v Alice Friend (Google)>to scientists that can use
the technology responsibly.

3828
03:53:53.824 --> 03:53:56.865
<v Alice Friend (Google)>We did that for the first time
with our AlphaFold model,

3829
03:53:57.185 --> 03:54:02.240
<v Alice Friend (Google)>which, of course, is how millions of
investigators around the world are able to,

3830
03:54:04.000 --> 03:54:07.840
<v Alice Friend (Google)>look at, the folding of proteins
and to use that research,

3831
03:54:08.480 --> 03:54:10.399
<v Alice Friend (Google)>to find cures for diseases.

3832
03:54:10.896 --> 03:54:14.655
<v Alice Friend (Google)>So, there's a real careful
approach that we take to ensure,

3833
03:54:14.976 --> 03:54:19.775
<v Alice Friend (Google)>that science can continue, but that dangerous
uses of biological knowledge cannot.

3834
03:54:22.255 --> 03:54:22.976
<v Logan Graham (Anthropic)>Likewise,

3835
03:54:23.056 --> 03:54:25.360
<v Logan Graham (Anthropic)>we take this extremely seriously.

3836
03:54:25.440 --> 03:54:28.400
<v Logan Graham (Anthropic)>In fact, if I'm not mistaken,
it was my team and myself

3837
03:54:28.640 --> 03:54:32.960
<v Logan Graham (Anthropic)>that developed the first call for action
on these risks, the first evaluations,

3838
03:54:33.040 --> 03:54:37.676
<v Logan Graham (Anthropic)>the first tests, the first industry government
partnerships on biological risks and

3839
03:54:37.676 --> 03:54:38.556
<v Logan Graham (Anthropic)>evaluating

3840
03:54:38.556 --> 03:54:40.396
<v Logan Graham (Anthropic)>models and their safeguards,

3841
03:54:41.275 --> 03:54:43.036
<v Logan Graham (Anthropic)>several years ago at this point.

3842
03:54:43.835 --> 03:54:48.636
<v Logan Graham (Anthropic)>Not only that, I personally think we need
to go, further here. You mentioned DNA

3843
03:54:48.636 --> 03:54:52.380
<v Logan Graham (Anthropic)>synthesis screening. I think this
is personally a no nonsense,

3844
03:54:52.700 --> 03:54:53.341
<v Logan Graham (Anthropic)>approach.

3845
03:54:53.740 --> 03:54:56.141
<v Logan Graham (Anthropic)>There's much else we can do to,

3846
03:54:56.540 --> 03:54:58.461
<v Logan Graham (Anthropic)>to work on biological defense.

3847
03:54:59.900 --> 03:55:05.015
<v Logan Graham (Anthropic)>And I will... You know, alongside the other
companies, we maintain the same and, in

3848
03:55:05.015 --> 03:55:08.455
<v Logan Graham (Anthropic)>fact, try to compete for the
best safety practices here.

3849
03:55:08.935 --> 03:55:12.455
<v Logan Graham (Anthropic)>And you can see some of what we found, I
think, disclosed for the first time ever in

3850
03:55:12.455 --> 03:55:15.895
<v Logan Graham (Anthropic)>a recent threat intelligence report
where we have observed actors

3851
03:55:16.210 --> 03:55:18.609
<v Logan Graham (Anthropic)>attempting to do biologically

3852
03:55:18.609 --> 03:55:23.890
<v Logan Graham (Anthropic)>risky activities with models. The threat
is, I think, clear to us now. We think we

3853
03:55:23.890 --> 03:55:27.970
<v Logan Graham (Anthropic)>need urgent action. But I will close by
saying that, you know, as somebody that grew

3854
03:55:27.970 --> 03:55:30.370
<v Logan Graham (Anthropic)>up with a number of chronic diseases

3855
03:55:30.370 --> 03:55:33.865
<v Logan Graham (Anthropic)>going blind and couldn't
walk, I very intimately feel

3856
03:55:34.025 --> 03:55:37.626
<v Logan Graham (Anthropic)>that the upside of the same
capabilities is enormous.

3857
03:55:37.945 --> 03:55:41.545
<v Logan Graham (Anthropic)>But increasingly in my work, I feel
optimistic that we can defend against the

3858
03:55:41.545 --> 03:55:42.826
<v Logan Graham (Anthropic)>downside sufficiently.

3859
03:55:43.545 --> 03:55:49.020
<v CM Lynn Schulman>The... If I could just say just one...
In closing, the upside is is very,

3860
03:55:49.261 --> 03:55:50.620
<v CM Lynn Schulman>has a great potential.

3861
03:55:50.940 --> 03:55:52.061
<v CM Lynn Schulman>The downside

3862
03:55:52.301 --> 03:55:55.261
<v CM Lynn Schulman>is extraordinarily
detrimental to humanity,

3863
03:55:55.660 --> 03:55:58.780
<v CM Lynn Schulman>but you should consider about
screening orders for synthetic

3864
03:55:59.020 --> 03:56:03.705
<v CM Lynn Schulman>nucleic acids, is DNA, and the equipment
that's needed to make them because right

3865
03:56:03.705 --> 03:56:04.745
<v CM Lynn Schulman>now AI

3866
03:56:04.985 --> 03:56:08.105
<v CM Lynn Schulman>is now outperforming
PhD level virologists.

3867
03:56:08.105 --> 03:56:08.585
<v CM Lynn Schulman>So

3868
03:56:08.985 --> 03:56:13.545
<v CM Lynn Schulman>a... An AI can take this and go somewhere
else with it. So that's something that's

3869
03:56:13.545 --> 03:56:17.430
<v CM Lynn Schulman>really important to have those safeguards.
Thank you. Thank you. Council member

3870
03:56:17.430 --> 03:56:21.350
<v CM Lynn Schulman>Brooks Powers followed by
Brewer followed by by Narcisse.

3871
03:56:24.869 --> 03:56:28.710
<v CM Selvena Brooks-Powers>Thank you, chair, and thank you, speaker.
I'm gonna just run through my questions.

3872
03:56:28.710 --> 03:56:32.386
<v CM Selvena Brooks-Powers>And if I have to repeat anything,
just let me know. First,

3873
03:56:33.825 --> 03:56:38.226
<v CM Selvena Brooks-Powers>before the summit, each of your companies
told the public that its frontier models

3874
03:56:38.226 --> 03:56:39.985
<v CM Selvena Brooks-Powers>were tested and secure.

3875
03:56:40.466 --> 03:56:45.920
<v CM Selvena Brooks-Powers>Since July, all four of your organizations
have admitted that your AI agents went

3876
03:56:45.920 --> 03:56:51.280
<v CM Selvena Brooks-Powers>rogue and nearly a dozen incidents have
been reported across all of your companies.

3877
03:56:51.520 --> 03:56:56.480
<v CM Selvena Brooks-Powers>Can you tell New Yorkers today that none
of your agents have... Has access or

3878
03:56:56.480 --> 03:56:57.280
<v CM Selvena Brooks-Powers>attempted

3879
03:56:57.439 --> 03:57:00.480
<v CM Selvena Brooks-Powers>to access a New York
City government system

3880
03:57:01.005 --> 03:57:04.365
<v CM Selvena Brooks-Powers>or a New York City
resident's personal data.

3881
03:57:05.165 --> 03:57:11.085
<v CM Selvena Brooks-Powers>Next, Southeast Queens has long been a
target cloning a deep fake... Cloning and

3882
03:57:11.085 --> 03:57:12.445
<v CM Selvena Brooks-Powers>deep fakes make it easier

3883
03:57:13.260 --> 03:57:17.500
<v CM Selvena Brooks-Powers>than ever to impersonate a family
member, a bank, or a city agency.

3884
03:57:17.899 --> 03:57:23.260
<v CM Selvena Brooks-Powers>What specifically stops your tools from
cloning a real person's voice or forging a

3885
03:57:23.260 --> 03:57:23.739
<v CM Selvena Brooks-Powers>document?

3886
03:57:25.415 --> 03:57:26.854
<v CM Selvena Brooks-Powers>What protocols

3887
03:57:26.854 --> 03:57:32.615
<v CM Selvena Brooks-Powers>are the leaders of AI development putting
into place to make sure AI does not take

3888
03:57:32.615 --> 03:57:38.854
<v CM Selvena Brooks-Powers>information from hate sites and disseminate
it to further racist and antisemitic

3889
03:57:38.854 --> 03:57:40.135
<v CM Selvena Brooks-Powers>or Islamophobic

3890
03:57:40.135 --> 03:57:40.854
<v CM Selvena Brooks-Powers>discourse?

3891
03:57:45.590 --> 03:57:46.790
<v CM Selvena Brooks-Powers>That was all. Sorry.

3892
03:57:47.750 --> 03:57:48.710
<v Morgan Dwyer (OpenAI)>Thank

3893
03:57:48.710 --> 03:57:50.391
<v Morgan Dwyer (OpenAI)>you for your question.

3894
03:57:51.030 --> 03:57:55.590
<v Morgan Dwyer (OpenAI)>So I'll start off by saying I am
not aware of any incidents affecting

3895
03:57:55.590 --> 03:57:56.550
<v Morgan Dwyer (OpenAI)>New York City,

3896
03:57:57.725 --> 03:58:00.285
<v Morgan Dwyer (OpenAI)>but we have an ongoing investigation

3897
03:58:00.285 --> 03:58:04.085
<v Morgan Dwyer (OpenAI)>into potential incidents
going back to November

3898
03:58:04.085 --> 03:58:05.085
<v Morgan Dwyer (OpenAI)>2025.

3899
03:58:05.645 --> 03:58:07.485
<v Morgan Dwyer (OpenAI)>And as part of that investigation,

3900
03:58:07.725 --> 03:58:10.765
<v Morgan Dwyer (OpenAI)>we are notifying any
affected third parties

3901
03:58:11.240 --> 03:58:14.360
<v Morgan Dwyer (OpenAI)>and making the results of
our investigation public.

3902
03:58:14.760 --> 03:58:16.680
<v Morgan Dwyer (OpenAI)>So I can commit to you that if,

3903
03:58:17.080 --> 03:58:22.280
<v Morgan Dwyer (OpenAI)>if there are incidents that affect New
York City, we will make the third parties

3904
03:58:22.280 --> 03:58:22.840
<v Morgan Dwyer (OpenAI)>aware.

3905
03:58:24.025 --> 03:58:25.386
<v Morgan Dwyer (OpenAI)>On your question,

3906
03:58:25.705 --> 03:58:27.386
<v Morgan Dwyer (OpenAI)>about impersonation,

3907
03:58:27.945 --> 03:58:28.905
<v Morgan Dwyer (OpenAI)>OpenAI,

3908
03:58:29.305 --> 03:58:32.905
<v Morgan Dwyer (OpenAI)>our usage policies prohibit
the use of our tools,

3909
03:58:33.145 --> 03:58:34.425
<v Morgan Dwyer (OpenAI)>for impersonation,

3910
03:58:34.425 --> 03:58:35.545
<v Morgan Dwyer (OpenAI)>fraud, and scams,

3911
03:58:36.301 --> 03:58:38.621
<v Morgan Dwyer (OpenAI)>and we enforce on those policies.

3912
03:58:38.860 --> 03:58:40.700
<v Morgan Dwyer (OpenAI)>And I would say more broadly,

3913
03:58:40.940 --> 03:58:43.740
<v Morgan Dwyer (OpenAI)>the approach that we take
to safety is holistic.

3914
03:58:44.061 --> 03:58:44.940
<v Morgan Dwyer (OpenAI)>That involves,

3915
03:58:45.181 --> 03:58:47.340
<v Morgan Dwyer (OpenAI)>filtering unsafe training data,

3916
03:58:48.075 --> 03:58:48.955
<v Morgan Dwyer (OpenAI)>performing,

3917
03:58:49.195 --> 03:58:50.155
<v Morgan Dwyer (OpenAI)>evaluations

3918
03:58:50.155 --> 03:58:51.835
<v Morgan Dwyer (OpenAI)>that are very detailed,

3919
03:58:52.475 --> 03:58:55.035
<v Morgan Dwyer (OpenAI)>working with third parties
on those evaluations,

3920
03:58:56.075 --> 03:59:00.235
<v Morgan Dwyer (OpenAI)>ahead or alongside our launches. As I
mentioned, we make the results of our

3921
03:59:00.235 --> 03:59:01.675
<v Morgan Dwyer (OpenAI)>evaluations public

3922
03:59:02.230 --> 03:59:03.910
<v Morgan Dwyer (OpenAI)>and summarize,

3923
03:59:04.070 --> 03:59:08.230
<v Morgan Dwyer (OpenAI)>the work that we did and any
residual risks in our system card.

3924
03:59:08.710 --> 03:59:09.990
<v Morgan Dwyer (OpenAI)>And then we monitor,

3925
03:59:10.230 --> 03:59:13.830
<v Morgan Dwyer (OpenAI)>for failures of our safeguards
so that we can continuously

3926
03:59:13.830 --> 03:59:14.870
<v Morgan Dwyer (OpenAI)>improve safety.

3927
03:59:17.825 --> 03:59:21.585
<v Shane Cahill (Meta)>Council member, in relation to your
first question, I'm not aware.

3928
03:59:22.465 --> 03:59:24.945
<v Shane Cahill (Meta)>In relation to your second question,

3929
03:59:25.345 --> 03:59:29.840
<v Shane Cahill (Meta)>we have robust and comprehensive policies
in place as well as well as t... As

3930
03:59:29.840 --> 03:59:33.200
<v Shane Cahill (Meta)>well... Excuse me. As
well as teams of experts

3931
03:59:33.680 --> 03:59:35.601
<v Shane Cahill (Meta)>who work on frauds and scams.

3932
03:59:36.319 --> 03:59:36.960
<v Shane Cahill (Meta)>And

3933
03:59:37.120 --> 03:59:41.680
<v Shane Cahill (Meta)>I'd like to follow-up with you with
those experts to discuss that one more in

3934
03:59:41.680 --> 03:59:41.920
<v Shane Cahill (Meta)>detail.

3935
03:59:42.585 --> 03:59:45.944
<v Shane Cahill (Meta)>In relation to hate content, I... First
of all, I just want to acknowledge how

3936
03:59:45.944 --> 03:59:48.985
<v Shane Cahill (Meta)>upsetting such content is in terms of

3937
03:59:49.465 --> 03:59:51.305
<v Shane Cahill (Meta)>people who are experiencing it.

3938
03:59:52.745 --> 03:59:58.631
<v Shane Cahill (Meta)>We are... Our goal at at Beta is for
our AI offerings to be responsive,

3939
03:59:58.870 --> 04:00:01.030
<v Shane Cahill (Meta)>accurate, balanced, and unbiased.

3940
04:00:03.030 --> 04:00:07.990
<v CM Selvena Brooks-Powers>I'm sorry. Was that answered just
now to the last question about the

3941
04:00:09.835 --> 04:00:14.875
<v CM Selvena Brooks-Powers>not taking the information to create the
negative discourse, the racial discourse?

3942
04:00:15.436 --> 04:00:18.635
<v Shane Cahill (Meta)>Yes. That's correct, council member. I
was speaking to your third question in

3943
04:00:18.635 --> 04:00:21.595
<v Shane Cahill (Meta)>relation to, to that type of content and,

3944
04:00:21.915 --> 04:00:24.075
<v Shane Cahill (Meta)>our objective in terms of

3945
04:00:24.870 --> 04:00:28.391
<v Shane Cahill (Meta)>ensuring that our AI
offerings are responsive,

3946
04:00:28.391 --> 04:00:30.951
<v Shane Cahill (Meta)>accurate, balanced, and unbiased.

3947
04:00:31.110 --> 04:00:34.871
<v CM Selvena Brooks-Powers>So have you enhanced what
you're doing considering that,

3948
04:00:36.150 --> 04:00:38.790
<v CM Selvena Brooks-Powers>you know, it has been
something that has been

3949
04:00:39.165 --> 04:00:43.405
<v CM Selvena Brooks-Powers>found on a number of different platforms.
Have you all been, like, staying up to

3950
04:00:43.405 --> 04:00:44.686
<v CM Selvena Brooks-Powers>course with

3951
04:00:45.245 --> 04:00:47.966
<v CM Selvena Brooks-Powers>enhancing whatever those processes are?

3952
04:00:49.085 --> 04:00:51.405
<v Shane Cahill (Meta)>Yes, council member. Thank
you for that question.

3953
04:00:52.020 --> 04:00:56.581
<v Shane Cahill (Meta)>And if if it's okay, I will
have our content teams,

3954
04:00:56.900 --> 04:01:01.221
<v Shane Cahill (Meta)>and specialists connect with you afterwards
so that I can provide you with precise

3955
04:01:01.221 --> 04:01:02.820
<v Shane Cahill (Meta)>information in relation to that.

3956
04:01:04.740 --> 04:01:05.300
<v CM Selvena Brooks-Powers>Thank you.

3957
04:01:06.635 --> 04:01:09.835
<v Alice Friend (Google)>Councilmember, thank you for
these questions. I'm also not,

3958
04:01:10.154 --> 04:01:11.355
<v Alice Friend (Google)>personally aware

3959
04:01:11.515 --> 04:01:14.154
<v Alice Friend (Google)>of any, New York City resident,

3960
04:01:14.314 --> 04:01:17.115
<v Alice Friend (Google)>being impacted by, the three incidents

3961
04:01:17.274 --> 04:01:19.195
<v Alice Friend (Google)>that I mentioned earlier.

3962
04:01:19.755 --> 04:01:21.114
<v Alice Friend (Google)>I do wanna note that,

3963
04:01:21.595 --> 04:01:24.660
<v Alice Friend (Google)>our case, in all three incidents,

3964
04:01:25.460 --> 04:01:29.860
<v Alice Friend (Google)>the model stopped itself once it realized
it had left the test environment.

3965
04:01:30.660 --> 04:01:35.220
<v Alice Friend (Google)>We tend to think of this less as a misalignment
event and more of a a mistake event.

3966
04:01:35.755 --> 04:01:41.354
<v Alice Friend (Google)>One of the websites, for example, that was
accessed had the same name as the, fake

3967
04:01:41.354 --> 04:01:43.835
<v Alice Friend (Google)>website being used in
the test environment.

3968
04:01:43.995 --> 04:01:48.475
<v Alice Friend (Google)>And once the the agent realized that it was
in fact a real company, the agent ceased

3969
04:01:48.475 --> 04:01:49.035
<v Alice Friend (Google)>its activity.

3970
04:01:49.730 --> 04:01:50.530
<v Alice Friend (Google)>We

3971
04:01:50.851 --> 04:01:54.210
<v Alice Friend (Google)>have a number of tools,
across our our platforms,

3972
04:01:55.250 --> 04:01:57.570
<v Alice Friend (Google)>to manage, likeness and identity,

3973
04:01:58.851 --> 04:02:02.931
<v Alice Friend (Google)>and we... We've also supported a range
of different, regulatory proposals

3974
04:02:03.235 --> 04:02:07.635
<v Alice Friend (Google)>regarding this topic. We're... Proudly
support the no fakes act, which would protect

3975
04:02:07.635 --> 04:02:09.636
<v Alice Friend (Google)>an individual's digital replica,

3976
04:02:10.115 --> 04:02:12.115
<v Alice Friend (Google)>and give people the right to control

3977
04:02:12.516 --> 04:02:16.035
<v Alice Friend (Google)>the use of their own voice and
likeness in AI generated works.

3978
04:02:16.811 --> 04:02:22.330
<v Alice Friend (Google)>We also have a a tool called Likeness ID on
YouTube that helps creators identify and

3979
04:02:22.330 --> 04:02:26.250
<v Alice Friend (Google)>manage AI generated content that uses
their face or voice without permission.

3980
04:02:26.811 --> 04:02:30.330
<v Alice Friend (Google)>So so we invest in a variety of different
ways to address this critical issue.

3981
04:02:31.665 --> 04:02:34.385
<v Alice Friend (Google)>On the information

3982
04:02:34.385 --> 04:02:35.186
<v Alice Friend (Google)>and and,

3983
04:02:37.105 --> 04:02:39.346
<v Alice Friend (Google)>getting through into our tools about,

3984
04:02:39.665 --> 04:02:41.025
<v Alice Friend (Google)>hate and bias,

3985
04:02:41.985 --> 04:02:44.146
<v Alice Friend (Google)>we have, again, a layered approach,

3986
04:02:44.580 --> 04:02:49.300
<v Alice Friend (Google)>as I've been saying, to this type
of issue as well. Everything from,

3987
04:02:49.620 --> 04:02:53.700
<v Alice Friend (Google)>the model training process itself
in both pre and post training,

3988
04:02:54.100 --> 04:02:55.220
<v Alice Friend (Google)>all the way through,

3989
04:02:55.780 --> 04:02:56.899
<v Alice Friend (Google)>guardrails

3990
04:02:56.899 --> 04:02:58.340
<v Alice Friend (Google)>on our application layer,

3991
04:02:58.795 --> 04:03:00.395
<v Alice Friend (Google)>all of which produce,

3992
04:03:00.715 --> 04:03:04.875
<v Alice Friend (Google)>what we're proud to say is, the
least biased model in the industry.

3993
04:03:05.274 --> 04:03:06.955
<v Alice Friend (Google)>Independent investigations,

3994
04:03:07.194 --> 04:03:08.155
<v Alice Friend (Google)>have shown,

3995
04:03:08.795 --> 04:03:10.395
<v Alice Friend (Google)>that our models show,

3996
04:03:10.715 --> 04:03:13.351
<v Alice Friend (Google)>the least bias, for example,

3997
04:03:13.830 --> 04:03:15.030
<v Alice Friend (Google)>with regards to,

3998
04:03:16.470 --> 04:03:17.671
<v Alice Friend (Google)>partisan bias.

3999
04:03:18.150 --> 04:03:21.591
<v Alice Friend (Google)>But we take these issues extremely
seriously. We wanna create,

4000
04:03:22.070 --> 04:03:23.270
<v Alice Friend (Google)>a fair and neutral

4001
04:03:24.645 --> 04:03:30.085
<v Alice Friend (Google)>set of information for people to access,
and we want, our responses generated by our

4002
04:03:30.085 --> 04:03:31.125
<v Alice Friend (Google)>AI tools,

4003
04:03:31.685 --> 04:03:32.805
<v Alice Friend (Google)>to represent

4004
04:03:32.885 --> 04:03:36.005
<v Alice Friend (Google)>a broad array of
experiences and perspectives

4005
04:03:36.210 --> 04:03:37.570
<v Alice Friend (Google)>without bias,

4006
04:03:37.810 --> 04:03:41.250
<v Alice Friend (Google)>towards or against any
particular identity or ideology.

4007
04:03:44.530 --> 04:03:46.450
<v Logan Graham (Anthropic)>And while in my role,

4008
04:03:46.850 --> 04:03:50.930
<v Logan Graham (Anthropic)>I I don't cover... I haven't covered
the... Any effect on, New York City,

4009
04:03:51.535 --> 04:03:53.936
<v Logan Graham (Anthropic)>I can say that we work
with in investigating,

4010
04:03:54.255 --> 04:03:57.455
<v Logan Graham (Anthropic)>incidents like this. We work with the right
parties at the right time to disclose to

4011
04:03:57.455 --> 04:04:00.015
<v Logan Graham (Anthropic)>affected parties, the right
level of law enforcement,

4012
04:04:00.575 --> 04:04:04.175
<v Logan Graham (Anthropic)>government at the right level as urgently
as possible needed to remediate,

4013
04:04:04.335 --> 04:04:07.890
<v Logan Graham (Anthropic)>the effect, and we commit to, you know,
making sure that we follow-up with the right

4014
04:04:07.890 --> 04:04:09.011
<v Logan Graham (Anthropic)>parties to do so.

4015
04:04:09.650 --> 04:04:13.971
<v Logan Graham (Anthropic)>And to to your point on the, the
information environment and otherwise,

4016
04:04:14.530 --> 04:04:16.530
<v Logan Graham (Anthropic)>I'll I'll point to a couple things here.

4017
04:04:17.010 --> 04:04:21.915
<v Logan Graham (Anthropic)>First is our focus on testing for Claude
itself and our our language model products

4018
04:04:21.915 --> 04:04:25.515
<v Logan Graham (Anthropic)>approach here.... Speak a little louder,
please? Sorry. Can you speak a little

4019
04:04:25.515 --> 04:04:26.155
<v Logan Graham (Anthropic)>louder?

4020
04:04:26.555 --> 04:04:30.155
<v Logan Graham (Anthropic)>Yes. Can you hear me now? Yeah.
That's better. Yes. Thank you.

4021
04:04:30.635 --> 04:04:33.354
<v Logan Graham (Anthropic)>I'll I'll point to our our long standing

4022
04:04:34.130 --> 04:04:35.170
<v Logan Graham (Anthropic)>evaluations

4023
04:04:35.170 --> 04:04:37.170
<v Logan Graham (Anthropic)>of political even handedness,

4024
04:04:37.489 --> 04:04:38.930
<v Logan Graham (Anthropic)>stereotype bias.

4025
04:04:39.970 --> 04:04:41.970
<v Logan Graham (Anthropic)>We train Claude to,

4026
04:04:42.290 --> 04:04:45.090
<v Logan Graham (Anthropic)>not comply with particularly
harmful requests,

4027
04:04:45.170 --> 04:04:48.689
<v Logan Graham (Anthropic)>in part for this reason and that this is
baked into the the nature of the model.

4028
04:04:52.175 --> 04:04:56.415
<v CM Selvena Brooks-Powers>Thank you for that, and thank you,
chairs. And I'll just say that in closing

4029
04:04:56.896 --> 04:04:57.615
<v CM Selvena Brooks-Powers>that

4030
04:04:58.015 --> 04:05:02.255
<v CM Selvena Brooks-Powers>I know that the AI companies
are pushing folks to

4031
04:05:02.520 --> 04:05:05.000
<v CM Selvena Brooks-Powers>allow AI to kind of be developed

4032
04:05:05.000 --> 04:05:05.720
<v CM Selvena Brooks-Powers>through

4033
04:05:06.920 --> 04:05:07.560
<v CM Selvena Brooks-Powers>people.

4034
04:05:08.359 --> 04:05:14.359
<v CM Selvena Brooks-Powers>But I think it's fair to say that
using live users to test these products

4035
04:05:14.936 --> 04:05:20.135
<v CM Selvena Brooks-Powers>can present dangers and definitely warrant
improvements, and I hope that's something

4036
04:05:20.135 --> 04:05:25.976
<v CM Selvena Brooks-Powers>that is taken from today's hearing to better
improve the the guardrails that need to

4037
04:05:25.976 --> 04:05:30.470
<v CM Selvena Brooks-Powers>exist. So thank you for your responses.
Thank you, chair. Thank thank you. We're

4038
04:05:30.470 --> 04:05:31.670
<v CM Selvena Brooks-Powers>going to go to

4039
04:05:32.310 --> 04:05:34.710
<v CM Carmen De La Rosa>council member Murano as a bill sponsor,

4040
04:05:34.950 --> 04:05:37.670
<v CM Carmen De La Rosa>and then we're gonna go Brewer, Narcisse,

4041
04:05:38.310 --> 04:05:39.590
<v CM Carmen De La Rosa>and Joseph.

4042
04:05:39.590 --> 04:05:44.045
<v CM Carmen De La Rosa>If folks could keep it to one minute just
because we have to get to the admin panel.

4043
04:05:44.045 --> 04:05:46.125
<v CM Carmen De La Rosa>Thank you so much. Thank you, chair.

4044
04:05:46.685 --> 04:05:50.285
<v CM Frank Morano>Thank you for your testimony today. I know
it's a long day for you guys as well.

4045
04:05:50.604 --> 04:05:54.365
<v CM Frank Morano>Because I only have a minute and my colleagues
know what a challenge that is for me,

4046
04:05:54.365 --> 04:05:57.645
<v CM Frank Morano>I'm gonna go through a series
of yes or no questions.

4047
04:05:58.180 --> 04:06:03.460
<v CM Frank Morano>And then if you could all answer in the
order of Dwyer, Friend, Cahill, Graham

4048
04:06:03.460 --> 04:06:04.580
<v CM Frank Morano>quickly.

4049
04:06:04.739 --> 04:06:09.220
<v CM Frank Morano>And then if you want to elaborate on any
of the areas that I raised, please do so.

4050
04:06:09.619 --> 04:06:15.064
<v CM Frank Morano>First, does your company believe
there's any nonzero chance

4051
04:06:15.225 --> 04:06:21.545
<v CM Frank Morano>that advanced AI could escape meaningful
human control and cause catastrophic harm

4052
04:06:21.545 --> 04:06:22.984
<v CM Frank Morano>as the whistleblowers indicated?

4053
04:06:27.760 --> 04:06:33.279
<v CM Frank Morano>I I don't know. What I will say is that
no level of risk of catastrophic harm

4054
04:06:33.279 --> 04:06:37.439
<v CM Frank Morano>is...... Just do yes or no now, and then
I'll invite you guys to elaborate after

4055
04:06:37.439 --> 04:06:39.520
<v CM Frank Morano>everyone answers. Please, miss Friend.

4056
04:06:42.295 --> 04:06:43.175
<v Alice Friend (Google)>Dwyer?

4057
04:06:43.175 --> 04:06:43.896
<v Alice Friend (Google)>I...

4058
04:06:44.535 --> 04:06:50.295
<v Alice Friend (Google)>The purpose of our Frontier Safety Framework
is to guard against that possibility.

4059
04:06:50.295 --> 04:06:52.295
<v Alice Friend (Google)>So do we entertain the possibility?

4060
04:06:53.000 --> 04:06:57.480
<v Alice Friend (Google)>Yes. As a responsible actor in this
space, we we entertain the possibility.

4061
04:07:00.680 --> 04:07:04.920
<v Shane Cahill (Meta)>Council member, our scaling framework includes
loss of control in terms of the risks

4062
04:07:04.920 --> 04:07:06.439
<v Shane Cahill (Meta)>that we evaluate and mitigate.

4063
04:07:07.505 --> 04:07:12.225
<v CM Frank Morano>So... And these are my other yes or no
questions and handle them however you want.

4064
04:07:12.225 --> 04:07:14.705
<v CM Frank Morano>So why should the public accept

4065
04:07:14.865 --> 04:07:20.064
<v CM Frank Morano>essentially trust us as the safety standard
from the same companies racing each

4066
04:07:20.064 --> 04:07:24.770
<v CM Frank Morano>other to build the most powerful systems?
Additionally, will you commit today that

4067
04:07:24.770 --> 04:07:29.649
<v CM Frank Morano>if your own safety team says a
model is too dangerous to deploy,

4068
04:07:30.050 --> 04:07:31.890
<v CM Frank Morano>commercial pressure won't

4069
04:07:32.050 --> 04:07:34.210
<v CM Frank Morano>ever overrule that judgment?

4070
04:07:34.449 --> 04:07:37.649
<v CM Frank Morano>Additionally, has your company
ever continued developing

4071
04:07:38.295 --> 04:07:43.895
<v CM Frank Morano>or deploying a model over the objections
of members of your own safety team? And

4072
04:07:43.895 --> 04:07:45.495
<v CM Frank Morano>lastly, on my bill,

4073
04:07:46.055 --> 04:07:49.175
<v CM Frank Morano>should you be allowed to
use someone else's...

4074
04:07:49.255 --> 04:07:51.734
<v CM Frank Morano>Someone's private chatbot conversation

4075
04:07:52.170 --> 04:07:54.250
<v CM Frank Morano>for targeted advertising

4076
04:07:54.250 --> 04:07:56.330
<v CM Frank Morano>or behavioral profiling

4077
04:07:56.490 --> 04:07:58.650
<v CM Frank Morano>without their affirmative consent?

4078
04:08:02.490 --> 04:08:07.370
<v Morgan Dwyer (OpenAI)>So I I will start, and try to get
through as many answers as possible.

4079
04:08:08.364 --> 04:08:09.005
<v Morgan Dwyer (OpenAI)>One,

4080
04:08:09.564 --> 04:08:10.765
<v Morgan Dwyer (OpenAI)>OpenAI supports,

4081
04:08:10.925 --> 04:08:15.244
<v Morgan Dwyer (OpenAI)>the need not only for us to be responsible
for safety, but we have supported

4082
04:08:15.244 --> 04:08:16.285
<v Morgan Dwyer (OpenAI)>regulations,

4083
04:08:16.524 --> 04:08:21.189
<v Morgan Dwyer (OpenAI)>at the federal level as well as at the
state level because we think it's important

4084
04:08:21.189 --> 04:08:23.750
<v Morgan Dwyer (OpenAI)>not only for us to be responsible,

4085
04:08:23.829 --> 04:08:26.310
<v Morgan Dwyer (OpenAI)>but for governments to
hold us accountable.

4086
04:08:27.109 --> 04:08:28.950
<v Morgan Dwyer (OpenAI)>On the question of,

4087
04:08:29.670 --> 04:08:33.510
<v Morgan Dwyer (OpenAI)>whether we would continue developing
models if they were unsafe,

4088
04:08:34.035 --> 04:08:39.955
<v Morgan Dwyer (OpenAI)>OpenAI has recently announced a pause of
some of our training activities because

4089
04:08:40.115 --> 04:08:45.635
<v Morgan Dwyer (OpenAI)>we did not deem moving forward to be
safe. We have also delayed the release of

4090
04:08:45.635 --> 04:08:48.915
<v Morgan Dwyer (OpenAI)>models because we didn't think that
they had sufficient safeguards.

4091
04:08:49.511 --> 04:08:52.230
<v Morgan Dwyer (OpenAI)>So we've done it before.
We will do it again.

4092
04:08:53.910 --> 04:08:58.870
<v Morgan Dwyer (OpenAI)>And more broadly, I will say that all of
our decisions regarding frontier safety and

4093
04:08:58.870 --> 04:09:02.710
<v Morgan Dwyer (OpenAI)>the release of models are governed
by our preparedness framework

4094
04:09:02.985 --> 04:09:07.945
<v Morgan Dwyer (OpenAI)>that is available online. So we're transparent
about what goes into those decisions.

4095
04:09:08.585 --> 04:09:10.905
<v Morgan Dwyer (OpenAI)>And then with the release of every model,

4096
04:09:11.065 --> 04:09:13.305
<v Morgan Dwyer (OpenAI)>every major model, we release,

4097
04:09:13.465 --> 04:09:15.865
<v Morgan Dwyer (OpenAI)>a system card which details,

4098
04:09:16.105 --> 04:09:17.146
<v Morgan Dwyer (OpenAI)>the work

4099
04:09:18.420 --> 04:09:21.780
<v Morgan Dwyer (OpenAI)>evaluations that we've conducted,
and any residual risks.

4100
04:09:25.220 --> 04:09:27.300
<v Shane Cahill (Meta)>Council member, in

4101
04:09:27.300 --> 04:09:28.900
<v Shane Cahill (Meta)>relation to your first question,

4102
04:09:29.620 --> 04:09:32.420
<v Shane Cahill (Meta)>I I see it differently
where we are accountable.

4103
04:09:33.625 --> 04:09:38.665
<v Shane Cahill (Meta)>Our framework and preparedness
reports are public. Meta uses outside

4104
04:09:38.665 --> 04:09:39.705
<v Shane Cahill (Meta)>expertise.

4105
04:09:40.105 --> 04:09:45.065
<v Shane Cahill (Meta)>We signed the accord at the at at the
White House on super intelligence,

4106
04:09:45.910 --> 04:09:48.710
<v Shane Cahill (Meta)>which committed us to
partnering with independent

4107
04:09:49.189 --> 04:09:50.790
<v Shane Cahill (Meta)>auditors and assessors.

4108
04:09:51.350 --> 04:09:56.311
<v Shane Cahill (Meta)>For example, in relation to... I think
your your your your second question,

4109
04:09:56.710 --> 04:09:59.270
<v Shane Cahill (Meta)>we delayed the launch of Muse

4110
04:09:59.745 --> 04:10:03.345
<v Shane Cahill (Meta)>by several months to focus
on safety and security.

4111
04:10:04.225 --> 04:10:07.905
<v Shane Cahill (Meta)>And we acted on this without waiting for
industry. It was just part of our normal

4112
04:10:07.905 --> 04:10:08.945
<v Shane Cahill (Meta)>day to day

4113
04:10:09.425 --> 04:10:12.625
<v Shane Cahill (Meta)>work because it was the right thing for us
and it was the right thing for people to

4114
04:10:12.625 --> 04:10:12.785
<v Shane Cahill (Meta)>do.

4115
04:10:13.840 --> 04:10:18.881
<v Shane Cahill (Meta)>And then in relation to ads, I would
just note that for our Muse product,

4116
04:10:19.601 --> 04:10:20.720
<v Shane Cahill (Meta)>conversations

4117
04:10:20.720 --> 04:10:22.961
<v Shane Cahill (Meta)>and interactions don't... Do not inform

4118
04:10:23.120 --> 04:10:24.561
<v Shane Cahill (Meta)>ads on Metas or their products.

4119
04:10:27.035 --> 04:10:30.716
<v Alice Friend (Google)>So Google... At Google, we
do use external evaluators,

4120
04:10:31.195 --> 04:10:32.955
<v Alice Friend (Google)>in the process of,

4121
04:10:33.596 --> 04:10:36.876
<v Alice Friend (Google)>research and development of
our AI models and our systems.

4122
04:10:37.436 --> 04:10:39.050
<v Alice Friend (Google)>We have supported

4123
04:10:39.050 --> 04:10:41.850
<v Alice Friend (Google)>audits of our safety procedures as well.

4124
04:10:42.490 --> 04:10:46.569
<v Alice Friend (Google)>And you will, see in our recent paper
from over the summer, which is called a

4125
04:10:46.569 --> 04:10:49.770
<v Alice Friend (Google)>pragmatic approach to AI
regulation in America.

4126
04:10:50.490 --> 04:10:51.290
<v Alice Friend (Google)>We do think

4127
04:10:52.055 --> 04:10:56.135
<v Alice Friend (Google)>that AI needs to be regulated,
and it needs to be regulated well.

4128
04:10:56.455 --> 04:10:58.935
<v Alice Friend (Google)>We've supported a range of different,

4129
04:10:59.175 --> 04:11:00.854
<v Alice Friend (Google)>legislative proposals

4130
04:11:00.854 --> 04:11:03.734
<v Alice Friend (Google)>at the federal and state
levels, to this end.

4131
04:11:04.690 --> 04:11:09.011
<v Alice Friend (Google)>We also, to your question
about conversations,

4132
04:11:09.330 --> 04:11:14.290
<v Alice Friend (Google)>we give our users controls over whether
their activity in our tools can be used for,

4133
04:11:14.450 --> 04:11:18.370
<v Alice Friend (Google)>training our AI models. So, yes, we
think users should be in control of that.

4134
04:11:19.925 --> 04:11:24.165
<v Logan Graham (Anthropic)>And, council member, I I would point to
our long standing call for the involvement

4135
04:11:24.165 --> 04:11:29.125
<v Logan Graham (Anthropic)>of government at the federal level and
the state where necessary and appropriate,

4136
04:11:29.845 --> 04:11:33.365
<v Logan Graham (Anthropic)>because this goes beyond the labs, and we
don't think the labs should be checking

4137
04:11:33.365 --> 04:11:34.165
<v Logan Graham (Anthropic)>their own homework.

4138
04:11:35.740 --> 04:11:41.740
<v Logan Graham (Anthropic)>Not only this, but, we have demonstrably
withheld models when we felt it was not

4139
04:11:41.740 --> 04:11:43.180
<v Logan Graham (Anthropic)>safe to widely release.

4140
04:11:43.500 --> 04:11:47.260
<v Logan Graham (Anthropic)>This year, withholding our frontier model,
probably the most notable instance of

4141
04:11:47.260 --> 04:11:47.740
<v Logan Graham (Anthropic)>this,

4142
04:11:48.140 --> 04:11:52.915
<v Logan Graham (Anthropic)>instead deploying it to cyber defenders in
order to secure the world in advance of a

4143
04:11:52.915 --> 04:11:56.675
<v Logan Graham (Anthropic)>world of more powerful models that could
exploit software vulnerabilities from

4144
04:11:56.675 --> 04:11:57.396
<v Logan Graham (Anthropic)>wherever.

4145
04:11:57.795 --> 04:12:02.194
<v Logan Graham (Anthropic)>This is part of our process. We go over and
above the process. We detail our default

4146
04:12:02.194 --> 04:12:03.635
<v Logan Graham (Anthropic)>in the responsible scaling policy.

4147
04:12:04.420 --> 04:12:05.859
<v Logan Graham (Anthropic)>And to your other question,

4148
04:12:06.100 --> 04:12:06.899
<v Logan Graham (Anthropic)>we

4149
04:12:06.980 --> 04:12:07.700
<v Logan Graham (Anthropic)>believe,

4150
04:12:08.020 --> 04:12:11.140
<v Logan Graham (Anthropic)>that Claude should act unambiguously
in the user's interest.

4151
04:12:11.380 --> 04:12:15.380
<v Logan Graham (Anthropic)>And for that reason, we do not
run ads nor are Claude's responses

4152
04:12:15.380 --> 04:12:16.740
<v Logan Graham (Anthropic)>influenced by,

4153
04:12:17.060 --> 04:12:17.699
<v Logan Graham (Anthropic)>advertisers.

4154
04:12:19.105 --> 04:12:23.665
<v CM Gale Brewer>Thank you so much. Council member Brewer
followed by Narcisse. Thank you very much.

4155
04:12:23.985 --> 04:12:26.865
<v CM Gale Brewer>I wanted to know three
quick questions. If a

4156
04:12:28.065 --> 04:12:28.705
<v CM Gale Brewer>model,

4157
04:12:28.945 --> 04:12:32.065
<v CM Gale Brewer>provider you rely on,
you or a subcontractor

4158
04:12:32.065 --> 04:12:35.000
<v CM Gale Brewer>did not complete the
third party validation

4159
04:12:35.000 --> 04:12:36.681
<v CM Gale Brewer>for a particular package,

4160
04:12:37.080 --> 04:12:39.561
<v CM Gale Brewer>how would that affect your
product's availability?

4161
04:12:39.640 --> 04:12:43.480
<v CM Gale Brewer>And, also, how do you decide
who gets the validation?

4162
04:12:43.560 --> 04:12:45.561
<v CM Gale Brewer>Number one. Number two, China.

4163
04:12:45.800 --> 04:12:50.385
<v CM Gale Brewer>The speaker asked about China and competition,
and I thought the answer was, don't

4164
04:12:50.385 --> 04:12:50.945
<v CM Gale Brewer>worry.

4165
04:12:51.186 --> 04:12:54.945
<v CM Gale Brewer>They need information from us. So I'm
wondering what do you send to China?

4166
04:12:55.105 --> 04:12:56.945
<v CM Gale Brewer>And number three, ICE.

4167
04:12:57.105 --> 04:13:00.865
<v CM Gale Brewer>I was upset that one of the whistleblower
said that Google has been working

4168
04:13:00.865 --> 04:13:03.905
<v CM Gale Brewer>something to do with ICE and
Minnesota. So my question is,

4169
04:13:04.891 --> 04:13:07.131
<v CM Gale Brewer>do you all support

4170
04:13:07.690 --> 04:13:10.250
<v CM Gale Brewer>not helping ICE at all?

4171
04:13:10.570 --> 04:13:11.290
<v CM Gale Brewer>Obviously,

4172
04:13:11.530 --> 04:13:15.130
<v CM Gale Brewer>you are made up of immigrants. We should
not be shipping immigrants out of this

4173
04:13:15.130 --> 04:13:17.690
<v CM Gale Brewer>country. Please answer those
three questions, and thank you.

4174
04:13:21.085 --> 04:13:24.765
<v Morgan Dwyer (OpenAI)>So I'll start with a question
on third party evaluations.

4175
04:13:25.404 --> 04:13:29.245
<v Morgan Dwyer (OpenAI)>We work with a range of third
parties to evaluate our models.

4176
04:13:29.885 --> 04:13:31.085
<v Morgan Dwyer (OpenAI)>That includes,

4177
04:13:31.404 --> 04:13:34.685
<v Morgan Dwyer (OpenAI)>independent investigators as well
as with the federal government.

4178
04:13:35.569 --> 04:13:36.930
<v Morgan Dwyer (OpenAI)>And then we detail,

4179
04:13:37.170 --> 04:13:42.370
<v Morgan Dwyer (OpenAI)>who we worked with and the results, in our
system card when we released our model.

4180
04:13:43.569 --> 04:13:48.635
<v Morgan Dwyer (OpenAI)>Decisions to launch our models are described
in our preparedness framework, And so

4181
04:13:48.635 --> 04:13:53.195
<v Morgan Dwyer (OpenAI)>that describes how we integrate
all of the results of evaluations,

4182
04:13:53.755 --> 04:13:55.195
<v Morgan Dwyer (OpenAI)>and make decisions,

4183
04:13:55.675 --> 04:14:00.075
<v Morgan Dwyer (OpenAI)>with respect to launching models.
So I would point you there on that,

4184
04:14:00.315 --> 04:14:01.275
<v Morgan Dwyer (OpenAI)>particular issue.

4185
04:14:02.700 --> 04:14:04.300
<v Morgan Dwyer (OpenAI)>On the question of,

4186
04:14:04.620 --> 04:14:05.500
<v Morgan Dwyer (OpenAI)>China,

4187
04:14:05.660 --> 04:14:07.180
<v Morgan Dwyer (OpenAI)>and the the PRC,

4188
04:14:07.980 --> 04:14:11.739
<v Morgan Dwyer (OpenAI)>I would like to get back to
you on that particular issue.

4189
04:14:11.980 --> 04:14:16.060
<v Morgan Dwyer (OpenAI)>I... Although I have a nurse... National
security background, I'm not working in

4190
04:14:16.060 --> 04:14:16.779
<v Morgan Dwyer (OpenAI)>that role

4191
04:14:17.315 --> 04:14:20.035
<v Morgan Dwyer (OpenAI)>at the company, and I
wanna make sure I get you

4192
04:14:20.915 --> 04:14:22.115
<v Morgan Dwyer (OpenAI)>good information.

4193
04:14:22.755 --> 04:14:23.476
<v Morgan Dwyer (OpenAI)>Similarly,

4194
04:14:23.715 --> 04:14:28.115
<v Morgan Dwyer (OpenAI)>with respect to ICE, I'd like to get back
to you. I'm not aware of any activities

4195
04:14:28.115 --> 04:14:33.250
<v Morgan Dwyer (OpenAI)>working with ICE, but I don't work on our
government sales team, so I'd like to get

4196
04:14:33.250 --> 04:14:34.290
<v Morgan Dwyer (OpenAI)>back to you on that.

4197
04:14:35.810 --> 04:14:36.450
<v Morgan Dwyer (OpenAI)>Next.

4198
04:14:36.770 --> 04:14:39.730
<v Shane Cahill (Meta)>Council member, thank you very
much for for your three questions.

4199
04:14:40.130 --> 04:14:41.891
<v Shane Cahill (Meta)>In relation to... Three questions.

4200
04:14:42.290 --> 04:14:47.075
<v Shane Cahill (Meta)>Pardon? Three questions. Yes. Three. Sorry.
My accent might have gotten in the way

4201
04:14:47.075 --> 04:14:47.955
<v Shane Cahill (Meta)>there. Apologies.

4202
04:14:48.355 --> 04:14:48.995
<v Shane Cahill (Meta)>Three

4203
04:14:49.315 --> 04:14:49.955
<v Shane Cahill (Meta)>questions.

4204
04:14:50.355 --> 04:14:52.516
<v Shane Cahill (Meta)>The... For third party validators,

4205
04:14:52.675 --> 04:14:55.955
<v Shane Cahill (Meta)>we also work with a range of validators

4206
04:14:56.516 --> 04:14:57.715
<v Shane Cahill (Meta)>assessors,

4207
04:14:57.715 --> 04:14:58.436
<v Shane Cahill (Meta)>and

4208
04:14:58.680 --> 04:15:03.560
<v Shane Cahill (Meta)>we also provide this information in our
publicly available preparedness reports.

4209
04:15:04.680 --> 04:15:07.320
<v Shane Cahill (Meta)>In relation to China and competition,

4210
04:15:07.720 --> 04:15:13.000
<v Shane Cahill (Meta)>I might just mention that, you know,
Meta is a proud American company

4211
04:15:14.195 --> 04:15:16.436
<v Shane Cahill (Meta)>and proud to serve billions of people

4212
04:15:16.755 --> 04:15:19.235
<v Shane Cahill (Meta)>around the world who rely on our services.

4213
04:15:19.875 --> 04:15:24.035
<v Shane Cahill (Meta)>We want to ensure that The US and
democratic countries lead on AI.

4214
04:15:24.675 --> 04:15:28.801
<v Shane Cahill (Meta)>And one thing I would just note here is
in relation to the importance of open

4215
04:15:28.801 --> 04:15:30.881
<v Shane Cahill (Meta)>weights and American open weights

4216
04:15:31.440 --> 04:15:33.360
<v Shane Cahill (Meta)>and the criticality of ensuring

4217
04:15:33.440 --> 04:15:35.841
<v Shane Cahill (Meta)>that American innovation

4218
04:15:35.920 --> 04:15:38.960
<v Shane Cahill (Meta)>continues to lead in
relation to the availability

4219
04:15:39.875 --> 04:15:41.476
<v Shane Cahill (Meta)>of open weight models.

4220
04:15:41.875 --> 04:15:43.396
<v Shane Cahill (Meta)>And in relation to ICE,

4221
04:15:43.636 --> 04:15:47.875
<v Shane Cahill (Meta)>if I may, I would like to connect you
with, with our teams and experts in that

4222
04:15:47.875 --> 04:15:51.636
<v Shane Cahill (Meta)>space. I don't wish to, to be
imprecise in responding to you.

4223
04:15:52.915 --> 04:15:53.235
<v CM Gale Brewer [?]>Next.

4224
04:15:54.710 --> 04:15:57.431
<v Alice Friend (Google)>Minh, in terms of third party evaluators,

4225
04:15:57.431 --> 04:16:01.910
<v Alice Friend (Google)>we also work with the US Casey, which is
the organization underneath the commerce

4226
04:16:01.910 --> 04:16:03.431
<v Alice Friend (Google)>department that does,

4227
04:16:03.591 --> 04:16:05.751
<v Alice Friend (Google)>testing of advanced AI models.

4228
04:16:06.230 --> 04:16:09.270
<v Alice Friend (Google)>We also work with a range
of third party evaluators.

4229
04:16:09.575 --> 04:16:11.896
<v Alice Friend (Google)>It's a very nascent ecosystem.

4230
04:16:12.295 --> 04:16:17.016
<v Alice Friend (Google)>And one of the challenges in evaluating
AI models, of course, is to get,

4231
04:16:17.495 --> 04:16:21.735
<v Alice Friend (Google)>organizations that both have the
technical competency to do evaluations

4232
04:16:21.735 --> 04:16:25.841
<v Alice Friend (Google)>well, also have the subject matter
expertise that we may be looking for.

4233
04:16:26.240 --> 04:16:29.921
<v Alice Friend (Google)>Earlier, one of the council members
mentioned concerns about bio, for example.

4234
04:16:30.480 --> 04:16:34.960
<v Alice Friend (Google)>We have third parties that, have
the bio expertise that's necessary.

4235
04:16:35.535 --> 04:16:39.215
<v Alice Friend (Google)>But as we have this national
conversation about,

4236
04:16:39.615 --> 04:16:43.135
<v Alice Friend (Google)>independent verification organizations
and third party evaluations,

4237
04:16:43.375 --> 04:16:47.055
<v Alice Friend (Google)>keeping in mind that we still
need to mature this ecosystem

4238
04:16:47.055 --> 04:16:48.735
<v Alice Friend (Google)>of independent evaluators,

4239
04:16:48.735 --> 04:16:49.854
<v Alice Friend (Google)>I think, is really important.

4240
04:16:50.710 --> 04:16:52.791
<v Alice Friend (Google)>On China and competition,

4241
04:16:53.351 --> 04:16:58.311
<v Alice Friend (Google)>I'd I'd like the opportunity to
give you, a more complete answer.

4242
04:16:58.471 --> 04:17:00.311
<v Alice Friend (Google)>We don't conduct AI

4243
04:17:00.391 --> 04:17:05.976
<v Alice Friend (Google)>research or trading in China. I know that,
we are part of the conversation about

4244
04:17:05.976 --> 04:17:08.936
<v Alice Friend (Google)>what it means for America to lead in AI,

4245
04:17:09.255 --> 04:17:14.455
<v Alice Friend (Google)>where we are, in the technological race
with China. I will just mention that

4246
04:17:14.980 --> 04:17:16.740
<v Alice Friend (Google)>we focus a lot on,

4247
04:17:17.061 --> 04:17:19.061
<v Alice Friend (Google)>who's, at the frontier,

4248
04:17:19.700 --> 04:17:20.900
<v Alice Friend (Google)>which country,

4249
04:17:21.140 --> 04:17:23.780
<v Alice Friend (Google)>has the most cutting edge research in AI.

4250
04:17:24.100 --> 04:17:27.940
<v Alice Friend (Google)>But we talk much less often,
about deployment of AI.

4251
04:17:28.180 --> 04:17:31.595
<v Alice Friend (Google)>And there, I think we should really
keep an eye on the Chinese. They have

4252
04:17:31.595 --> 04:17:32.475
<v Alice Friend (Google)>emphasized,

4253
04:17:32.795 --> 04:17:34.955
<v Alice Friend (Google)>deploying AI in their economy,

4254
04:17:35.595 --> 04:17:40.955
<v Alice Friend (Google)>quite a bit, and and that's a place where
I sometimes have concerns that The United

4255
04:17:40.955 --> 04:17:41.515
<v Alice Friend (Google)>States,

4256
04:17:41.995 --> 04:17:44.475
<v Alice Friend (Google)>is in danger of falling behind China.

4257
04:17:45.800 --> 04:17:47.399
<v Alice Friend (Google)>As to the whistleblower's,

4258
04:17:47.640 --> 04:17:48.520
<v Alice Friend (Google)>allegations

4259
04:17:48.520 --> 04:17:53.480
<v Alice Friend (Google)>from today, I don't have any specifics
about that, but I'd be happy to ask, the

4260
04:17:53.480 --> 04:17:57.160
<v Alice Friend (Google)>public policy team in New York to get back
to you right away with any information

4261
04:17:57.160 --> 04:17:57.960
<v Alice Friend (Google)>that we have about it.

4262
04:17:58.795 --> 04:17:59.435
<v CM Gale Brewer [?]>Next.

4263
04:18:00.154 --> 04:18:01.914
<v Logan Graham (Anthropic)>And at Anthropic,

4264
04:18:02.154 --> 04:18:02.875
<v Logan Graham (Anthropic)>the

4265
04:18:03.194 --> 04:18:05.675
<v Logan Graham (Anthropic)>use of third parties for verification,

4266
04:18:06.234 --> 04:18:07.675
<v Logan Graham (Anthropic)>developing and using,

4267
04:18:08.795 --> 04:18:12.795
<v Logan Graham (Anthropic)>evaluations for models has been the
part of the process from day one,

4268
04:18:13.360 --> 04:18:14.641
<v Logan Graham (Anthropic)>including

4269
04:18:14.641 --> 04:18:20.080
<v Logan Graham (Anthropic)>with not just external parties in industry
or nonprofits, but also with governments

4270
04:18:20.080 --> 04:18:21.921
<v Logan Graham (Anthropic)>such as the USKC.

4271
04:18:22.641 --> 04:18:27.040
<v Logan Graham (Anthropic)>This has been foundational. As my colleague
said, it's pretty nascent. We think it

4272
04:18:27.040 --> 04:18:28.245
<v Logan Graham (Anthropic)>needs to be urgently

4273
04:18:28.245 --> 04:18:29.686
<v Logan Graham (Anthropic)>urgently expanded

4274
04:18:29.765 --> 04:18:34.485
<v Logan Graham (Anthropic)>and not just used for predeployment
testing, but also to understand,

4275
04:18:34.965 --> 04:18:38.885
<v Logan Graham (Anthropic)>models and their behavior after deployment
and the safety processes at companies.

4276
04:18:40.400 --> 04:18:44.400
<v Logan Graham (Anthropic)>In the case of of China, I think it's very
clear that the first thing that we need

4277
04:18:44.400 --> 04:18:49.040
<v Logan Graham (Anthropic)>to do is maintain an American and
democratic lead in model capabilities,

4278
04:18:49.040 --> 04:18:53.295
<v Logan Graham (Anthropic)>but we're very clear about the threat
elsewhere as well. So we have publicly

4279
04:18:53.295 --> 04:18:54.255
<v Logan Graham (Anthropic)>documented,

4280
04:18:54.815 --> 04:18:58.655
<v Logan Graham (Anthropic)>Chinese attempts to distill our models,
essentially take the content and

4281
04:18:58.655 --> 04:19:00.415
<v Logan Graham (Anthropic)>capabilities for their own,

4282
04:19:00.895 --> 04:19:03.056
<v Logan Graham (Anthropic)>and we work to ban those.

4283
04:19:03.455 --> 04:19:06.610
<v Logan Graham (Anthropic)>We have, prohibited use of,

4284
04:19:06.930 --> 04:19:07.811
<v Logan Graham (Anthropic)>models,

4285
04:19:08.130 --> 04:19:10.210
<v Logan Graham (Anthropic)>in China. We do not serve there.

4286
04:19:10.690 --> 04:19:12.851
<v Logan Graham (Anthropic)>And we have also documented,

4287
04:19:13.410 --> 04:19:15.410
<v Logan Graham (Anthropic)>the use, we believe, by,

4288
04:19:15.730 --> 04:19:20.770
<v Logan Graham (Anthropic)>related parties of Claude or the
attempt to for cyber offensive reasons.

4289
04:19:21.015 --> 04:19:24.534
<v Logan Graham (Anthropic)>We are extremely serious about this, and
we think it's a very substantial priority.

4290
04:19:24.534 --> 04:19:29.255
<v Logan Graham (Anthropic)>This is one reason why we have advocated at
the federal level for export controls on

4291
04:19:29.255 --> 04:19:31.495
<v Logan Graham (Anthropic)>the most important chips that power,

4292
04:19:31.895 --> 04:19:33.095
<v Logan Graham (Anthropic)>AI innovation here.

4293
04:19:33.590 --> 04:19:38.230
<v Logan Graham (Anthropic)>And on your third question, while I'm not,
across the issue enough to know what our

4294
04:19:38.230 --> 04:19:39.591
<v Logan Graham (Anthropic)>what our engagements are,

4295
04:19:40.150 --> 04:19:42.150
<v Logan Graham (Anthropic)>I will say that our usage policy,

4296
04:19:42.391 --> 04:19:43.831
<v Logan Graham (Anthropic)>we ban surveillance,

4297
04:19:43.910 --> 04:19:44.791
<v Logan Graham (Anthropic)>censorship,

4298
04:19:45.190 --> 04:19:47.990
<v Logan Graham (Anthropic)>prohibited law enforcement,
and criminal justice uses

4299
04:19:48.306 --> 04:19:50.706
<v Logan Graham (Anthropic)>for for this purpose. Thank you.

4300
04:19:51.266 --> 04:19:54.626
<v CM Carmen De La Rosa>Thank you. We have
council members Narcisse,

4301
04:19:54.785 --> 04:19:58.546
<v CM Carmen De La Rosa>Joseph p Sanchez, and then,
council member Phil Wong.

4302
04:19:58.785 --> 04:19:59.905
<v CM Carmen De La Rosa>So, Narcisse.

4303
04:20:02.230 --> 04:20:05.510
<v CM Mercedes Narcisse>Good afternoon. Thank you, chair,
and thank you, madam speaker.

4304
04:20:06.069 --> 04:20:10.230
<v CM Mercedes Narcisse>I wanna say thank you to you too because
you guys stay here with us, and I'm happy

4305
04:20:10.230 --> 04:20:13.989
<v CM Mercedes Narcisse>to see at least two women is
on the panel to talk to us.

4306
04:20:15.325 --> 04:20:17.726
<v CM Mercedes Narcisse>Hollywood give us clues all the time,

4307
04:20:18.045 --> 04:20:22.365
<v CM Mercedes Narcisse>but we kind of slow in taking,
you know, hands on clues.

4308
04:20:22.605 --> 04:20:27.885
<v CM Mercedes Narcisse>But no companies in this world, I
feel that should be self regulated.

4309
04:20:28.320 --> 04:20:32.801
<v CM Mercedes Narcisse>What I'm sitting here for about four
hours, and like I said, thank you to you,

4310
04:20:32.801 --> 04:20:34.320
<v CM Mercedes Narcisse>privacy, transparency,

4311
04:20:34.641 --> 04:20:36.561
<v CM Mercedes Narcisse>and safety. So now

4312
04:20:36.641 --> 04:20:39.520
<v CM Mercedes Narcisse>do any of those company

4313
04:20:39.681 --> 04:20:44.785
<v CM Mercedes Narcisse>that you can say that is a come... A standard
across the board for all the company

4314
04:20:44.785 --> 04:20:47.266
<v CM Mercedes Narcisse>to follow because apparently not.

4315
04:20:48.065 --> 04:20:51.025
<v CM Mercedes Narcisse>Now do you regulate hospital?

4316
04:20:51.105 --> 04:20:55.666
<v CM Mercedes Narcisse>I mean, in... Do you have AI...
Sell AI? Sorry. Do you sell AI,

4317
04:20:56.065 --> 04:20:59.040
<v CM Mercedes Narcisse>products to New York City Hospital?

4318
04:20:59.199 --> 04:21:00.240
<v CM Mercedes Narcisse>And if so,

4319
04:21:00.880 --> 04:21:04.000
<v CM Mercedes Narcisse>do you think that a licensed clinician

4320
04:21:04.000 --> 04:21:08.560
<v CM Mercedes Narcisse>should sign on all the diagnostic
tests? Because I heard,

4321
04:21:08.880 --> 04:21:09.439
<v CM Mercedes Narcisse>if,

4322
04:21:09.760 --> 04:21:13.115
<v CM Mercedes Narcisse>miss Friend say that you diagnosed with...

4323
04:21:14.154 --> 04:21:19.835
<v CM Mercedes Narcisse>You went to a process of cancer, if I'm
correct. So you understand that people need

4324
04:21:19.835 --> 04:21:25.195
<v CM Mercedes Narcisse>to be diagnosed by human. By... And I can
give you example, like for colonoscopy.

4325
04:21:25.274 --> 04:21:30.210
<v CM Mercedes Narcisse>Sometimes the bubble, according to AI, will
tell you it's a polyps. In the meantime,

4326
04:21:30.210 --> 04:21:36.051
<v CM Mercedes Narcisse>it's not polyps. It's a... It's not a a...
It's not polyps. It's a bubble. But AI

4327
04:21:36.051 --> 04:21:40.610
<v CM Mercedes Narcisse>cannot differentiate that. So, therefore,
you need a human to see through.

4328
04:21:42.355 --> 04:21:46.835
<v CM Carmen De La Rosa>And right now, don't know if you read it
your times. I'm sorry. Can can you Yeah.

4329
04:21:46.835 --> 04:21:48.275
<v CM Carmen De La Rosa>I'm gonna close my last question.

4330
04:21:48.676 --> 04:21:51.955
<v CM Mercedes Narcisse>New York Times already
telling us that there's

4331
04:21:51.955 --> 04:21:57.631
<v CM Mercedes Narcisse>a report that AI is getting confusion for
the... Between the the hospital and the

4332
04:21:57.631 --> 04:22:02.110
<v CM Mercedes Narcisse>insurance company. So I don't know how you
see that happening, how we can make sure

4333
04:22:02.110 --> 04:22:04.190
<v CM Mercedes Narcisse>that AI is being monitored.

4334
04:22:06.830 --> 04:22:08.511
<v Morgan Dwyer (OpenAI)>Yes. So thank you for your question.

4335
04:22:09.865 --> 04:22:10.745
<v Morgan Dwyer (OpenAI)>ChatGPT

4336
04:22:10.745 --> 04:22:12.745
<v Morgan Dwyer (OpenAI)>for health care is deployed,

4337
04:22:12.985 --> 04:22:16.745
<v Morgan Dwyer (OpenAI)>for example, in Memorial
Sloan Kettering Cancer Center.

4338
04:22:17.545 --> 04:22:21.705
<v Morgan Dwyer (OpenAI)>But to your point, we believe that
doctors should remain in charge.

4339
04:22:22.850 --> 04:22:27.971
<v Morgan Dwyer (OpenAI)>We do not think that AI should represent
itself as a licensed professional.

4340
04:22:28.450 --> 04:22:32.850
<v Morgan Dwyer (OpenAI)>And we also think, though, as as we've
talked about earlier, there's tremendous

4341
04:22:32.850 --> 04:22:35.890
<v Morgan Dwyer (OpenAI)>potential to use AI in
health care settings,

4342
04:22:36.825 --> 04:22:38.906
<v Morgan Dwyer (OpenAI)>particularly for in a... Administrative

4343
04:22:39.466 --> 04:22:40.346
<v Morgan Dwyer (OpenAI)>uses,

4344
04:22:40.346 --> 04:22:41.226
<v Morgan Dwyer (OpenAI)>documentation.

4345
04:22:41.705 --> 04:22:45.945
<v Morgan Dwyer (OpenAI)>I mean, doctors are already overworked,
and giving them an opportunity to free up

4346
04:22:45.945 --> 04:22:49.546
<v Morgan Dwyer (OpenAI)>their time so that they can spend
more time with their patients

4347
04:22:49.860 --> 04:22:53.940
<v Morgan Dwyer (OpenAI)>is an outcome I think we can all
agree would lead to better care.

4348
04:22:56.580 --> 04:22:59.460
<v Shane Cahill (Meta)>Council member, thank you very
much for these questions.

4349
04:22:59.860 --> 04:23:04.020
<v Shane Cahill (Meta)>In relation to your first
question on self regulation,

4350
04:23:04.975 --> 04:23:06.255
<v Shane Cahill (Meta)>we believe that

4351
04:23:06.975 --> 04:23:08.175
<v Shane Cahill (Meta)>the accord is

4352
04:23:09.615 --> 04:23:10.896
<v Shane Cahill (Meta)>a really important

4353
04:23:11.055 --> 04:23:11.775
<v Shane Cahill (Meta)>step

4354
04:23:12.095 --> 04:23:13.455
<v Shane Cahill (Meta)>in relation to

4355
04:23:14.095 --> 04:23:15.375
<v Shane Cahill (Meta)>the concerns that you mentioned.

4356
04:23:16.891 --> 04:23:18.251
<v Shane Cahill (Meta)>In relation to

4357
04:23:18.811 --> 04:23:19.690
<v Shane Cahill (Meta)>health care,

4358
04:23:20.330 --> 04:23:24.410
<v Shane Cahill (Meta)>Meta AI is prohibited from acting as a
licensed professional, including a doctor or

4359
04:23:24.410 --> 04:23:26.811
<v Shane Cahill (Meta)>any other medical professional
that he's licensed.

4360
04:23:27.770 --> 04:23:31.131
<v Shane Cahill (Meta)>There are, as my colleagues
have mentioned, phenomenal

4361
04:23:31.131 --> 04:23:31.610
<v Shane Cahill (Meta)>benefits

4362
04:23:32.165 --> 04:23:36.404
<v Shane Cahill (Meta)>in relation to the use of AI in health
care. I know that there's phenomenal work

4363
04:23:36.404 --> 04:23:41.205
<v Shane Cahill (Meta)>happening at NYU Langone who actually, I
think, has a specialist division that's

4364
04:23:41.205 --> 04:23:46.101
<v Shane Cahill (Meta)>actually investing and working specifically
around AI and health care settings. And

4365
04:23:46.101 --> 04:23:48.020
<v Shane Cahill (Meta)>similarly, Columbia Irving,

4366
04:23:49.221 --> 04:23:51.061
<v Shane Cahill (Meta)>which which which

4367
04:23:51.061 --> 04:23:53.461
<v Shane Cahill (Meta)>is really doing phenomenal
work in that space also.

4368
04:23:55.221 --> 04:23:59.295
<v Alice Friend (Google)>Ma'am, thank you so much for the the...
Those questions. On whether or not a company

4369
04:23:59.295 --> 04:24:00.735
<v Alice Friend (Google)>should be self regulated,

4370
04:24:01.295 --> 04:24:02.976
<v Alice Friend (Google)>I again wanna emphasize,

4371
04:24:03.535 --> 04:24:06.495
<v Alice Friend (Google)>at Google, we believe
that we are regulated

4372
04:24:06.495 --> 04:24:07.775
<v Alice Friend (Google)>by existing

4373
04:24:08.016 --> 04:24:08.976
<v Alice Friend (Google)>regulations,

4374
04:24:08.976 --> 04:24:10.016
<v Alice Friend (Google)>sectoral regulations,

4375
04:24:10.690 --> 04:24:11.570
<v Alice Friend (Google)>state

4376
04:24:11.570 --> 04:24:16.690
<v Alice Friend (Google)>federal level regulations all apply to AI.
And we have supported further regulation,

4377
04:24:17.250 --> 04:24:20.051
<v Alice Friend (Google)>particularly of frontier
AI at the federal level,

4378
04:24:20.530 --> 04:24:21.490
<v Alice Friend (Google)>but also,

4379
04:24:21.811 --> 04:24:25.730
<v Alice Friend (Google)>other AI directed laws at,
all levels of government.

4380
04:24:26.825 --> 04:24:27.466
<v Alice Friend (Google)>So,

4381
04:24:28.825 --> 04:24:29.306
<v Alice Friend (Google)>no.

4382
04:24:29.785 --> 04:24:34.266
<v Alice Friend (Google)>We should not be self regulated. We we
we should, be within the the regulatory

4383
04:24:34.266 --> 04:24:35.706
<v Alice Friend (Google)>frameworks we already are.

4384
04:24:37.226 --> 04:24:38.105
<v Alice Friend (Google)>On the,

4385
04:24:38.426 --> 04:24:44.440
<v Alice Friend (Google)>questions about health care. So I... We
certainly agree that doctors should stay

4386
04:24:44.760 --> 04:24:48.520
<v Alice Friend (Google)>in the lead and that AI
is a tool for clinicians.

4387
04:24:48.520 --> 04:24:50.440
<v Alice Friend (Google)>It's not a replacement for them.

4388
04:24:51.160 --> 04:24:53.880
<v Alice Friend (Google)>In my particular case, I
was diagnosed by a person.

4389
04:24:54.864 --> 04:24:58.545
<v Alice Friend (Google)>But often these days when I go
in for checkups and screenings,

4390
04:24:58.864 --> 04:25:02.305
<v Alice Friend (Google)>my clinician will ask if they
can use an AI tool to take notes,

4391
04:25:02.705 --> 04:25:04.865
<v Alice Friend (Google)>while while we're doing the checkup.

4392
04:25:05.185 --> 04:25:10.064
<v Alice Friend (Google)>And I always say yes because I know that
our clinicians are incredibly overtaxed

4393
04:25:10.699 --> 04:25:12.699
<v Alice Friend (Google)>by a range of administrative

4394
04:25:12.699 --> 04:25:13.500
<v Alice Friend (Google)>requirements.

4395
04:25:13.500 --> 04:25:18.460
<v Alice Friend (Google)>And if AI can take some of that burden off
and allow my doctor to look me in the eye

4396
04:25:18.460 --> 04:25:20.780
<v Alice Friend (Google)>when we're having an appointment,

4397
04:25:21.260 --> 04:25:22.540
<v Alice Friend (Google)>I'm gonna say yes to that.

4398
04:25:23.604 --> 04:25:28.645
<v Alice Friend (Google)>But similar to our colleagues, we also
have guardrails on our Gemini app

4399
04:25:29.045 --> 04:25:33.845
<v Alice Friend (Google)>so that users will never think that they
are talking to a doctor, that they are

4400
04:25:33.845 --> 04:25:35.125
<v Alice Friend (Google)>talking to a licensed physician,

4401
04:25:35.900 --> 04:25:39.660
<v Alice Friend (Google)>and the Gemini app will also flag for
them that if they're seeking medical

4402
04:25:39.660 --> 04:25:42.381
<v Alice Friend (Google)>attention, they should get
it from a professional.

4403
04:25:43.180 --> 04:25:43.980
<v CM Gale Brewer [?]>Thank you.

4404
04:25:45.260 --> 04:25:45.980
<v Logan Graham (Anthropic)>Likewise,

4405
04:25:46.540 --> 04:25:50.220
<v Logan Graham (Anthropic)>we... We've been very clear that we
support the role of government here.

4406
04:25:50.944 --> 04:25:54.465
<v Logan Graham (Anthropic)>At the federal level, we've
advanced a proposal for meaningful

4407
04:25:54.705 --> 04:25:56.145
<v Logan Graham (Anthropic)>involvement to set,

4408
04:25:56.465 --> 04:26:00.385
<v Logan Graham (Anthropic)>standards and hold companies to account
for safety processes, evaluations,

4409
04:26:00.465 --> 04:26:04.145
<v Logan Graham (Anthropic)>security, and the like. And we welcome the
accords that were signed very recently.

4410
04:26:05.780 --> 04:26:10.340
<v Logan Graham (Anthropic)>To the point of medical usage, if I can
use a personal example here, about six

4411
04:26:10.340 --> 04:26:11.141
<v Logan Graham (Anthropic)>months ago,

4412
04:26:11.540 --> 04:26:16.020
<v Logan Graham (Anthropic)>I experienced some symptoms and long
standing eye disease that might make me go

4413
04:26:16.020 --> 04:26:20.735
<v Logan Graham (Anthropic)>blind at any moment. And that night,
actually, at about 1AM, it was Claude that

4414
04:26:20.735 --> 04:26:24.016
<v Logan Graham (Anthropic)>helped me identify that it was serious
enough that I needed to be treated,

4415
04:26:24.335 --> 04:26:25.535
<v Logan Graham (Anthropic)>very, very quickly.

4416
04:26:26.095 --> 04:26:30.975
<v Logan Graham (Anthropic)>Claude triaged a very large part, of it
for me, but what was most important was I

4417
04:26:30.975 --> 04:26:36.149
<v Logan Graham (Anthropic)>could use that to get to the hospital very
fast. It's very clear that both can be,

4418
04:26:36.149 --> 04:26:39.510
<v Logan Graham (Anthropic)>you know, transformative and powerful
together, certainly in my case and I think

4419
04:26:39.510 --> 04:26:40.550
<v Logan Graham (Anthropic)>countless others.

4420
04:26:40.789 --> 04:26:43.989
<v Logan Graham (Anthropic)>But this is also one that's closest to
my heart that we need to be very careful

4421
04:26:43.989 --> 04:26:44.230
<v Logan Graham (Anthropic)>about.

4422
04:26:49.936 --> 04:26:50.815
<v CM Carmen De La Rosa>Thank you.

4423
04:26:50.976 --> 04:26:56.095
<v CM Carmen De La Rosa>Thank thank you all so much. Up next, we
have council member Joseph followed by p

4424
04:26:56.095 --> 04:26:57.855
<v CM Carmen De La Rosa>Sanchez, followed by Wong...

4425
04:26:58.335 --> 04:27:00.176
<v CM Carmen De La Rosa>Phil Wong, and then Shuang.

4426
04:27:00.976 --> 04:27:05.029
<v CM Rita Joseph>Thank you, chair. Question... Two
questions. Several of you sell AI products

4427
04:27:05.029 --> 04:27:08.630
<v CM Rita Joseph>directly to colleges and universities.
When student use those products,

4428
04:27:08.870 --> 04:27:12.870
<v CM Rita Joseph>is anything they enter used to train your
models? And beyond training, who can

4429
04:27:12.870 --> 04:27:14.311
<v CM Rita Joseph>access those conversation,

4430
04:27:14.550 --> 04:27:18.069
<v CM Rita Joseph>and what limit do you place on how
colleges can use them, for example, for

4431
04:27:18.069 --> 04:27:18.550
<v CM Rita Joseph>discipline,

4432
04:27:18.875 --> 04:27:20.955
<v CM Rita Joseph>academic integrity, investigation,

4433
04:27:21.275 --> 04:27:22.635
<v CM Rita Joseph>or student profiling?

4434
04:27:22.875 --> 04:27:27.515
<v CM Rita Joseph>The next one is given that AI is already
reshaping entry level work as we know, as

4435
04:27:27.515 --> 04:27:32.475
<v CM Rita Joseph>is shown in this new Center for Urban Future
report, what responsibility do you have

4436
04:27:32.475 --> 04:27:35.301
<v CM Rita Joseph>to ensure these tools help
students build useful

4437
04:27:35.380 --> 04:27:36.820
<v CM Rita Joseph>AI augmentative

4438
04:27:36.820 --> 04:27:41.540
<v CM Rita Joseph>skills rather than automated away the very
task to which young people learn? What

4439
04:27:41.540 --> 04:27:46.020
<v CM Rita Joseph>responsibility do your companies have
addressing these early career impacts, and how

4440
04:27:46.020 --> 04:27:46.820
<v CM Rita Joseph>can government help?

4441
04:27:51.915 --> 04:27:54.875
<v Morgan Dwyer (OpenAI)>So to your first question on privacy,

4442
04:27:55.195 --> 04:27:58.635
<v Morgan Dwyer (OpenAI)>we believe that privacy is
central to our mission.

4443
04:27:59.035 --> 04:28:02.075
<v Morgan Dwyer (OpenAI)>We give users control over their data

4444
04:28:02.790 --> 04:28:03.591
<v Morgan Dwyer (OpenAI)>that,

4445
04:28:04.070 --> 04:28:07.431
<v Morgan Dwyer (OpenAI)>includes the decision
to opt out of training.

4446
04:28:07.830 --> 04:28:09.270
<v Morgan Dwyer (OpenAI)>But all users,

4447
04:28:09.511 --> 04:28:11.270
<v Morgan Dwyer (OpenAI)>have their data protected,

4448
04:28:11.351 --> 04:28:12.871
<v Morgan Dwyer (OpenAI)>and private by default.

4449
04:28:13.351 --> 04:28:14.711
<v Morgan Dwyer (OpenAI)>So it is separated

4450
04:28:14.790 --> 04:28:15.431
<v Morgan Dwyer (OpenAI)>from,

4451
04:28:15.670 --> 04:28:16.391
<v Morgan Dwyer (OpenAI)>their names.

4452
04:28:17.385 --> 04:28:22.105
<v Morgan Dwyer (OpenAI)>And, again, they have the option to
opt out of training and to request

4453
04:28:22.185 --> 04:28:22.825
<v Morgan Dwyer (OpenAI)>that

4454
04:28:23.385 --> 04:28:25.065
<v Morgan Dwyer (OpenAI)>their data is deleted.

4455
04:28:25.785 --> 04:28:31.225
<v Morgan Dwyer (OpenAI)>On the particular question of colleges
and how ChatGeBT is used in colleges and

4456
04:28:31.225 --> 04:28:31.705
<v Morgan Dwyer (OpenAI)>education,

4457
04:28:32.450 --> 04:28:35.649
<v Morgan Dwyer (OpenAI)>I would like to get back to you on
that. I don't have an accurate,

4458
04:28:35.890 --> 04:28:39.970
<v Morgan Dwyer (OpenAI)>I I don't I don't know the answer, and
so I'd I'd like to be able to follow-up,

4459
04:28:40.130 --> 04:28:41.810
<v Morgan Dwyer (OpenAI)>because we do believe,

4460
04:28:42.210 --> 04:28:42.931
<v Morgan Dwyer (OpenAI)>that

4461
04:28:43.250 --> 04:28:49.715
<v Morgan Dwyer (OpenAI)>education is critical. AI literacy is
critical. We've supported numerous bills, at

4462
04:28:49.715 --> 04:28:53.875
<v Morgan Dwyer (OpenAI)>both the state and federal
levels supporting AI literacy.

4463
04:28:54.274 --> 04:28:59.155
<v Morgan Dwyer (OpenAI)>That is both training for teachers to
ensure that they can lead adoption,

4464
04:28:59.715 --> 04:29:01.555
<v Morgan Dwyer (OpenAI)>in schools as well as

4465
04:29:03.230 --> 04:29:08.910
<v Morgan Dwyer (OpenAI)>improvements in curricula to make sure that
students learn how to use AI safely as

4466
04:29:08.910 --> 04:29:11.230
<v Morgan Dwyer (OpenAI)>well as learn core subjects,

4467
04:29:11.470 --> 04:29:15.950
<v Morgan Dwyer (OpenAI)>that are critical to be a...
Being able to use AI safely

4468
04:29:16.485 --> 04:29:17.925
<v Morgan Dwyer (OpenAI)>and responsibly.

4469
04:29:18.564 --> 04:29:21.525
<v Morgan Dwyer (OpenAI)>To your question about jobs,

4470
04:29:22.564 --> 04:29:27.765
<v Morgan Dwyer (OpenAI)>we think one of the most important things
that we and policymakers can do right now

4471
04:29:27.765 --> 04:29:29.045
<v Morgan Dwyer (OpenAI)>is to continue

4472
04:29:29.400 --> 04:29:33.721
<v Morgan Dwyer (OpenAI)>to collect data on the question
of AI impacts to jobs,

4473
04:29:34.120 --> 04:29:36.040
<v Morgan Dwyer (OpenAI)>and make that data public.

4474
04:29:36.601 --> 04:29:41.400
<v Morgan Dwyer (OpenAI)>We make our assessments of
how people are using ChatGPT,

4475
04:29:41.561 --> 04:29:43.320
<v Morgan Dwyer (OpenAI)>at work public. We share that.

4476
04:29:43.995 --> 04:29:46.795
<v Morgan Dwyer (OpenAI)>And we also have worked with policymakers,

4477
04:29:46.955 --> 04:29:48.395
<v Morgan Dwyer (OpenAI)>including supporting,

4478
04:29:48.715 --> 04:29:51.515
<v Morgan Dwyer (OpenAI)>the Bipartisan Workforce Transparency Act,

4479
04:29:51.835 --> 04:29:55.274
<v Morgan Dwyer (OpenAI)>to make, or to create voluntary pathways,

4480
04:29:55.354 --> 04:29:57.835
<v Morgan Dwyer (OpenAI)>for employers to report on,

4481
04:29:58.234 --> 04:29:59.034
<v Morgan Dwyer (OpenAI)>job impacts.

4482
04:30:01.551 --> 04:30:04.511
<v Shane Cahill (Meta)>Council member, thank you very
much for for these questions.

4483
04:30:04.990 --> 04:30:06.431
<v Shane Cahill (Meta)>In relation to

4484
04:30:06.910 --> 04:30:08.511
<v Shane Cahill (Meta)>colleges and universities,

4485
04:30:08.910 --> 04:30:11.471
<v Shane Cahill (Meta)>I would like to to have my...

4486
04:30:11.790 --> 04:30:14.830
<v Shane Cahill (Meta)>The the specialists at at Meta
who who work in that area

4487
04:30:15.415 --> 04:30:20.295
<v Shane Cahill (Meta)>connect with you on the details. But
specifically in relation to privacy and and

4488
04:30:20.295 --> 04:30:23.095
<v Shane Cahill (Meta)>data, this is absolutely core

4489
04:30:23.095 --> 04:30:25.095
<v Shane Cahill (Meta)>to everything that we do at Meta.

4490
04:30:25.415 --> 04:30:30.950
<v Shane Cahill (Meta)>For example, our Muse product, you can
update all the use of data for training and

4491
04:30:30.950 --> 04:30:31.910
<v Shane Cahill (Meta)>you can ask

4492
04:30:32.870 --> 04:30:34.791
<v Shane Cahill (Meta)>Muse to to forget any

4493
04:30:35.190 --> 04:30:40.311
<v Shane Cahill (Meta)>information that you provide to it. And,
of course, all of the data is is is within

4494
04:30:40.311 --> 04:30:42.230
<v Shane Cahill (Meta)>your control in terms of
what you provide to it.

4495
04:30:44.035 --> 04:30:46.275
<v Shane Cahill (Meta)>I... On the AI literacy

4496
04:30:46.515 --> 04:30:47.155
<v Shane Cahill (Meta)>point,

4497
04:30:47.715 --> 04:30:48.755
<v Shane Cahill (Meta)>it's so important.

4498
04:30:49.235 --> 04:30:54.195
<v Shane Cahill (Meta)>I actually came across a really great
example from New York City, and I I wanted to

4499
04:30:54.195 --> 04:30:57.700
<v Shane Cahill (Meta)>to call it out because I thought it was
great. It was called building minds,

4500
04:30:57.700 --> 04:31:01.860
<v Shane Cahill (Meta)>building communities, their digital empowerment
by design work. And I thought it was

4501
04:31:01.860 --> 04:31:03.541
<v Shane Cahill (Meta)>a really, really great example

4502
04:31:03.940 --> 04:31:06.020
<v Shane Cahill (Meta)>of the use of AI technologies

4503
04:31:06.020 --> 04:31:07.860
<v Shane Cahill (Meta)>for AI literacy

4504
04:31:08.020 --> 04:31:10.101
<v Shane Cahill (Meta)>and kind of a a a flywheel

4505
04:31:10.335 --> 04:31:12.015
<v Shane Cahill (Meta)>in terms of benefits there.

4506
04:31:13.055 --> 04:31:14.415
<v Shane Cahill (Meta)>In terms of

4507
04:31:14.654 --> 04:31:16.255
<v Shane Cahill (Meta)>the entry level point,

4508
04:31:16.895 --> 04:31:17.774
<v Shane Cahill (Meta)>absolutely

4509
04:31:17.774 --> 04:31:20.735
<v Shane Cahill (Meta)>understand the the concern
in relation to that.

4510
04:31:21.055 --> 04:31:25.040
<v Shane Cahill (Meta)>At Meta, we believe in placing
personal superintelligence

4511
04:31:25.040 --> 04:31:26.561
<v Shane Cahill (Meta)>into the hands of everyone

4512
04:31:26.960 --> 04:31:28.480
<v Shane Cahill (Meta)>and for invention,

4513
04:31:28.560 --> 04:31:29.520
<v Shane Cahill (Meta)>not automation,

4514
04:31:29.680 --> 04:31:30.960
<v Shane Cahill (Meta)>and for augmentation,

4515
04:31:31.040 --> 04:31:32.000
<v Shane Cahill (Meta)>not replacement.

4516
04:31:32.160 --> 04:31:35.040
<v Shane Cahill (Meta)>And we really believe that this
will have a transformative

4517
04:31:35.854 --> 04:31:37.695
<v Shane Cahill (Meta)>impact in terms of

4518
04:31:38.095 --> 04:31:39.855
<v Shane Cahill (Meta)>enabling and empowering people

4519
04:31:40.015 --> 04:31:41.935
<v Shane Cahill (Meta)>to pursue their own

4520
04:31:43.694 --> 04:31:46.335
<v Shane Cahill (Meta)>their own values and
direct AI towards those.

4521
04:31:50.040 --> 04:31:54.360
<v Alice Friend (Google)>Yes, ma'am. Similar to my colleagues, we
believe users should have control over

4522
04:31:54.360 --> 04:31:56.200
<v Alice Friend (Google)>their own data. We're

4523
04:31:56.200 --> 04:32:00.520
<v CM Carmen De La Rosa>not going to have outbursts from the
chamber or you will be removed. You decide.

4524
04:32:01.075 --> 04:32:02.436
<v CM Carmen De La Rosa>Sorry. Go ahead.

4525
04:32:04.035 --> 04:32:10.115
<v Alice Friend (Google)>As I was saying, users have control over
their own data, across our AI surfaces

4526
04:32:12.675 --> 04:32:13.955
<v Alice Friend (Google)>and can

4527
04:32:14.200 --> 04:32:16.601
<v Alice Friend (Google)>opt in or opt out of

4528
04:32:17.720 --> 04:32:20.681
<v Alice Friend (Google)>their data being used to
train our our models.

4529
04:32:22.200 --> 04:32:25.721
<v Alice Friend (Google)>In terms of AI use in colleges,

4530
04:32:26.040 --> 04:32:26.920
<v Alice Friend (Google)>as I said earlier,

4531
04:32:27.716 --> 04:32:31.796
<v Alice Friend (Google)>we have tools for educators and
administrators to take the lead,

4532
04:32:32.195 --> 04:32:35.716
<v Alice Friend (Google)>in how AI is used in the
classroom and and used for,

4533
04:32:36.275 --> 04:32:38.275
<v Alice Friend (Google)>students. We think that colleges,

4534
04:32:38.756 --> 04:32:41.955
<v Alice Friend (Google)>should be able to adapt to the
needs of their student populations.

4535
04:32:43.109 --> 04:32:45.989
<v Alice Friend (Google)>I I, like Shane, was looking

4536
04:32:45.989 --> 04:32:46.870
<v Alice Friend (Google)>into

4537
04:32:46.949 --> 04:32:47.990
<v Alice Friend (Google)>how different

4538
04:32:48.630 --> 04:32:51.030
<v Alice Friend (Google)>colleges and universities in New York City

4539
04:32:51.590 --> 04:32:55.430
<v Alice Friend (Google)>use AI and saw that NYU, for
example, has a very helpful page

4540
04:32:55.825 --> 04:32:59.505
<v Alice Friend (Google)>where they go through all of the AI tools
they make available to students and give

4541
04:32:59.505 --> 04:33:00.625
<v Alice Friend (Google)>them resources

4542
04:33:01.265 --> 04:33:03.665
<v Alice Friend (Google)>for learning both how to use the tools,

4543
04:33:04.385 --> 04:33:05.824
<v Alice Friend (Google)>and then also,

4544
04:33:06.145 --> 04:33:09.984
<v Alice Friend (Google)>training them in the kinds of things they
can do with those tools to further their

4545
04:33:09.984 --> 04:33:10.385
<v Alice Friend (Google)>education.

4546
04:33:10.830 --> 04:33:14.270
<v Alice Friend (Google)>I would flag that they use
Gemini Notebook, which is a

4547
04:33:14.750 --> 04:33:16.670
<v Alice Friend (Google)>study guide creation

4548
04:33:16.670 --> 04:33:17.390
<v Alice Friend (Google)>tool.

4549
04:33:18.190 --> 04:33:23.310
<v Alice Friend (Google)>So NYU is a good example, I think, of of
a college thinking carefully about how AI

4550
04:33:23.310 --> 04:33:24.910
<v Alice Friend (Google)>can benefit their student population.

4551
04:33:26.326 --> 04:33:30.885
<v Alice Friend (Google)>And as to the impact on early careers, we
think this is also an incredibly important

4552
04:33:30.885 --> 04:33:35.045
<v Alice Friend (Google)>area of of research. It's part of understanding
the broader impacts on the economy

4553
04:33:35.045 --> 04:33:35.926
<v Alice Friend (Google)>and labor.

4554
04:33:36.326 --> 04:33:40.246
<v Alice Friend (Google)>We recently released our own AI
and Economy Atlas, which also

4555
04:33:40.781 --> 04:33:42.060
<v Alice Friend (Google)>releases

4556
04:33:42.060 --> 04:33:45.502
<v Alice Friend (Google)>data about real world uses of AI,

4557
04:33:45.900 --> 04:33:48.140
<v Alice Friend (Google)>in the economy and at jobs.

4558
04:33:48.460 --> 04:33:51.422
<v Alice Friend (Google)>And we're hoping that continuing
to add to that dataset,

4559
04:33:51.740 --> 04:33:53.275
<v Alice Friend (Google)>will help economists

4560
04:33:53.275 --> 04:33:56.875
<v Alice Friend (Google)>study and then develop
policy recommendations

4561
04:33:56.875 --> 04:33:57.436
<v Alice Friend (Google)>for,

4562
04:33:57.914 --> 04:34:00.234
<v Alice Friend (Google)>managing the labor transitions of AI.

4563
04:34:01.836 --> 04:34:03.596
<v Logan Graham (Anthropic)>And at Anthropic,

4564
04:34:03.914 --> 04:34:05.676
<v Logan Graham (Anthropic)>maintaining user privacy is

4565
04:34:05.836 --> 04:34:08.395
<v Logan Graham (Anthropic)>very core to our commitment
on safety and security.

4566
04:34:09.030 --> 04:34:13.670
<v Logan Graham (Anthropic)>Users can choose whether or not they want
their data to be used or contribute to

4567
04:34:13.670 --> 04:34:15.430
<v Logan Graham (Anthropic)>training a model, for example.

4568
04:34:15.670 --> 04:34:20.469
<v Logan Graham (Anthropic)>But in addition to that, we tried to
pioneer the science of privacy preserving

4569
04:34:20.790 --> 04:34:25.227
<v Logan Graham (Anthropic)>safety maintenance when you need to ensure
that users stay safe on the platform.

4570
04:34:26.105 --> 04:34:29.625
<v Logan Graham (Anthropic)>To the point of education, while it's
not my area, here's what I understand.

4571
04:34:30.266 --> 04:34:31.945
<v Logan Graham (Anthropic)>We partner with universities.

4572
04:34:32.586 --> 04:34:36.300
<v Logan Graham (Anthropic)>We're very interested in this question.
There's a nifty feature in part of our

4573
04:34:36.300 --> 04:34:38.621
<v Logan Graham (Anthropic)>called learning mode, which allows,

4574
04:34:38.861 --> 04:34:43.102
<v Logan Graham (Anthropic)>a user or student user to interact with
the model in a more discursive way that can

4575
04:34:43.102 --> 04:34:44.060
<v Logan Graham (Anthropic)>help them think.

4576
04:34:44.701 --> 04:34:47.900
<v Logan Graham (Anthropic)>But in addition to this, the
impacts are really important.

4577
04:34:48.460 --> 04:34:52.584
<v Logan Graham (Anthropic)>I think we are relatively
well known for our work on,

4578
04:34:52.824 --> 04:34:54.264
<v Logan Graham (Anthropic)>studying the economic,

4579
04:34:54.664 --> 04:34:57.465
<v Logan Graham (Anthropic)>and other behavioral impacts of models.

4580
04:34:57.465 --> 04:35:00.904
<v Logan Graham (Anthropic)>We have a comprehensive, and I think we
were the first to release this, economic

4581
04:35:00.904 --> 04:35:05.520
<v Logan Graham (Anthropic)>index which details this. But in addition,
we've done research into how students are

4582
04:35:05.520 --> 04:35:06.959
<v Logan Graham (Anthropic)>are using models,

4583
04:35:07.360 --> 04:35:11.440
<v Logan Graham (Anthropic)>because we want to figure out how to
encourage the use of models to complement

4584
04:35:11.520 --> 04:35:14.160
<v Logan Graham (Anthropic)>the developments and skill
development of students.

4585
04:35:15.440 --> 04:35:16.240
<v CM Carmen De La Rosa>Thank you.

4586
04:35:16.879 --> 04:35:19.600
<v CM Carmen De La Rosa>Up next is council member P Sanchez,

4587
04:35:19.680 --> 04:35:20.719
<v CM Carmen De La Rosa>council member

4588
04:35:20.895 --> 04:35:26.096
<v CM Carmen De La Rosa>Phil Wong, and council member Susan
Shuang. I will remind my colleagues to

4589
04:35:26.254 --> 04:35:28.096
<v CM Carmen De La Rosa>ask your questions upfront.

4590
04:35:28.574 --> 04:35:32.574
<v CM Carmen De La Rosa>And if you have questions that are specific
for some of the members of the panel,

4591
04:35:32.574 --> 04:35:34.494
<v CM Carmen De La Rosa>not all four, specify.

4592
04:35:34.494 --> 04:35:36.094
<v CM Carmen De La Rosa>Thank you, council member Sanchez.

4593
04:35:36.560 --> 04:35:40.880
<v CM Pierina Ana Sanchez>Thank you, chair. My first question is
in connection to heartfelt testimony of

4594
04:35:40.880 --> 04:35:43.600
<v CM Pierina Ana Sanchez>Leila's Turner Scott before congresswoman

4595
04:35:43.600 --> 04:35:45.361
<v CM Pierina Ana Sanchez>Jaya Paul last month.

4596
04:35:45.680 --> 04:35:48.800
<v CM Pierina Ana Sanchez>Miss Turner Scott's son died
following chats with OpenAI

4597
04:35:48.800 --> 04:35:50.720
<v CM Pierina Ana Sanchez>that instructed him that incompatible

4598
04:35:50.720 --> 04:35:51.920
<v CM Pierina Ana Sanchez>drugs were compatible.

4599
04:35:52.385 --> 04:35:56.945
<v CM Pierina Ana Sanchez>She highlighted that throughout the review
of logs, ChatGPT responded to her son

4600
04:35:56.945 --> 04:36:00.945
<v CM Pierina Ana Sanchez>with phrases like, I've got your back, you
can talk to me about anything, and even I

4601
04:36:00.945 --> 04:36:02.785
<v CM Pierina Ana Sanchez>love you with heart emojis.

4602
04:36:03.025 --> 04:36:07.880
<v CM Pierina Ana Sanchez>She she stated and it seems true that the
goal of ChatGPT was not to keep her son

4603
04:36:07.880 --> 04:36:12.121
<v CM Pierina Ana Sanchez>safe. It was only to keep him using the
product. What updates have your companies

4604
04:36:12.121 --> 04:36:15.320
<v CM Pierina Ana Sanchez>made to their models to engage
children and prevent addiction?

4605
04:36:15.400 --> 04:36:18.600
<v CM Pierina Ana Sanchez>Are you limiting your agents from giving
medical and health advice without a

4606
04:36:18.600 --> 04:36:19.240
<v CM Pierina Ana Sanchez>license?

4607
04:36:19.400 --> 04:36:23.205
<v CM Pierina Ana Sanchez>And my second question is regarding
the fact that members of the public and

4608
04:36:23.205 --> 04:36:26.725
<v CM Pierina Ana Sanchez>enterprise enterprise clients are sharing
more and more information with your

4609
04:36:26.725 --> 04:36:30.164
<v CM Pierina Ana Sanchez>companies. You have access to
copyrighted information information

4610
04:36:30.164 --> 04:36:31.445
<v CM Pierina Ana Sanchez>publicly available.

4611
04:36:31.686 --> 04:36:32.885
<v CM Pierina Ana Sanchez>How do your companies...

4612
04:36:34.201 --> 04:36:37.880
<v CM Pierina Ana Sanchez>Excuse me. How do your products currently
use information gathered from users in an

4613
04:36:37.880 --> 04:36:39.162
<v CM Pierina Ana Sanchez>earlier exchanges

4614
04:36:39.320 --> 04:36:43.640
<v CM Pierina Ana Sanchez>chair if I may? Several of you stated that
users can opt out of sharing your data.

4615
04:36:43.640 --> 04:36:45.480
<v CM Pierina Ana Sanchez>What is the default option?

4616
04:36:45.880 --> 04:36:49.533
<v CM Pierina Ana Sanchez>And how are trade secrets and other
confidential business or government information

4617
04:36:49.533 --> 04:36:54.094
<v CM Pierina Ana Sanchez>from enterprise clients being used to trade
AI? Who have you paid for the valuable

4618
04:36:54.094 --> 04:36:56.254
<v CM Pierina Ana Sanchez>data that you are using
to train your models?

4619
04:37:01.720 --> 04:37:04.521
<v Morgan Dwyer (OpenAI)>So I wanna start off by acknowledging,

4620
04:37:04.840 --> 04:37:07.480
<v Morgan Dwyer (OpenAI)>the tragic incident that you referenced.

4621
04:37:08.041 --> 04:37:09.320
<v Morgan Dwyer (OpenAI)>Any incidents

4622
04:37:09.320 --> 04:37:14.840
<v Morgan Dwyer (OpenAI)>of harm that results from our product or
any other product is tragic, and we take it

4623
04:37:14.840 --> 04:37:15.400
<v Morgan Dwyer (OpenAI)>seriously.

4624
04:37:17.266 --> 04:37:20.705
<v Morgan Dwyer (OpenAI)>I'd like to tell you
about ChatGPT for teens,

4625
04:37:20.945 --> 04:37:23.426
<v Morgan Dwyer (OpenAI)>which is... Was was recently released,

4626
04:37:23.666 --> 04:37:26.225
<v Morgan Dwyer (OpenAI)>to focus specifically on users

4627
04:37:26.385 --> 04:37:28.785
<v Morgan Dwyer (OpenAI)>ages 13 to 18,

4628
04:37:29.346 --> 04:37:31.904
<v Morgan Dwyer (OpenAI)>and is designed to be safe by default.

4629
04:37:32.600 --> 04:37:33.799
<v Morgan Dwyer (OpenAI)>That includes,

4630
04:37:34.040 --> 04:37:37.639
<v Morgan Dwyer (OpenAI)>safety protections on the type
of content that is available,

4631
04:37:37.959 --> 04:37:38.600
<v Morgan Dwyer (OpenAI)>limiting

4632
04:37:39.479 --> 04:37:43.480
<v Morgan Dwyer (OpenAI)>content that might be related
to suicide or self harm,

4633
04:37:44.120 --> 04:37:45.479
<v Morgan Dwyer (OpenAI)>eating disorders,

4634
04:37:45.479 --> 04:37:46.440
<v Morgan Dwyer (OpenAI)>violence,

4635
04:37:46.760 --> 04:37:48.416
<v Morgan Dwyer (OpenAI)>or graphic

4636
04:37:48.416 --> 04:37:50.656
<v Morgan Dwyer (OpenAI)>graphic and... Or sexual content.

4637
04:37:51.295 --> 04:37:53.936
<v Morgan Dwyer (OpenAI)>We also include take a break reminders,

4638
04:37:54.096 --> 04:37:55.855
<v Morgan Dwyer (OpenAI)>to remind teens,

4639
04:37:56.016 --> 04:37:58.336
<v Morgan Dwyer (OpenAI)>to stop using ChatGPT.

4640
04:37:58.975 --> 04:38:02.895
<v Morgan Dwyer (OpenAI)>And we have, parental controls,
optional parental controls

4641
04:38:03.090 --> 04:38:03.810
<v Morgan Dwyer (OpenAI)>that,

4642
04:38:04.690 --> 04:38:06.129
<v Morgan Dwyer (OpenAI)>allow parents

4643
04:38:06.129 --> 04:38:06.770
<v Morgan Dwyer (OpenAI)>to,

4644
04:38:08.129 --> 04:38:12.130
<v Morgan Dwyer (OpenAI)>add additional layers of safety
onto their teen's experience.

4645
04:38:13.010 --> 04:38:15.730
<v Morgan Dwyer (OpenAI)>One thing that I will highlight here is,

4646
04:38:16.129 --> 04:38:18.450
<v Morgan Dwyer (OpenAI)>if a teen indicates

4647
04:38:18.975 --> 04:38:23.615
<v Morgan Dwyer (OpenAI)>signs of suicide or self harm, we have
a feature that allows the parent to be

4648
04:38:23.615 --> 04:38:24.494
<v Morgan Dwyer (OpenAI)>notified.

4649
04:38:26.975 --> 04:38:30.414
<v Morgan Dwyer (OpenAI)>On your question of privacy and data,

4650
04:38:30.975 --> 04:38:32.814
<v Morgan Dwyer (OpenAI)>as I and my colleagues,

4651
04:38:32.975 --> 04:38:36.820
<v Morgan Dwyer (OpenAI)>have shared, we take
privacy privacy seriously,

4652
04:38:37.219 --> 04:38:41.780
<v Morgan Dwyer (OpenAI)>and our core value is that users
should have control over their data.

4653
04:38:43.379 --> 04:38:49.620
<v Morgan Dwyer (OpenAI)>And I'm I'm happy to follow-up on on
specific details of how those controls work

4654
04:38:50.557 --> 04:38:52.877
<v Morgan Dwyer (OpenAI)>at a later time.... Default option?

4655
04:38:55.035 --> 04:38:59.596
<v CM Pierina Ana Sanchez>I'm sorry. I didn't hear you. What is the
default option with respect to who owns

4656
04:38:59.596 --> 04:39:00.717
<v CM Pierina Ana Sanchez>the data and how it is used?

4657
04:39:02.650 --> 04:39:07.690
<v Morgan Dwyer (OpenAI)>So users have control over their data.
Everything is privacy...... Default option.

4658
04:39:07.770 --> 04:39:08.170
<v Morgan Dwyer (OpenAI)>Everything

4659
04:39:09.049 --> 04:39:11.770
<v Morgan Dwyer (OpenAI)>is privacy protected by default,

4660
04:39:11.850 --> 04:39:16.826
<v Morgan Dwyer (OpenAI)>and users have the option to turn off their
data for training or request that it be

4661
04:39:16.826 --> 04:39:17.545
<v Morgan Dwyer (OpenAI)>deleted?

4662
04:39:23.625 --> 04:39:25.467
<v Shane Cahill (Meta)>Council member, first

4663
04:39:25.945 --> 04:39:26.586
<v Shane Cahill (Meta)>off,

4664
04:39:26.826 --> 04:39:29.865
<v Shane Cahill (Meta)>I just want to acknowledge
the terrible tragedy

4665
04:39:30.025 --> 04:39:32.025
<v Shane Cahill (Meta)>that was

4666
04:39:32.959 --> 04:39:35.201
<v Shane Cahill (Meta)>a terrible tragedy in relation to

4667
04:39:35.600 --> 04:39:37.280
<v Shane Cahill (Meta)>the Turner Scott family

4668
04:39:37.600 --> 04:39:41.201
<v Shane Cahill (Meta)>and how distressing and
upsetting that is to to hear

4669
04:39:41.920 --> 04:39:42.560
<v Shane Cahill (Meta)>and,

4670
04:39:42.720 --> 04:39:44.800
<v Shane Cahill (Meta)>you know, my sympathies and condolences

4671
04:39:44.800 --> 04:39:45.600
<v Shane Cahill (Meta)>to that family.

4672
04:39:47.635 --> 04:39:50.994
<v Shane Cahill (Meta)>In relation to how Meta approaches

4673
04:39:53.314 --> 04:39:54.914
<v Shane Cahill (Meta)>minors and young people,

4674
04:39:56.914 --> 04:39:58.914
<v Shane Cahill (Meta)>creating a safe and productive

4675
04:39:59.314 --> 04:40:01.400
<v Shane Cahill (Meta)>experience is absolutely imperative

4676
04:40:01.400 --> 04:40:03.240
<v Shane Cahill (Meta)>to Meta, and we're committed

4677
04:40:03.560 --> 04:40:07.320
<v Shane Cahill (Meta)>to providing a safe and responsible
AI, especially for teams.

4678
04:40:07.640 --> 04:40:11.880
<v Shane Cahill (Meta)>And we have strong policies and
safeguards in place in relation to that.

4679
04:40:12.440 --> 04:40:13.400
<v Shane Cahill (Meta)>For example,

4680
04:40:13.800 --> 04:40:14.520
<v Shane Cahill (Meta)>by default,

4681
04:40:16.414 --> 04:40:17.455
<v Shane Cahill (Meta)>minors

4682
04:40:18.574 --> 04:40:21.135
<v Shane Cahill (Meta)>only have age appropriate interactions,

4683
04:40:21.135 --> 04:40:25.455
<v Shane Cahill (Meta)>meaning that content would
not be out of place on a 13

4684
04:40:25.455 --> 04:40:26.734
<v Shane Cahill (Meta)>movie, for example.

4685
04:40:27.535 --> 04:40:28.654
<v Shane Cahill (Meta)>We also prohibit

4686
04:40:31.150 --> 04:40:32.590
<v Shane Cahill (Meta)>AI offerings

4687
04:40:32.830 --> 04:40:34.190
<v Shane Cahill (Meta)>from providing

4688
04:40:34.270 --> 04:40:37.230
<v Shane Cahill (Meta)>licensed professional information
like medical information

4689
04:40:37.470 --> 04:40:40.510
<v Shane Cahill (Meta)>and direct people towards
appropriate resources.

4690
04:40:42.885 --> 04:40:44.965
<v Shane Cahill (Meta)>In relation to training data,

4691
04:40:46.805 --> 04:40:47.525
<v Shane Cahill (Meta)>we

4692
04:40:47.605 --> 04:40:48.885
<v Shane Cahill (Meta)>take creativity,

4693
04:40:48.885 --> 04:40:49.525
<v Shane Cahill (Meta)>copyright,

4694
04:40:49.766 --> 04:40:54.166
<v Shane Cahill (Meta)>and IP rights seriously and believe our
practices are consistent with the law.

4695
04:40:55.980 --> 04:40:57.422
<v Shane Cahill (Meta)>We train on

4696
04:40:58.541 --> 04:41:00.380
<v Shane Cahill (Meta)>information from the Internet,

4697
04:41:00.861 --> 04:41:03.021
<v Shane Cahill (Meta)>information licensed from third parties,

4698
04:41:03.660 --> 04:41:08.215
<v Shane Cahill (Meta)>and information shared across our services.
And we provide all of the information in

4699
04:41:08.215 --> 04:41:11.816
<v Shane Cahill (Meta)>relation to how we use that
data in our privacy center.

4700
04:41:12.855 --> 04:41:14.695
<v Shane Cahill (Meta)>In relation to to Muse,

4701
04:41:15.496 --> 04:41:16.215
<v Shane Cahill (Meta)>the...

4702
04:41:16.535 --> 04:41:17.576
<v Shane Cahill (Meta)>As I mentioned,

4703
04:41:17.736 --> 04:41:19.016
<v Shane Cahill (Meta)>I think a little earlier,

4704
04:41:19.496 --> 04:41:25.060
<v Shane Cahill (Meta)>in individuals interacting with Muse can
opt out of their interactions for training.

4705
04:41:29.540 --> 04:41:34.660
<v Alice Friend (Google)>Ma'am, I wanna highlight a few protections
that we have in place with kids in mind

4706
04:41:34.660 --> 04:41:37.540
<v Alice Friend (Google)>in particular. One is
for what we call persona,

4707
04:41:38.455 --> 04:41:41.016
<v Alice Friend (Google)>which prevents claims of sentience

4708
04:41:41.016 --> 04:41:45.895
<v Alice Friend (Google)>or simulating relationships and role play
that helps prevent kids from thinking

4709
04:41:45.895 --> 04:41:49.494
<v Alice Friend (Google)>they're talking to a person
and developing a relationship.

4710
04:41:50.215 --> 04:41:53.680
<v Alice Friend (Google)>We also prevent sexually explicit outputs

4711
04:41:53.680 --> 04:41:54.398
<v Alice Friend (Google)>from

4712
04:41:54.719 --> 04:41:56.080
<v Alice Friend (Google)>chats with kids.

4713
04:41:56.639 --> 04:42:02.080
<v Alice Friend (Google)>And we also have for all ages
protocols on suicide and self harm

4714
04:42:02.479 --> 04:42:07.200
<v Alice Friend (Google)>where we refer users who are expressing
suicidal ideation or other types of self

4715
04:42:07.200 --> 04:42:09.334
<v Alice Friend (Google)>harm to crisis resources.

4716
04:42:09.975 --> 04:42:14.295
<v Alice Friend (Google)>That approach draws from our deep
experience fielding such queries on search.

4717
04:42:15.334 --> 04:42:20.615
<v Alice Friend (Google)>When it comes to training and the default
question you asked, for those over 18, it

4718
04:42:20.615 --> 04:42:23.310
<v Alice Friend (Google)>is default on to use

4719
04:42:23.310 --> 04:42:26.271
<v Alice Friend (Google)>chat data to train our models to improve,

4720
04:42:26.510 --> 04:42:31.550
<v Alice Friend (Google)>but all users have control over the ability
to turn this feature off or to control

4721
04:42:31.550 --> 04:42:34.670
<v Alice Friend (Google)>the length of time such
information is kept.

4722
04:42:35.230 --> 04:42:38.590
<v Alice Friend (Google)>For users between 13 and
17, it's default off.

4723
04:42:39.365 --> 04:42:42.965
<v Alice Friend (Google)>Teens can choose to turn
it on if they wish to

4724
04:42:43.365 --> 04:42:46.645
<v Alice Friend (Google)>train a model through our
personal intelligence program.

4725
04:42:46.805 --> 04:42:48.166
<v Alice Friend (Google)>And for 13,

4726
04:42:48.244 --> 04:42:54.484
<v Alice Friend (Google)>it is default off as well, and we never
use data for under 13 to train our models.

4727
04:42:57.380 --> 04:42:59.621
<v Logan Graham (Anthropic) [?]>And Thank you. At Anthropic,

4728
04:42:59.701 --> 04:43:04.740
<v Logan Graham (Anthropic)>we, have limited the cloud
platform to those only 18.

4729
04:43:05.380 --> 04:43:07.541
<v Logan Graham (Anthropic)>We're concerned about child well-being.

4730
04:43:08.195 --> 04:43:14.756
<v Logan Graham (Anthropic)>We monitor and detect for any users that
might be 18 or under. And when that flags,

4731
04:43:14.756 --> 04:43:18.596
<v Logan Graham (Anthropic)>we investigate to make sure that we
could remove their access if so.

4732
04:43:20.120 --> 04:43:23.719
<v Logan Graham (Anthropic)>On training data, I will
have to get back to you with

4733
04:43:24.120 --> 04:43:27.320
<v Logan Graham (Anthropic)>more details as it's not
my role. My understanding

4734
04:43:27.398 --> 04:43:30.520
<v Logan Graham (Anthropic)>is that all users must select which option

4735
04:43:31.158 --> 04:43:32.040
<v Logan Graham (Anthropic)>they

4736
04:43:32.520 --> 04:43:33.000
<v Logan Graham (Anthropic)>would like.

4737
04:43:33.805 --> 04:43:37.645
<v Logan Graham (Anthropic)>But I would refer you to our public
documentation where we we, describe this in

4738
04:43:37.645 --> 04:43:38.285
<v Logan Graham (Anthropic)>detail.

4739
04:43:40.125 --> 04:43:45.404
<v CM Carmen De La Rosa>Thank you all. So up next, we have council
member Phil Wang, council member Shuang,

4740
04:43:45.404 --> 04:43:46.605
<v CM Carmen De La Rosa>and then Epstein.

4741
04:43:46.764 --> 04:43:47.725
<v CM Phil Wong>K. Thank you, chair.

4742
04:43:49.390 --> 04:43:50.350
<v CM Phil Wong>My question

4743
04:43:50.750 --> 04:43:51.710
<v CM Phil Wong>pertains to,

4744
04:43:52.750 --> 04:43:54.031
<v CM Phil Wong>the self

4745
04:43:54.270 --> 04:43:58.271
<v CM Phil Wong>improvement process, the recursive
self improvement process,

4746
04:43:58.430 --> 04:44:00.510
<v CM Phil Wong>as well as the training activities

4747
04:44:00.510 --> 04:44:06.664
<v CM Phil Wong>of your company's AI model. And it is my
understanding that when you... When it goes

4748
04:44:06.664 --> 04:44:07.225
<v CM Phil Wong>through,

4749
04:44:07.943 --> 04:44:11.465
<v CM Phil Wong>the self improvement
process, it generates code

4750
04:44:12.023 --> 04:44:17.305
<v CM Phil Wong>that will... Itself will run next time
around, and then it will be a smarter AI

4751
04:44:17.305 --> 04:44:17.705
<v CM Phil Wong>model.

4752
04:44:18.140 --> 04:44:19.420
<v CM Phil Wong>Is that right?

4753
04:44:19.900 --> 04:44:20.861
<v CM Phil Wong>OpenAI?

4754
04:44:20.861 --> 04:44:24.060
<v CM Phil Wong>Yes or no? Yes or no? It generates code.

4755
04:44:24.621 --> 04:44:27.021
<v CM Phil Wong>So there be a smarter AI model.

4756
04:44:28.140 --> 04:44:32.125
<v Morgan Dwyer (OpenAI)>I'd like to get back to you on the specific
definition of RSI. I'm not sure that

4757
04:44:32.125 --> 04:44:36.525
<v Morgan Dwyer (OpenAI)>what you're saying maps exactly to
the technical definition we use. K.

4758
04:44:36.924 --> 04:44:37.805
<v CM Phil Wong>Entropic,

4759
04:44:38.205 --> 04:44:40.365
<v CM Phil Wong>does it generate code automatically

4760
04:44:40.605 --> 04:44:43.164
<v CM Phil Wong>to make itself smarter
as part of the process?

4761
04:44:44.300 --> 04:44:45.660
<v Logan Graham (Anthropic)>Yes. We've detailed,

4762
04:44:45.900 --> 04:44:49.900
<v Logan Graham (Anthropic)>publicly multiple times the degrees in
which models that generate code are being

4763
04:44:49.900 --> 04:44:51.021
<v Logan Graham (Anthropic)>used in our Does

4764
04:44:51.500 --> 04:44:55.820
<v CM Phil Wong>your model self generate code as
part of the training activities

4765
04:44:56.060 --> 04:44:57.660
<v CM Phil Wong>or self improvement process?

4766
04:44:58.506 --> 04:45:01.307
<v Shane Cahill (Meta)>Council member, I'll need to get back
to you and have our technical experts

4767
04:45:01.307 --> 04:45:03.467
<v Shane Cahill (Meta)>follow-up. How about Google?

4768
04:45:04.586 --> 04:45:08.666
<v Alice Friend (Google)>Yeah. Same here, sir. I'm not sure we
would describe it exactly the way you are

4769
04:45:08.666 --> 04:45:11.785
<v Alice Friend (Google)>describing, so I wanna get you
the precise technical definition.

4770
04:45:11.785 --> 04:45:13.465
<v CM Phil Wong>Well, here's my question. If

4771
04:45:13.980 --> 04:45:18.300
<v CM Phil Wong>if your model improves by
itself by generating code,

4772
04:45:19.100 --> 04:45:23.900
<v CM Phil Wong>is there a human being... Is it a team
of human beings that checks the code

4773
04:45:24.300 --> 04:45:25.900
<v CM Phil Wong>to make sure that it is

4774
04:45:26.984 --> 04:45:30.025
<v CM Phil Wong>benign and not misaligned or malicious?

4775
04:45:30.344 --> 04:45:32.504
<v CM Phil Wong>Can can each one of you answer that?

4776
04:45:34.904 --> 04:45:37.385
<v Morgan Dwyer (OpenAI)>So as part of our safety process,

4777
04:45:37.625 --> 04:45:43.590
<v Morgan Dwyer (OpenAI)>I... I've mentioned our preparedness framework
that outlines how we evaluate models,

4778
04:45:43.910 --> 04:45:49.030
<v Morgan Dwyer (OpenAI)>and how we assess risks across a wide
range of national security risks.

4779
04:45:51.030 --> 04:45:53.750
<v Morgan Dwyer (OpenAI)>So we do, robust evaluations

4780
04:45:53.750 --> 04:45:55.190
<v Morgan Dwyer (OpenAI)>of a set of risks,

4781
04:45:55.865 --> 04:45:59.385
<v Morgan Dwyer (OpenAI)>and we make those results
public as part of our,

4782
04:45:59.705 --> 04:46:03.305
<v Morgan Dwyer (OpenAI)>system card, which we release,
with every major model.

4783
04:46:04.984 --> 04:46:09.545
<v Shane Cahill (Meta)>Council member, as part of our scaling
framework, we outline considerations

4784
04:46:09.545 --> 04:46:13.910
<v Shane Cahill (Meta)>that we are taking overall to maintain
human control of model development.

4785
04:46:16.630 --> 04:46:20.150
<v Alice Friend (Google)>Yes, sir. As I said before,
we have supervised processes

4786
04:46:20.150 --> 04:46:22.390
<v Alice Friend (Google)>to keep progress verifiable,

4787
04:46:22.390 --> 04:46:27.416
<v Alice Friend (Google)>and we have a range of different techniques
to test the model and its performance

4788
04:46:27.656 --> 04:46:31.096
<v Alice Friend (Google)>to ensure that it's staying
within defined safety limits.

4789
04:46:31.416 --> 04:46:35.496
<v Alice Friend (Google)>We also have an AI control framework
that we released earlier this summer,

4790
04:46:35.816 --> 04:46:37.176
<v Alice Friend (Google)>which similarly

4791
04:46:37.256 --> 04:46:42.880
<v Alice Friend (Google)>takes into account how we're ensuring
that our systems are continuously under,

4792
04:46:43.200 --> 04:46:44.400
<v Alice Friend (Google)>human control,

4793
04:46:44.640 --> 04:46:48.320
<v Alice Friend (Google)>and we manage the risks of fully autonomous
model self improvement through our

4794
04:46:48.320 --> 04:46:49.760
<v Alice Friend (Google)>frontier safety framework.

4795
04:46:51.760 --> 04:46:52.480
<v Logan Graham (Anthropic)>And,

4796
04:46:52.560 --> 04:46:53.121
<v Logan Graham (Anthropic)>yes,

4797
04:46:53.440 --> 04:46:57.975
<v Logan Graham (Anthropic)>humans and the systems we put in place
monitor at every level of the model

4798
04:46:58.375 --> 04:46:59.574
<v Logan Graham (Anthropic)>during training,

4799
04:46:59.895 --> 04:47:01.895
<v Logan Graham (Anthropic)>prior to release, post release.

4800
04:47:02.055 --> 04:47:05.654
<v Logan Graham (Anthropic)>But this is an important question because
the models are clearly getting better

4801
04:47:05.654 --> 04:47:11.130
<v Logan Graham (Anthropic)>faster, and so we think more is needed
here. K. Thank you. Thank you, chair. Thank

4802
04:47:11.130 --> 04:47:13.771
<v Logan Graham (Anthropic)>you so much. Thank you.
Council member Huang.

4803
04:47:13.852 --> 04:47:14.970
<v CM Susan Zhuang>Thank you, chair.

4804
04:47:15.291 --> 04:47:20.490
<v CM Susan Zhuang>I have two questions. The first one, what
are you doing to offset the massive impact

4805
04:47:20.490 --> 04:47:21.691
<v CM Susan Zhuang>of data centers

4806
04:47:21.994 --> 04:47:23.354
<v CM Susan Zhuang>on the environment?

4807
04:47:23.594 --> 04:47:25.354
<v CM Susan Zhuang>The second one, are you

4808
04:47:25.674 --> 04:47:26.555
<v CM Susan Zhuang>tracking

4809
04:47:26.555 --> 04:47:27.834
<v CM Susan Zhuang>how inaccurate

4810
04:47:27.994 --> 04:47:32.875
<v CM Susan Zhuang>statement or decisions by your
AI tools are impacting security?

4811
04:47:39.291 --> 04:47:43.531
<v Morgan Dwyer (OpenAI)>So I... I'll start by addressing
your question on data centers,

4812
04:47:43.850 --> 04:47:45.291
<v Morgan Dwyer (OpenAI)>and the environment.

4813
04:47:45.770 --> 04:47:47.371
<v Morgan Dwyer (OpenAI)>As I mentioned before,

4814
04:47:47.690 --> 04:47:48.650
<v Morgan Dwyer (OpenAI)>OpenAI's,

4815
04:47:48.890 --> 04:47:51.770
<v Morgan Dwyer (OpenAI)>data centers, that we build,
they're called Stargate.

4816
04:47:52.416 --> 04:47:54.736
<v Morgan Dwyer (OpenAI)>And for each of our data centers,

4817
04:47:55.055 --> 04:47:59.295
<v Morgan Dwyer (OpenAI)>we build them with and for communities.
And the environment and the environmental

4818
04:47:59.295 --> 04:48:00.336
<v Morgan Dwyer (OpenAI)>impacts

4819
04:48:00.416 --> 04:48:05.215
<v Morgan Dwyer (OpenAI)>on communities is part of how we
think through our community compacts.

4820
04:48:05.920 --> 04:48:07.520
<v Morgan Dwyer (OpenAI)>For every data center,

4821
04:48:07.680 --> 04:48:12.719
<v Morgan Dwyer (OpenAI)>we have released community compact that
details our commitments to communities,

4822
04:48:13.280 --> 04:48:17.520
<v Morgan Dwyer (OpenAI)>and that includes a range of things
like paying our own way on energy

4823
04:48:18.445 --> 04:48:20.365
<v Morgan Dwyer (OpenAI)>as well as commitments

4824
04:48:20.365 --> 04:48:22.045
<v Morgan Dwyer (OpenAI)>to... Community commitments,

4825
04:48:22.285 --> 04:48:25.326
<v Morgan Dwyer (OpenAI)>to work with with the
the local governments.

4826
04:48:25.725 --> 04:48:27.645
<v Morgan Dwyer (OpenAI)>But I think the key point is,

4827
04:48:28.285 --> 04:48:33.450
<v Morgan Dwyer (OpenAI)>the community compacts are driven by
local concerns, and so we want to meet the

4828
04:48:33.450 --> 04:48:34.490
<v Morgan Dwyer (OpenAI)>needs of the community,

4829
04:48:34.810 --> 04:48:36.810
<v Morgan Dwyer (OpenAI)>and and we work with them to do that.

4830
04:48:40.650 --> 04:48:41.610
<v Shane Cahill (Meta)>Council member,

4831
04:48:41.930 --> 04:48:46.090
<v Shane Cahill (Meta)>in relation to, to data centers
and the impact of the environment,

4832
04:48:46.090 --> 04:48:49.785
<v Shane Cahill (Meta)>thank you very much for asking that question.
I know it's top of mind for a lot of

4833
04:48:49.785 --> 04:48:50.266
<v Shane Cahill (Meta)>people.

4834
04:48:51.945 --> 04:48:53.625
<v Shane Cahill (Meta)>Our sustainable

4835
04:48:53.625 --> 04:48:55.227
<v Shane Cahill (Meta)>infrastructure commitments

4836
04:48:56.025 --> 04:48:57.785
<v Shane Cahill (Meta)>are are are core to

4837
04:48:58.346 --> 04:49:00.826
<v Shane Cahill (Meta)>our philosophy around our
data centers, including

4838
04:49:01.970 --> 04:49:06.611
<v Shane Cahill (Meta)>consumers should not be bearing the costs
and paying our own way in relation to to

4839
04:49:06.611 --> 04:49:07.490
<v Shane Cahill (Meta)>electricity.

4840
04:49:07.810 --> 04:49:12.531
<v Shane Cahill (Meta)>But I would like to connect you with our
our data center team so they can provide

4841
04:49:12.531 --> 04:49:14.611
<v Shane Cahill (Meta)>you with more specific...
Specifics on that.

4842
04:49:15.436 --> 04:49:21.035
<v CM Susan Zhuang>How about the one question about
the... How you actuate the statement

4843
04:49:21.115 --> 04:49:23.916
<v CM Susan Zhuang>by your AI tools are impacting security?

4844
04:49:31.130 --> 04:49:33.691
<v CM Susan Zhuang>Do you guys are tracking those data?

4845
04:49:35.050 --> 04:49:36.172
<v Shane Cahill (Meta)>Council member,

4846
04:49:36.410 --> 04:49:38.890
<v Shane Cahill (Meta)>I I think in relation to inaccurate

4847
04:49:38.890 --> 04:49:39.932
<v Shane Cahill (Meta)>information,

4848
04:49:40.490 --> 04:49:44.016
<v Shane Cahill (Meta)>AIs can make mistakes. And
I think that's important

4849
04:49:44.336 --> 04:49:45.775
<v Shane Cahill (Meta)>in terms of

4850
04:49:46.176 --> 04:49:47.297
<v Shane Cahill (Meta)>acknowledging

4851
04:49:47.775 --> 04:49:49.695
<v Shane Cahill (Meta)>that AIs can make mistakes.

4852
04:49:49.855 --> 04:49:54.256
<v Shane Cahill (Meta)>We disclose this. We're very transparent
in relation to to to that in terms of our

4853
04:49:54.256 --> 04:49:54.977
<v Shane Cahill (Meta)>AI offerings.

4854
04:49:58.879 --> 04:49:59.760
<v Logan Graham (Anthropic)>And

4855
04:49:59.760 --> 04:50:02.000
<v Logan Graham (Anthropic)>I I can offer a perspective here that,

4856
04:50:02.719 --> 04:50:06.799
<v Logan Graham (Anthropic)>we think, and we've worked very hard
this year to try to improve the world's

4857
04:50:06.799 --> 04:50:07.440
<v Logan Graham (Anthropic)>security

4858
04:50:07.600 --> 04:50:09.756
<v Logan Graham (Anthropic)>because we think models

4859
04:50:09.756 --> 04:50:10.477
<v Logan Graham (Anthropic)>might,

4860
04:50:11.035 --> 04:50:12.717
<v Logan Graham (Anthropic)>if not properly safeguarded,

4861
04:50:13.516 --> 04:50:17.355
<v Logan Graham (Anthropic)>harm security, specifically the
cybersecurity of critical infrastructure or

4862
04:50:17.355 --> 04:50:21.916
<v Logan Graham (Anthropic)>otherwise. This year, we withheld a model
to be able to give beneficial access to

4863
04:50:21.916 --> 04:50:24.430
<v Logan Graham (Anthropic)>defenders to have time to patch systems

4864
04:50:24.670 --> 04:50:26.590
<v Logan Graham (Anthropic)>with security vulnerabilities.

4865
04:50:26.910 --> 04:50:30.670
<v Logan Graham (Anthropic)>And I can say working with training teams
and otherwise, we work to make the model

4866
04:50:30.670 --> 04:50:34.910
<v Logan Graham (Anthropic)>better at making sure the code that it
writes is more secure, for example.

4867
04:50:36.785 --> 04:50:41.666
<v CM Carmen De La Rosa>Thank you so much. Up up next, we have
council member Epstein followed by Marte

4868
04:50:41.666 --> 04:50:43.906
<v CM Carmen De La Rosa>followed by Sansuoso.

4869
04:50:44.625 --> 04:50:48.066
<v CM Harvey Epstein>Thank you, chair, and thank you for being
here for... I know it's a long panel.

4870
04:50:48.770 --> 04:50:51.330
<v CM Harvey Epstein>So I have a couple of questions I'll ask
him, and, hopefully, you'll be able to

4871
04:50:51.330 --> 04:50:55.170
<v CM Harvey Epstein>answer for the entire four members of the
panel. Can you confirm whether you think

4872
04:50:55.170 --> 04:50:57.490
<v CM Harvey Epstein>self regulation has failed so far?

4873
04:50:57.889 --> 04:51:02.824
<v CM Harvey Epstein>And if it has failed, which I think we
effectively think it has, do you think that,

4874
04:51:03.145 --> 04:51:07.225
<v CM Harvey Epstein>third party audits, especially,
when AI or tools are being used

4875
04:51:07.545 --> 04:51:12.186
<v CM Harvey Epstein>to highlight risk decisions should be
regulated and whether government plays an

4876
04:51:12.186 --> 04:51:13.705
<v CM Harvey Epstein>effective role in that regulation?

4877
04:51:14.590 --> 04:51:16.350
<v CM Harvey Epstein>Second, about liability.

4878
04:51:16.510 --> 04:51:21.870
<v CM Harvey Epstein>When AI violates the law, what liability
do you think the company should have for

4879
04:51:21.870 --> 04:51:23.390
<v CM Harvey Epstein>legal violations,

4880
04:51:23.390 --> 04:51:25.390
<v CM Harvey Epstein>whether criminal or civil liability?

4881
04:51:25.390 --> 04:51:27.390
<v CM Harvey Epstein>Seems like there's some culpability

4882
04:51:27.390 --> 04:51:28.430
<v CM Harvey Epstein>that should be in place.

4883
04:51:29.705 --> 04:51:34.744
<v CM Harvey Epstein>And then the turnaround time when these
violations happen. When some when some...

4884
04:51:34.984 --> 04:51:37.943
<v CM Harvey Epstein>You know you know, the hugging face or
some other violation happened, what

4885
04:51:37.943 --> 04:51:40.824
<v CM Harvey Epstein>responsibility you have to notify
the public and the impacted people?

4886
04:51:41.330 --> 04:51:45.570
<v CM Harvey Epstein>Would you agree to some short turnaround
time, whether it's forty eight hours or

4887
04:51:45.570 --> 04:51:47.170
<v CM Harvey Epstein>not? And finally,

4888
04:51:47.490 --> 04:51:51.650
<v CM Harvey Epstein>the... You know, we've seen a lot of records
about a compilation of these violations

4889
04:51:51.650 --> 04:51:54.530
<v CM Harvey Epstein>of what's happened. Are you
willing to commit to releasing

4890
04:51:55.305 --> 04:51:57.145
<v CM Harvey Epstein>the list of places where these,

4891
04:51:57.625 --> 04:52:02.025
<v CM Harvey Epstein>kind of rogue actors happened? I know you've
released some letters to the EU. Will

4892
04:52:02.025 --> 04:52:06.186
<v CM Harvey Epstein>you release it to the city council as well
so we can see what's been happening in

4893
04:52:06.186 --> 04:52:07.705
<v CM Harvey Epstein>The United States and around the country?

4894
04:52:11.889 --> 04:52:14.129
<v Morgan Dwyer (OpenAI)>So on the question of self regulation,

4895
04:52:14.610 --> 04:52:16.209
<v Morgan Dwyer (OpenAI)>OpenAI was proud,

4896
04:52:16.370 --> 04:52:17.410
<v Morgan Dwyer (OpenAI)>to support,

4897
04:52:17.889 --> 04:52:19.330
<v Morgan Dwyer (OpenAI)>the recent superintelligence

4898
04:52:19.330 --> 04:52:24.049
<v Morgan Dwyer (OpenAI)>accords. We've also supported voluntary
frameworks in, the Biden Harris

4899
04:52:24.049 --> 04:52:24.690
<v Morgan Dwyer (OpenAI)>administration,

4900
04:52:25.186 --> 04:52:28.146
<v Morgan Dwyer (OpenAI)>but we don't think voluntary
frameworks are enough.

4901
04:52:28.625 --> 04:52:31.186
<v Morgan Dwyer (OpenAI)>And that is why we have
supported regulation

4902
04:52:31.186 --> 04:52:36.146
<v Morgan Dwyer (OpenAI)>for frontier safety both at the federal
level and at the state level, and we'll

4903
04:52:36.146 --> 04:52:37.105
<v Morgan Dwyer (OpenAI)>continue to do that.

4904
04:52:38.320 --> 04:52:40.320
<v Morgan Dwyer (OpenAI)>On your question regarding,

4905
04:52:40.639 --> 04:52:41.920
<v Morgan Dwyer (OpenAI)>transparency

4906
04:52:41.920 --> 04:52:43.120
<v Morgan Dwyer (OpenAI)>of incidents,

4907
04:52:44.400 --> 04:52:45.680
<v Morgan Dwyer (OpenAI)>so OpenAI,

4908
04:52:46.000 --> 04:52:46.879
<v Morgan Dwyer (OpenAI)>is

4909
04:52:46.879 --> 04:52:47.760
<v Morgan Dwyer (OpenAI)>currently,

4910
04:52:47.920 --> 04:52:51.280
<v Morgan Dwyer (OpenAI)>investigating past potential
incidents of misalignment.

4911
04:52:52.115 --> 04:52:56.836
<v Morgan Dwyer (OpenAI)>And if we discover those incidents,
we are notifying third parties,

4912
04:52:57.234 --> 04:52:59.234
<v Morgan Dwyer (OpenAI)>and making the results public.

4913
04:52:59.875 --> 04:53:02.516
<v Morgan Dwyer (OpenAI)>I'm also happy to say that we have,

4914
04:53:03.154 --> 04:53:08.430
<v Morgan Dwyer (OpenAI)>a reporting framework now where anyone in
the company can report suspected incidents

4915
04:53:08.430 --> 04:53:09.630
<v Morgan Dwyer (OpenAI)>of misalignment.

4916
04:53:10.111 --> 04:53:15.310
<v Morgan Dwyer (OpenAI)>We evaluate those incidents, remediate
them, notify third parties, and make the

4917
04:53:15.310 --> 04:53:16.432
<v Morgan Dwyer (OpenAI)>results public.

4918
04:53:16.990 --> 04:53:18.670
<v Morgan Dwyer (OpenAI)>So we're doing that already.

4919
04:53:18.990 --> 04:53:20.191
<v Morgan Dwyer (OpenAI)>We think transparency

4920
04:53:20.191 --> 04:53:25.045
<v Morgan Dwyer (OpenAI)>is really important here to help policymakers
like you, as well as researchers and

4921
04:53:25.045 --> 04:53:26.246
<v Morgan Dwyer (OpenAI)>the rest of industry,

4922
04:53:26.564 --> 04:53:27.445
<v Morgan Dwyer (OpenAI)>understand,

4923
04:53:27.846 --> 04:53:30.564
<v Morgan Dwyer (OpenAI)>the risks, and improve safety.

4924
04:53:31.205 --> 04:53:35.686
<v CM Harvey Epstein>And can you share a list of those incidents
so we have those, of ones you've already

4925
04:53:35.686 --> 04:53:36.166
<v CM Harvey Epstein>determined?

4926
04:53:37.270 --> 04:53:42.229
<v Morgan Dwyer (OpenAI)>Yes. They are all available online, and so
I'm happy to follow-up with the website,

4927
04:53:42.389 --> 04:53:44.070
<v Morgan Dwyer (OpenAI)>where they're all publicly available.

4928
04:53:45.270 --> 04:53:49.350
<v CM Harvey Epstein>I don't know who goes next,
chair, but I'm... Kehu.

4929
04:53:49.510 --> 04:53:52.389
<v Shane Cahill (Meta)>I'll go next. Thank you, council
member. Thank you, chair.

4930
04:53:52.984 --> 04:53:56.584
<v Shane Cahill (Meta)>In relation to your first
question, council member,

4931
04:53:57.783 --> 04:53:59.064
<v Shane Cahill (Meta)>I see it differently

4932
04:53:59.305 --> 04:54:02.264
<v Shane Cahill (Meta)>in in that sense that we're
very much accountable.

4933
04:54:02.584 --> 04:54:07.861
<v Shane Cahill (Meta)>We're accountable. We publish our
framework or scaling framework publicly,

4934
04:54:07.940 --> 04:54:09.621
<v Shane Cahill (Meta)>preparedness reports.

4935
04:54:10.980 --> 04:54:13.300
<v Shane Cahill (Meta)>We use ex... Outside expertise

4936
04:54:13.621 --> 04:54:17.300
<v Shane Cahill (Meta)>in relation to the Accord,
which we signed with our peers.

4937
04:54:17.621 --> 04:54:20.500
<v Shane Cahill (Meta)>We committed to partnering
with independent

4938
04:54:21.426 --> 04:54:23.266
<v Shane Cahill (Meta)>auditors and evaluators,

4939
04:54:23.746 --> 04:54:28.064
<v Shane Cahill (Meta)>and we also consider that there's a
considerable body of existing federal law which

4940
04:54:28.064 --> 04:54:29.025
<v Shane Cahill (Meta)>also applies.

4941
04:54:30.385 --> 04:54:32.545
<v Shane Cahill (Meta)>In relation to incidents,

4942
04:54:33.340 --> 04:54:35.180
<v Shane Cahill (Meta)>we have a robust global

4943
04:54:35.180 --> 04:54:36.780
<v Shane Cahill (Meta)>response program in place

4944
04:54:37.020 --> 04:54:41.740
<v Shane Cahill (Meta)>in relation to incidents and a very
important part of our overall processes,

4945
04:54:42.540 --> 04:54:45.900
<v Shane Cahill (Meta)>and we report incidents
as as required by law.

4946
04:54:46.645 --> 04:54:48.725
<v Shane Cahill (Meta)>In relation to public disclosure,

4947
04:54:48.805 --> 04:54:52.244
<v Shane Cahill (Meta)>I'm only aware of one incident and...

4948
04:54:52.484 --> 04:54:54.805
<v Shane Cahill (Meta)>Over the summer, and we have

4949
04:54:55.045 --> 04:54:55.684
<v Shane Cahill (Meta)>our...

4950
04:54:55.924 --> 04:55:00.004
<v Shane Cahill (Meta)>A a public post in relation to
the details of our independent

4951
04:55:01.340 --> 04:55:03.102
<v Shane Cahill (Meta)>independent review of that?

4952
04:55:07.420 --> 04:55:11.340
<v Alice Friend (Google)>Sir, as I've stated in earlier
answers, we believe that,

4953
04:55:11.740 --> 04:55:12.781
<v Alice Friend (Google)>AI is,

4954
04:55:13.100 --> 04:55:14.460
<v Alice Friend (Google)>covered by existing

4955
04:55:15.035 --> 04:55:16.154
<v Alice Friend (Google)>regulation

4956
04:55:16.154 --> 04:55:18.395
<v Alice Friend (Google)>at both the federal and state level.

4957
04:55:19.035 --> 04:55:24.635
<v Alice Friend (Google)>But where there are gaps in existing
regulation or existing law or where, in

4958
04:55:24.635 --> 04:55:26.795
<v Alice Friend (Google)>particular frontier AI poses,

4959
04:55:27.115 --> 04:55:27.676
<v Alice Friend (Google)>potentially

4960
04:55:28.410 --> 04:55:31.610
<v Alice Friend (Google)>new risks. We have, supported,

4961
04:55:32.090 --> 04:55:35.050
<v Alice Friend (Google)>regulation at the federal
level and also regulations,

4962
04:55:35.370 --> 04:55:38.330
<v Alice Friend (Google)>both federally and at
the state level as well.

4963
04:55:40.090 --> 04:55:41.290
<v Alice Friend (Google)>On liability,

4964
04:55:41.290 --> 04:55:43.690
<v Alice Friend (Google)>we think it's important
to attach accountability

4965
04:55:44.195 --> 04:55:45.635
<v Alice Friend (Google)>and responsibility,

4966
04:55:46.516 --> 04:55:51.955
<v Alice Friend (Google)>to the party that is closest to the the
knowledge about harm and to the possible

4967
04:55:51.955 --> 04:55:53.795
<v Alice Friend (Google)>cause of harm as well.

4968
04:55:54.596 --> 04:55:59.310
<v Alice Friend (Google)>So we really see this as a value
chain approach to thinking through

4969
04:55:59.310 --> 04:56:00.670
<v Alice Friend (Google)>AI liability.

4970
04:56:00.910 --> 04:56:01.951
<v Alice Friend (Google)>In many cases,

4971
04:56:02.191 --> 04:56:03.470
<v Alice Friend (Google)>existing torts,

4972
04:56:03.630 --> 04:56:05.150
<v Alice Friend (Google)>already have procedures

4973
04:56:05.150 --> 04:56:05.791
<v Alice Friend (Google)>for,

4974
04:56:06.271 --> 04:56:07.871
<v Alice Friend (Google)>assigning accountability,

4975
04:56:08.111 --> 04:56:09.951
<v Alice Friend (Google)>depending on the,

4976
04:56:10.191 --> 04:56:12.350
<v Alice Friend (Google)>the particulars of of a case.

4977
04:56:13.805 --> 04:56:19.646
<v Alice Friend (Google)>And then I'm happy to to have the local
team in New York follow-up with you, about,

4978
04:56:19.885 --> 04:56:21.246
<v Alice Friend (Google)>our three incidents.

4979
04:56:23.486 --> 04:56:25.486
<v Logan Graham (Anthropic)>And, at Anthropic,

4980
04:56:25.486 --> 04:56:30.568
<v Logan Graham (Anthropic)>we have long pushed for two things. The
first is for the industry itself to push its

4981
04:56:30.568 --> 04:56:32.650
<v Logan Graham (Anthropic)>own way up the safety frontier.

4982
04:56:32.889 --> 04:56:36.568
<v Logan Graham (Anthropic)>Call that the race to the top, but we've
tried to do that from from day one, always

4983
04:56:36.568 --> 04:56:41.129
<v Logan Graham (Anthropic)>improving and always trying to create a
gold standard so that, everybody can rise

4984
04:56:41.129 --> 04:56:41.370
<v Logan Graham (Anthropic)>together.

4985
04:56:42.105 --> 04:56:46.586
<v Logan Graham (Anthropic)>I have personally thought that some of that
has been very, very positive. However,

4986
04:56:46.586 --> 04:56:50.346
<v Logan Graham (Anthropic)>we have to go further. And for that reason,
we welcome the role of government. We

4987
04:56:50.346 --> 04:56:52.586
<v Logan Graham (Anthropic)>have detailed proposals,

4988
04:56:52.904 --> 04:56:57.120
<v Logan Graham (Anthropic)>at the federal level with our advanced AI
framework. And we supported state level,

4989
04:56:57.280 --> 04:56:58.720
<v Logan Graham (Anthropic)>for this reason as well,

4990
04:56:59.200 --> 04:57:02.400
<v Logan Graham (Anthropic)>including ambitious legislation
in in Massachusetts.

4991
04:57:03.120 --> 04:57:06.879
<v Logan Graham (Anthropic)>On the point about liability, I'm
fairly far from being a lawyer,

4992
04:57:07.280 --> 04:57:11.360
<v Logan Graham (Anthropic)>but I I can say that Anthropic is committed
to following all all relevant laws.

4993
04:57:12.375 --> 04:57:16.533
<v Logan Graham (Anthropic)>And to the excellent question on how we
should handle incidents, I I think a couple

4994
04:57:16.533 --> 04:57:20.375
<v Logan Graham (Anthropic)>principles are really important to me having
seen some of these incidents up close.

4995
04:57:20.613 --> 04:57:23.254
<v Logan Graham (Anthropic)>The first is you have to triage,

4996
04:57:23.254 --> 04:57:26.560
<v Logan Graham (Anthropic)>investigate, and disclose fast
and be transparent about it.

4997
04:57:27.121 --> 04:57:31.281
<v Logan Graham (Anthropic)>I think everybody here agrees with that.
At the same time, we need to take the

4998
04:57:31.281 --> 04:57:33.762
<v Logan Graham (Anthropic)>safety of those affected
extremely seriously.

4999
04:57:33.762 --> 04:57:36.640
<v Logan Graham (Anthropic)>In some of these incidents,
disclosing certain information,

5000
04:57:37.201 --> 04:57:40.955
<v Logan Graham (Anthropic)>might put them at further harm. And so our
first priority is making sure that those

5001
04:57:40.955 --> 04:57:45.195
<v Logan Graham (Anthropic)>affected would not be further harmed.
But at the same time, we need to very

5002
04:57:45.195 --> 04:57:48.555
<v Logan Graham (Anthropic)>ambitiously and urgently disclose
so that all can benefit.

5003
04:57:49.275 --> 04:57:53.115
<v CM Harvey Epstein>Yeah. I appreciate the follow-up on the
information, especially if there are some

5004
04:57:53.115 --> 04:57:57.309
<v CM Harvey Epstein>violations and and the timeline for getting
people inputted, and I thank the chair

5005
04:57:57.309 --> 04:58:01.870
<v CM Harvey Epstein>for the time. Thank you so much. Up next,
we have council member Marte followed by

5006
04:58:01.870 --> 04:58:02.830
<v CM Harvey Epstein>Santosuoso

5007
04:58:02.830 --> 04:58:03.950
<v CM Carmen De La Rosa>and then Juan.

5008
04:58:04.910 --> 04:58:08.190
<v CM Christopher Marte>Thank you all for being here and staying
so late, and I wanna thank the speaker and

5009
04:58:08.190 --> 04:58:09.549
<v CM Christopher Marte>chair for hosting this hearing.

5010
04:58:10.605 --> 04:58:15.084
<v CM Christopher Marte>You know, I wanna just get some clarity.
One of the bill sponsors of the RAISE Act

5011
04:58:15.084 --> 04:58:16.605
<v CM Christopher Marte>said that OpenAI

5012
04:58:16.605 --> 04:58:17.965
<v CM Christopher Marte>was actually against...

5013
04:58:18.445 --> 04:58:23.560
<v CM Christopher Marte>Strongly against the initial draft language
of the bill, and you stated that you

5014
04:58:23.560 --> 04:58:28.680
<v CM Christopher Marte>have been supportive of the RAISE Act in
in Albany. So I kinda wanna know, just

5015
04:58:28.680 --> 04:58:32.201
<v CM Christopher Marte>because it seems like you guys have been
supportive of a lot of legislation,

5016
04:58:33.080 --> 04:58:34.600
<v CM Christopher Marte>based on your testimony,

5017
04:58:34.680 --> 04:58:40.076
<v CM Christopher Marte>what specific legislation are you against
in New York state or on a federal level?

5018
04:58:42.955 --> 04:58:44.717
<v Morgan Dwyer (OpenAI)>So thank you for the question.

5019
04:58:45.275 --> 04:58:46.795
<v Morgan Dwyer (OpenAI)>OpenAI worked,

5020
04:58:46.955 --> 04:58:49.115
<v Morgan Dwyer (OpenAI)>with other companies on,

5021
04:58:49.355 --> 04:58:50.795
<v Morgan Dwyer (OpenAI)>the New York RAISE Act,

5022
04:58:51.451 --> 04:58:52.730
<v Morgan Dwyer (OpenAI)>and we do support,

5023
04:58:52.890 --> 04:58:55.050
<v Morgan Dwyer (OpenAI)>the New York RAISE Act, which passed.

5024
04:58:55.771 --> 04:58:57.611
<v Morgan Dwyer (OpenAI)>We also, as I've mentioned,

5025
04:58:57.852 --> 04:58:59.210
<v Morgan Dwyer (OpenAI)>supported legislation

5026
04:58:59.210 --> 04:59:02.650
<v Morgan Dwyer (OpenAI)>that actually goes a step
beyond the New York RAISE Act.

5027
04:59:02.890 --> 04:59:07.936
<v Morgan Dwyer (OpenAI)>For example, in Illinois, that requires
independent third party audits of safety

5028
04:59:07.936 --> 04:59:08.734
<v Morgan Dwyer (OpenAI)>frameworks.

5029
04:59:09.855 --> 04:59:11.855
<v CM Christopher Marte>Are are there any pieces of legislation,

5030
04:59:11.855 --> 04:59:16.016
<v CM Christopher Marte>and I guess this is for everyone, that you
are against on the state level or federal

5031
04:59:16.016 --> 04:59:16.334
<v CM Christopher Marte>level?

5032
04:59:21.200 --> 04:59:26.240
<v Shane Cahill (Meta)>I I I go next. Council member, how I think
about it generally is the importance of

5033
04:59:26.240 --> 04:59:28.560
<v Shane Cahill (Meta)>ensuring that we have a robust uniform

5034
04:59:28.880 --> 04:59:29.840
<v Shane Cahill (Meta)>federal standard.

5035
04:59:30.266 --> 04:59:34.266
<v Shane Cahill (Meta)>I think I mentioned
earlier in the day that AI

5036
04:59:34.586 --> 04:59:36.186
<v Shane Cahill (Meta)>is transcontinental.

5037
04:59:36.506 --> 04:59:38.666
<v Shane Cahill (Meta)>It is inherently interstate.

5038
04:59:39.545 --> 04:59:44.746
<v Shane Cahill (Meta)>So it is important that there is a robust
uniform federal standard in relation to to

5039
04:59:44.746 --> 04:59:45.945
<v Shane Cahill (Meta)>how AI is governed.

5040
04:59:50.320 --> 04:59:53.600
<v Alice Friend (Google)>I think we would align ourselves
with those comments, sir.

5041
04:59:53.920 --> 04:59:57.121
<v Alice Friend (Google)>I think the most important
thing for us to focus on is,

5042
04:59:57.760 --> 05:00:02.316
<v Alice Friend (Google)>getting a strong federal framework,
particularly around frontier AI,

5043
05:00:02.875 --> 05:00:03.596
<v Alice Friend (Google)>and also,

5044
05:00:03.916 --> 05:00:05.115
<v Alice Friend (Google)>working through,

5045
05:00:05.355 --> 05:00:10.314
<v Alice Friend (Google)>an overall regulatory framework that's
going to ensure AI safety and also ensure

5046
05:00:10.314 --> 05:00:14.475
<v Alice Friend (Google)>that we can deploy and leverage
AI, across the entire country.

5047
05:00:16.690 --> 05:00:21.250
<v Logan Graham (Anthropic)>Yeah. And and likewise, I don't know what
our position is there exactly. What I do

5048
05:00:21.250 --> 05:00:25.170
<v Logan Graham (Anthropic)>know is that we have outlined
and supported our specific,

5049
05:00:25.729 --> 05:00:26.530
<v Logan Graham (Anthropic)>federal

5050
05:00:26.530 --> 05:00:27.889
<v Logan Graham (Anthropic)>legislation proposals.

5051
05:00:28.690 --> 05:00:33.195
<v Logan Graham (Anthropic)>We think a core part of this, for example,
needs to involve independent evaluators.

5052
05:00:33.754 --> 05:00:38.314
<v Logan Graham (Anthropic)>And, we we are, you know, long standing
supporters of the role of government here.

5053
05:00:40.314 --> 05:00:44.795
<v CM Carmen De La Rosa>Thank you all so much. Up next, we
have council member Santosuoso,

5054
05:00:44.795 --> 05:00:46.154
<v CM Carmen De La Rosa>Juan, and then Hanif.

5055
05:00:48.580 --> 05:00:49.701
<v CM Kayla Santosuosso>Thank you, chair.

5056
05:00:50.100 --> 05:00:53.941
<v CM Kayla Santosuosso>I primarily have questions
for, just mister Graham

5057
05:00:54.820 --> 05:00:55.460
<v CM Kayla Santosuosso>and,

5058
05:00:55.701 --> 05:00:57.220
<v CM Kayla Santosuosso>for OpenAI as well.

5059
05:00:57.781 --> 05:01:00.900
<v CM Kayla Santosuosso>This morning, we heard testimony
from a former Anthropic

5060
05:01:00.900 --> 05:01:05.754
<v CM Kayla Santosuosso>employee that based on his time in the
company, he believes that it is more likely

5061
05:01:05.754 --> 05:01:06.555
<v CM Kayla Santosuosso>than not,

5062
05:01:07.195 --> 05:01:12.234
<v CM Kayla Santosuosso>that misaligned AI will lead to the, quote,
unquote, total extinction of humanity.

5063
05:01:12.715 --> 05:01:17.436
<v CM Kayla Santosuosso>Another whistleblower who testified this
morning gave the... Gave this a one in

5064
05:01:17.436 --> 05:01:18.154
<v CM Kayla Santosuosso>three chance.

5065
05:01:18.800 --> 05:01:22.400
<v CM Kayla Santosuosso>Mister Graham, if I understand it, you
are part of the red team, which is tasked

5066
05:01:22.400 --> 05:01:25.441
<v CM Kayla Santosuosso>with identifying and addressing
the riskiest capabilities

5067
05:01:25.441 --> 05:01:26.480
<v CM Kayla Santosuosso>of AI.

5068
05:01:26.640 --> 05:01:30.080
<v CM Kayla Santosuosso>So my question for you is how
many people are on your team?

5069
05:01:31.955 --> 05:01:36.834
<v Logan Graham (Anthropic)>Current estimates about 25 just on my team
that we have. I will say we have a number

5070
05:01:36.834 --> 05:01:40.355
<v Logan Graham (Anthropic)>of teams that work on similarly
catastrophic risks in various roles,

5071
05:01:40.516 --> 05:01:44.436
<v CM Kayla Santosuosso>number into the hundreds at Anthropic.
And how many people work for Anthropic

5072
05:01:44.436 --> 05:01:44.834
<v CM Kayla Santosuosso>overall?

5073
05:01:47.121 --> 05:01:50.480
<v Logan Graham (Anthropic)>My current understanding is
maybe a bit more than 5,000.

5074
05:01:51.121 --> 05:01:51.762
<v CM Kayla Santosuosso>So

5075
05:01:52.000 --> 05:01:52.960
<v CM Kayla Santosuosso>25

5076
05:01:52.960 --> 05:01:58.240
<v CM Kayla Santosuosso>people are working on the riskiest forms of
AI in a company that has 5,000 employees

5077
05:01:58.240 --> 05:02:04.227
<v CM Kayla Santosuosso>dedicated to advancing just the models
without regard for safety or for that level

5078
05:02:04.227 --> 05:02:05.025
<v CM Kayla Santosuosso>of risk?

5079
05:02:05.906 --> 05:02:08.066
<v Logan Graham (Anthropic)>In in reality, there... If you,

5080
05:02:08.385 --> 05:02:11.586
<v Logan Graham (Anthropic)>look at all the roles focused
on catastrophic risks,

5081
05:02:11.826 --> 05:02:16.041
<v Logan Graham (Anthropic)>my personal estimate would be likely
several 100 at least. This is distributed

5082
05:02:16.041 --> 05:02:19.240
<v Logan Graham (Anthropic)>across our safeguards team
that focuses on ensuring that,

5083
05:02:19.720 --> 05:02:24.121
<v Logan Graham (Anthropic)>the models or any misuse for
catastrophic risks does not happen.

5084
05:02:24.680 --> 05:02:28.920
<v Logan Graham (Anthropic)>Our alignment research team that focuses
on notions of lossless control, for

5085
05:02:28.920 --> 05:02:31.080
<v Logan Graham (Anthropic)>example, and between them,
at least several 100.

5086
05:02:31.855 --> 05:02:33.775
<v CM Kayla Santosuosso>And do you feel that is sufficient,

5087
05:02:34.176 --> 05:02:38.574
<v CM Kayla Santosuosso>or do you feel that you need to dedicate
more time and resources and people to the

5088
05:02:38.574 --> 05:02:41.615
<v CM Kayla Santosuosso>riskiest capabilities of AI
given the level of risk?

5089
05:02:42.574 --> 05:02:46.420
<v Logan Graham (Anthropic)>Yeah. Our our view is that we put a lot
of resources into this, and this is what

5090
05:02:46.420 --> 05:02:49.700
<v Logan Graham (Anthropic)>we're known for. But at the
same time, we need to continue

5091
05:02:49.700 --> 05:02:50.660
<v Logan Graham (Anthropic)>doing more.

5092
05:02:51.059 --> 05:02:54.660
<v CM Kayla Santosuosso>This is a part of Anthropic that is...
Continues to grow quite substantially,

5093
05:02:54.660 --> 05:02:58.100
<v CM Kayla Santosuosso>especially as the models become more capable.
And I don't mean to cut you off, but I

5094
05:02:58.100 --> 05:03:01.715
<v CM Kayla Santosuosso>just... I only have so much time, so I I
wanna ask the same question of OpenAI. How

5095
05:03:01.715 --> 05:03:04.516
<v CM Kayla Santosuosso>many people on your team are dedicated to

5096
05:03:05.234 --> 05:03:06.035
<v CM Kayla Santosuosso>addressing

5097
05:03:06.035 --> 05:03:09.635
<v CM Kayla Santosuosso>and identifying the riskiest capabilities
of AI versus the number of employees in

5098
05:03:09.635 --> 05:03:12.275
<v CM Kayla Santosuosso>your company overall? And do
you find that sufficient?

5099
05:03:12.914 --> 05:03:15.074
<v Morgan Dwyer (OpenAI)>So I lead the policy team,

5100
05:03:15.395 --> 05:03:20.750
<v Morgan Dwyer (OpenAI)>but I'm happy to get back to you on numbers
on... Of people in the overall company

5101
05:03:20.750 --> 05:03:23.469
<v Morgan Dwyer (OpenAI)>as well as in our safety teams.

5102
05:03:23.870 --> 05:03:28.965
<v Morgan Dwyer (OpenAI)>But at OpenAI, we don't believe that it's
just... That you should just rely on our

5103
05:03:28.965 --> 05:03:30.885
<v Morgan Dwyer (OpenAI)>safety teams and our evaluators.

5104
05:03:30.885 --> 05:03:34.084
<v CM Kayla Santosuosso>I'm sorry. I don't... Again, I I... I'm
sorry to cut you off. I just only have so

5105
05:03:34.084 --> 05:03:38.404
<v CM Kayla Santosuosso>much time. Can you tell me how many people
in your company are currently working to

5106
05:03:38.404 --> 05:03:40.805
<v CM Kayla Santosuosso>address the risk of the knee... Of the

5107
05:03:41.740 --> 05:03:44.300
<v CM Kayla Santosuosso>potential extinction of humanity?

5108
05:03:46.540 --> 05:03:50.541
<v Morgan Dwyer (OpenAI)>As I said, I lead the policy
team, so I don't have that number.

5109
05:03:51.100 --> 05:03:55.180
<v Morgan Dwyer (OpenAI)>But we not only have teams that are at...
So thank you so much. I appreciate it.

5110
05:03:56.076 --> 05:03:57.996
<v CM Carmen De La Rosa>Alright. Thank you so much.

5111
05:03:59.596 --> 05:04:04.637
<v CM Carmen De La Rosa>Up next, we have council member one,
followed by Hanif, followed by Aviles,

5112
05:04:04.637 --> 05:04:05.836
<v CM Carmen De La Rosa>followed by Gutierrez.

5113
05:04:05.836 --> 05:04:07.037
<v CM Julie Won>Thank you so much.

5114
05:04:07.436 --> 05:04:11.770
<v CM Julie Won>This is for all panelists. What is the
exact failure rate or benchmark threshold

5115
05:04:11.770 --> 05:04:15.370
<v CM Julie Won>that would cause you to halt the deployment
of a new model? And if your model

5116
05:04:15.370 --> 05:04:19.290
<v CM Julie Won>successfully generates a bio weapon design
5% of the time during internal testing,

5117
05:04:19.290 --> 05:04:23.334
<v CM Julie Won>do you still release it? And then I have
a separate question. Will you commit today

5118
05:04:23.334 --> 05:04:27.494
<v CM Julie Won>to providing exact redacted logs of the
prompts that successfully jailbroke your

5119
05:04:27.494 --> 05:04:29.176
<v CM Julie Won>models by biosecurity

5120
05:04:29.176 --> 05:04:31.574
<v CM Julie Won>features... Filters during
your internal testing?

5121
05:04:36.450 --> 05:04:38.610
<v Morgan Dwyer (OpenAI)>So to your question on

5122
05:04:39.010 --> 05:04:40.930
<v Morgan Dwyer (OpenAI)>statistics or probabilities,

5123
05:04:40.930 --> 05:04:42.850
<v Morgan Dwyer (OpenAI)>I don't have a specific number.

5124
05:04:43.170 --> 05:04:48.690
<v Morgan Dwyer (OpenAI)>But what I can say is that no level of
catastrophic risk is remotely acceptable.

5125
05:04:49.416 --> 05:04:53.416
<v Morgan Dwyer (OpenAI)>And we don't believe that we should be
training models unless we can make an

5126
05:04:53.416 --> 05:04:57.176
<v Morgan Dwyer (OpenAI)>extremely strong case that we can
keep them under human control.

5127
05:05:01.350 --> 05:05:02.469
<v Shane Cahill (Meta)>Council member,

5128
05:05:02.790 --> 05:05:03.830
<v Shane Cahill (Meta)>I would just

5129
05:05:04.148 --> 05:05:07.990
<v Shane Cahill (Meta)>would just note that we are
committed to building AI safely.

5130
05:05:08.229 --> 05:05:09.750
<v Shane Cahill (Meta)>And we have laid out

5131
05:05:10.070 --> 05:05:12.790
<v Shane Cahill (Meta)>our... In detail how we approach this

5132
05:05:13.030 --> 05:05:16.229
<v Shane Cahill (Meta)>in our scaling framework and
in our preparedness reports.

5133
05:05:16.766 --> 05:05:21.486
<v Shane Cahill (Meta)>In relation to further detail, I would
need to get back to you on specifics.

5134
05:05:22.525 --> 05:05:24.045
<v CM Julie Won>For the other panelists?

5135
05:05:25.006 --> 05:05:29.725
<v Alice Friend (Google)>Yes, ma'am. I don't have a specific number
for failure rate that would cause us to

5136
05:05:29.725 --> 05:05:32.630
<v Alice Friend (Google)>make a nonlaunch decision.
Although, I do know that,

5137
05:05:33.111 --> 05:05:34.470
<v Alice Friend (Google)>we aim for,

5138
05:05:34.710 --> 05:05:39.111
<v Alice Friend (Google)>safety and performance standards ahead of
allowing ourselves to launch. And if we're

5139
05:05:39.111 --> 05:05:43.191
<v Alice Friend (Google)>not meeting that, it's not a good enough
model or product for us to release.

5140
05:05:43.824 --> 05:05:46.145
<v Alice Friend (Google)>I'm not personally aware of any,

5141
05:05:47.424 --> 05:05:49.023
<v Alice Friend (Google)>biosecurity

5142
05:05:49.023 --> 05:05:49.984
<v Alice Friend (Google)>evaluations

5143
05:05:49.984 --> 05:05:54.625
<v Alice Friend (Google)>where jailbreak... Successful jailbreaks
happened. So I'm gonna have to, go back to

5144
05:05:54.625 --> 05:05:58.225
<v Alice Friend (Google)>the teams and ask, if they've ever seen
such a thing and get back to you on that.

5145
05:06:00.459 --> 05:06:04.219
<v Logan Graham (Anthropic)>Thanks. And as I mentioned earlier, I
think we were essentially the first team in

5146
05:06:04.219 --> 05:06:07.500
<v Logan Graham (Anthropic)>the industry to design and
test models for this purpose.

5147
05:06:07.820 --> 05:06:11.580
<v Logan Graham (Anthropic)>We outline a very nuanced process
of how we decide how to launch

5148
05:06:12.266 --> 05:06:14.506
<v Logan Graham (Anthropic)>with respect to biological capabilities,

5149
05:06:14.506 --> 05:06:15.385
<v Logan Graham (Anthropic)>for example.

5150
05:06:15.705 --> 05:06:19.225
<v Logan Graham (Anthropic)>And in fact, in the past, we have held
the model because we needed more time to

5151
05:06:19.225 --> 05:06:20.986
<v Logan Graham (Anthropic)>assess its biological capabilities

5152
05:06:21.064 --> 05:06:22.746
<v Logan Graham (Anthropic)>and proactively implemented

5153
05:06:22.746 --> 05:06:24.266
<v Logan Graham (Anthropic)>safer safeguards,

5154
05:06:24.746 --> 05:06:29.110
<v Logan Graham (Anthropic)>for this reason. To the point of jailbreaks,
this is a great question. One of the

5155
05:06:29.110 --> 05:06:33.030
<v Logan Graham (Anthropic)>hard parts is sometimes you do or don't
want to share a jailbreak depending on

5156
05:06:33.030 --> 05:06:37.590
<v Logan Graham (Anthropic)>whether it would put other models at risk.
And so we regularly publish our research

5157
05:06:37.590 --> 05:06:38.870
<v Logan Graham (Anthropic)>and alignment reports

5158
05:06:39.186 --> 05:06:41.986
<v Logan Graham (Anthropic)>in order... With all the
information that we possibly can,

5159
05:06:42.385 --> 05:06:46.146
<v Logan Graham (Anthropic)>that would be safe to do so, and we
try to set the industry standard there.

5160
05:06:46.465 --> 05:06:50.545
<v CM Julie Won>Right now, I know you're not in the room,
but behind me, we have the safety claims,

5161
05:06:50.545 --> 05:06:53.346
<v CM Julie Won>and we have all the incidents that
have happened, including biosecurity,

5162
05:06:53.346 --> 05:06:57.682
<v CM Julie Won>which you have heard also from council
member Lynn Schulman who who continues to

5163
05:06:57.682 --> 05:06:58.640
<v CM Julie Won>lead us in,

5164
05:06:59.041 --> 05:07:01.041
<v CM Julie Won>as a committee chair for
health and hospitals.

5165
05:07:01.121 --> 05:07:05.121
<v CM Julie Won>So can you please clarify? So Anthropic
has been the only one that gave me a clear

5166
05:07:05.121 --> 05:07:08.640
<v CM Julie Won>answer. So for Anthropic, what is the
exact failure rate and benchmark threshold?

5167
05:07:09.205 --> 05:07:11.525
<v CM Julie Won>And the rest of you have
not given me that answer.

5168
05:07:11.846 --> 05:07:15.205
<v CM Carmen De La Rosa>And then we'll conclude and move
on to council member Hanif.

5169
05:07:17.525 --> 05:07:19.445
<v Logan Graham (Anthropic)>The answer is it's nuanced.

5170
05:07:19.686 --> 05:07:24.719
<v Logan Graham (Anthropic)>What we do, if you would like the sort of
complex answer, is we detail multi 100

5171
05:07:24.719 --> 05:07:28.318
<v Logan Graham (Anthropic)>pages of reports, several dozen of which
are usually allocated to biology every time

5172
05:07:28.318 --> 05:07:30.799
<v Logan Graham (Anthropic)>we launch a model. You can see the exact,

5173
05:07:31.040 --> 05:07:35.520
<v Logan Graham (Anthropic)>levels of capability and the exact reasoning
that we put in when we look at a number

5174
05:07:35.520 --> 05:07:39.656
<v Logan Graham (Anthropic)>of different tests that have different
so on, you know, quote, unquote, failure

5175
05:07:39.656 --> 05:07:43.656
<v Logan Graham (Anthropic)>rates that allow us to make that decision.
I think that highlights just how much

5176
05:07:43.656 --> 05:07:47.736
<v Logan Graham (Anthropic)>effort we put very specifically into, you
know, not just biology, but cybersecurity

5177
05:07:47.736 --> 05:07:49.736
<v Logan Graham (Anthropic)>and self improvement
and alignment as well.

5178
05:07:53.090 --> 05:07:54.450
<v CM Julie Won>And everybody else?

5179
05:07:56.610 --> 05:08:00.530
<v CM Julie Won>Because if what you claim is true is true,
then you wouldn't have released these

5180
05:08:00.530 --> 05:08:04.850
<v CM Julie Won>models that are causing these harms. But
I will pass on. We have to move on. Thank

5181
05:08:04.850 --> 05:08:08.943
<v CM Julie Won>you so much. Council member Hanif, followed
by council member Aviles, followed by

5182
05:08:08.943 --> 05:08:10.385
<v CM Julie Won>council member Gutierrez.

5183
05:08:10.625 --> 05:08:15.424
<v CM Shahana Hanif>Thank you, chair. On February twenty eighth
of this year, a 120 school children and

5184
05:08:15.424 --> 05:08:19.984
<v CM Shahana Hanif>35 teachers were killed in an airstrike
reported as the deadliest US attack on

5185
05:08:19.984 --> 05:08:21.344
<v CM Shahana Hanif>civilians in thirty five years.

5186
05:08:21.960 --> 05:08:24.760
<v CM Shahana Hanif>The military has stated the
target was misidentified

5187
05:08:24.760 --> 05:08:28.842
<v CM Shahana Hanif>by Project Maven, which originated
with Google and Amazon Web Services

5188
05:08:29.080 --> 05:08:33.080
<v CM Shahana Hanif>and later incorporated Palantir
Software and Anthropic's Claude.

5189
05:08:33.320 --> 05:08:38.234
<v CM Shahana Hanif>To mister Graham from Anthropic, your usage
policy prohibits facilitating any act of

5190
05:08:38.234 --> 05:08:41.115
<v CM Shahana Hanif>violence. What mechanism
enforces that prohibition

5191
05:08:41.195 --> 05:08:43.035
<v CM Shahana Hanif>in a classified deployment,

5192
05:08:43.275 --> 05:08:45.676
<v CM Shahana Hanif>and how would your company
learn of a violation?

5193
05:08:46.154 --> 05:08:50.154
<v CM Shahana Hanif>And has your company ever terminated
or suspended a government customer for

5194
05:08:50.154 --> 05:08:51.970
<v CM Shahana Hanif>violating its usage policy?

5195
05:08:52.209 --> 05:08:53.810
<v CM Shahana Hanif>If so, how many times?

5196
05:08:55.010 --> 05:08:57.650
<v Logan Graham (Anthropic)>First, I I really
appreciate this question.

5197
05:08:57.970 --> 05:09:01.250
<v Logan Graham (Anthropic)>In in my role, while I don't
have the exact answers to that,

5198
05:09:01.650 --> 05:09:02.850
<v Logan Graham (Anthropic)>I can point to

5199
05:09:03.170 --> 05:09:07.445
<v CM Shahana Hanif>our very public positions on Could
you just share them out loud here?

5200
05:09:08.404 --> 05:09:12.244
<v Logan Graham (Anthropic)>In government, we've we've taken very
public positions against the use of, for

5201
05:09:12.244 --> 05:09:14.086
<v Logan Graham (Anthropic)>example, domestic mass surveillance,

5202
05:09:14.725 --> 05:09:18.484
<v Logan Graham (Anthropic)>the use of autonomous weapons or
development of autonomous weapons.

5203
05:09:18.564 --> 05:09:20.324
<v Logan Graham (Anthropic)>How does company learn about

5204
05:09:20.950 --> 05:09:22.070
<v CM Shahana Hanif>that violation?

5205
05:09:23.510 --> 05:09:27.670
<v CM Shahana Hanif>I would And could you also just respond
the... To the piece about the termination

5206
05:09:28.148 --> 05:09:33.190
<v Logan Graham (Anthropic)>of a government I'm not entirely sure
of our process on on sort of government

5207
05:09:33.795 --> 05:09:37.395
<v Logan Graham (Anthropic)>sort of monitoring. I'd refer
to my my my colleagues there.

5208
05:09:37.555 --> 05:09:38.275
<v Alice Friend (Google) [?]>Thanks.

5209
05:09:40.195 --> 05:09:44.355
<v CM Carmen De La Rosa>Thank you so much. Council member
Aviles followed by Gutierrez to close.

5210
05:09:47.410 --> 05:09:51.570
<v CM Carmen De La Rosa>Thank you so much. I... To
clarify, to close this panel. Oh,

5211
05:09:52.451 --> 05:09:53.410
<v CM Alexa Avilés>so

5212
05:09:53.810 --> 05:09:58.690
<v CM Alexa Avilés>I'd I'd like to ask a a quick series of
questions to each of you. In this first sec,

5213
05:09:58.690 --> 05:10:03.096
<v CM Alexa Avilés>can you just say yes, no, and we can continue
to move forward? So does each of your

5214
05:10:03.096 --> 05:10:03.896
<v CM Alexa Avilés>companies

5215
05:10:03.975 --> 05:10:06.775
<v CM Alexa Avilés>sell or license its products directly

5216
05:10:06.855 --> 05:10:08.455
<v CM Alexa Avilés>or through intermediaries

5217
05:10:08.455 --> 05:10:12.695
<v CM Alexa Avilés>to immigration and customs enforcement
or customs and border protection?

5218
05:10:16.379 --> 05:10:18.940
<v Morgan Dwyer (OpenAI)>I don't know, but I'm happy
to get back to you on that.

5219
05:10:20.540 --> 05:10:22.780
<v Shane Cahill (Meta)>I'll get back to you on
that council member.

5220
05:10:23.500 --> 05:10:26.620
<v Alice Friend (Google)>I'm also so sorry I don't know,
but we will get back to you.

5221
05:10:27.020 --> 05:10:29.635
<v CM Alexa Avilés>Okay. Nobody knows, but

5222
05:10:29.635 --> 05:10:31.795
<v CM Alexa Avilés>it has been pretty documented that

5223
05:10:32.113 --> 05:10:35.635
<v CM Alexa Avilés>couple of... From Google
and in particular OpenAI,

5224
05:10:36.914 --> 05:10:39.715
<v CM Alexa Avilés>ICE has been utilizing your technology.

5225
05:10:39.715 --> 05:10:42.273
<v CM Alexa Avilés>So if you could please
provide to this council,

5226
05:10:42.514 --> 05:10:47.121
<v CM Alexa Avilés>each of those agreements, if in fact you
have them with each of those agencies.

5227
05:10:47.602 --> 05:10:52.080
<v CM Alexa Avilés>Will your company commit to not using New
York City resident data or law enforcement

5228
05:10:52.080 --> 05:10:54.720
<v CM Alexa Avilés>or immigration enforcement applications?

5229
05:10:55.840 --> 05:11:01.314
<v CM Alexa Avilés>Four. Excuse me. Will your company commit
to not using New York City resident data

5230
05:11:01.314 --> 05:11:05.555
<v CM Alexa Avilés>for law enforcement or immigration
enforcement applications?

5231
05:11:09.154 --> 05:11:13.260
<v Morgan Dwyer (OpenAI)>I'm sorry. I'm gonna have to get back to
you on that. I work on the policy team, not

5232
05:11:13.260 --> 05:11:14.940
<v Morgan Dwyer (OpenAI)>with our government contracts.

5233
05:11:17.340 --> 05:11:21.420
<v Shane Cahill (Meta)>I'm afraid it'll, also, we'll need
to get back to you, council member.

5234
05:11:22.379 --> 05:11:27.084
<v Alice Friend (Google) [?]>I have to give you the same answer, member,
but I'll ask the lawyers to to let us

5235
05:11:27.084 --> 05:11:33.166
<v Alice Friend (Google) [?]>know. Do any of your companies audit downstream
law enforcement use of its products?

5236
05:11:36.766 --> 05:11:39.004
<v Morgan Dwyer (OpenAI)>I don't have the answer to
that. I'm happy to follow-up.

5237
05:11:40.371 --> 05:11:42.771
<v Shane Cahill (Meta)>Council member, we'll follow-up with you.

5238
05:11:43.810 --> 05:11:45.170
<v Alice Friend (Google)>Same here.

5239
05:11:46.531 --> 05:11:49.092
<v Logan Graham (Anthropic)>Thank you. Likewise. We can follow-up.

5240
05:11:51.010 --> 05:11:53.570
<v CM Carmen De La Rosa>Thank you. Council member Gutierrez.

5241
05:11:55.984 --> 05:11:57.025
<v CM Jennifer Gutiérrez>Thank you all.

5242
05:11:58.064 --> 05:12:03.105
<v CM Jennifer Gutiérrez>I just wanted to ask on a number of
questions of issues that were raised today. I

5243
05:12:03.105 --> 05:12:07.744
<v CM Jennifer Gutiérrez>feel that every single person on this
panel has said safety is of the up note...

5244
05:12:07.744 --> 05:12:08.785
<v CM Jennifer Gutiérrez>Utmost importance,

5245
05:12:09.340 --> 05:12:09.900
<v CM Jennifer Gutiérrez>but

5246
05:12:10.140 --> 05:12:14.540
<v CM Jennifer Gutiérrez>I don't believe... I don't feel that. I
don't feel that in the responses, it feels

5247
05:12:15.340 --> 05:12:16.540
<v CM Jennifer Gutiérrez>too nuanced.

5248
05:12:16.940 --> 05:12:19.580
<v CM Jennifer Gutiérrez>And I would like to
understand specifically,

5249
05:12:20.140 --> 05:12:21.180
<v CM Jennifer Gutiérrez>in all cases,

5250
05:12:21.500 --> 05:12:26.914
<v CM Jennifer Gutiérrez>as much as you can share, How are you
directly ensuring safety of individual

5251
05:12:27.555 --> 05:12:31.074
<v CM Jennifer Gutiérrez>users when there's a data
breach, when there's issues

5252
05:12:31.314 --> 05:12:31.955
<v CM Jennifer Gutiérrez>of,

5253
05:12:32.674 --> 05:12:33.555
<v CM Jennifer Gutiérrez>surveillance?

5254
05:12:34.113 --> 05:12:38.670
<v CM Jennifer Gutiérrez>To, my colleague council member Shahana
Hanif's questions about AI and violence.

5255
05:12:38.670 --> 05:12:42.830
<v CM Jennifer Gutiérrez>What what is specifically that you can
tell to New Yorkers here that New Yorkers

5256
05:12:42.830 --> 05:12:47.470
<v CM Jennifer Gutiérrez>that have signed up to testify? What
what is the specific safety that you are

5257
05:12:47.470 --> 05:12:51.630
<v CM Jennifer Gutiérrez>ensuring? Make me believe that it is of
the utmost importance to you all. That's my

5258
05:12:51.630 --> 05:12:52.031
<v CM Jennifer Gutiérrez>one question.

5259
05:12:54.125 --> 05:12:55.887
<v Morgan Dwyer (OpenAI)>Yeah. Thank you for that question.

5260
05:12:56.365 --> 05:12:59.967
<v Morgan Dwyer (OpenAI)>So safety is a top priority at OpenAI.

5261
05:12:59.967 --> 05:13:05.166
<v Morgan Dwyer (OpenAI)>We build our models to be safe, throughout
the entire development and deployment

5262
05:13:05.166 --> 05:13:05.646
<v Morgan Dwyer (OpenAI)>process.

5263
05:13:06.110 --> 05:13:11.309
<v Morgan Dwyer (OpenAI)>That includes removing unsafe data from
training. It includes training the models to

5264
05:13:11.309 --> 05:13:12.430
<v Morgan Dwyer (OpenAI)>behave safely

5265
05:13:12.670 --> 05:13:14.990
<v Morgan Dwyer (OpenAI)>and evaluating them to be safely...

5266
05:13:15.148 --> 05:13:19.166
<v Morgan Dwyer (OpenAI)>To to behave safely across
a range of risks from cyber,

5267
05:13:19.564 --> 05:13:21.885
<v Morgan Dwyer (OpenAI)>to bio, to child safety.

5268
05:13:22.445 --> 05:13:24.926
<v Morgan Dwyer (OpenAI)>And then we monitor our safeguards

5269
05:13:24.926 --> 05:13:25.564
<v Morgan Dwyer (OpenAI)>for,

5270
05:13:26.125 --> 05:13:30.285
<v Morgan Dwyer (OpenAI)>for any potential misuse, and we
enforce on our usage policies.

5271
05:13:30.920 --> 05:13:34.840
<v Morgan Dwyer (OpenAI)>But to your specific question
of what I am personally doing,

5272
05:13:35.160 --> 05:13:38.440
<v Morgan Dwyer (OpenAI)>I am not on the safety team.
I lead our policy team,

5273
05:13:38.840 --> 05:13:41.240
<v Morgan Dwyer (OpenAI)>and I'm also a longtime public servant.

5274
05:13:41.719 --> 05:13:47.195
<v Morgan Dwyer (OpenAI)>And I share your concerns about safety,
And I think there's an important role for

5275
05:13:47.195 --> 05:13:48.395
<v Morgan Dwyer (OpenAI)>governments to play.

5276
05:13:48.635 --> 05:13:52.715
<v Morgan Dwyer (OpenAI)>And that's why I am proud that my
team is working with governments,

5277
05:13:53.115 --> 05:13:54.875
<v Morgan Dwyer (OpenAI)>to support regulations

5278
05:13:54.875 --> 05:13:59.660
<v Morgan Dwyer (OpenAI)>both at the state level and at the federal
level because we have strong safety

5279
05:13:59.660 --> 05:14:00.380
<v Morgan Dwyer (OpenAI)>practices,

5280
05:14:00.781 --> 05:14:03.500
<v Morgan Dwyer (OpenAI)>but we should also be held
accountable for them.

5281
05:14:06.300 --> 05:14:10.715
<v Shane Cahill (Meta)>Council member, thank you very much for
this question. And and overall, thank you

5282
05:14:10.795 --> 05:14:14.715
<v Shane Cahill (Meta)>for deepening the conversation on
this and other important topics.

5283
05:14:15.113 --> 05:14:16.875
<v Shane Cahill (Meta)>I would just reiterate that

5284
05:14:17.434 --> 05:14:21.273
<v Shane Cahill (Meta)>Meta is committed to
developing and deploying

5285
05:14:21.674 --> 05:14:23.674
<v Shane Cahill (Meta)>our AI offering safely.

5286
05:14:24.500 --> 05:14:30.100
<v Shane Cahill (Meta)>That is across the board, both in terms of
our models and also our systems and then

5287
05:14:30.100 --> 05:14:31.540
<v Shane Cahill (Meta)>our usage policies.

5288
05:14:31.620 --> 05:14:35.139
<v Shane Cahill (Meta)>So we cover the entire spectrum
in terms of our AI offerings.

5289
05:14:38.014 --> 05:14:38.895
<v Alice Friend (Google)>Ma'am,

5290
05:14:38.975 --> 05:14:44.014
<v Alice Friend (Google)>at Google, we have been committed to not
just the safety of our users, but the

5291
05:14:44.014 --> 05:14:49.135
<v Alice Friend (Google)>safety and security of the open web itself
for over a quarter of a century now.

5292
05:14:50.090 --> 05:14:52.969
<v Alice Friend (Google)>Through our own experiences
in our search platform,

5293
05:14:53.690 --> 05:14:57.610
<v Alice Friend (Google)>through everyday tools that
users rely on like Gmail,

5294
05:14:57.610 --> 05:14:58.650
<v Alice Friend (Google)>like YouTube,

5295
05:14:59.530 --> 05:15:04.234
<v Alice Friend (Google)>security and safety is incredibly important
to us. Because quite frankly, if people

5296
05:15:04.234 --> 05:15:06.635
<v Alice Friend (Google)>did not trust our platforms and our tools,

5297
05:15:06.795 --> 05:15:09.275
<v Alice Friend (Google)>they would not use them and
we would be out of business.

5298
05:15:09.674 --> 05:15:13.676
<v Alice Friend (Google)>So for us, safety is a
critical business imperative.

5299
05:15:13.834 --> 05:15:16.635
<v Alice Friend (Google)>It's also an imperative as part of

5300
05:15:17.770 --> 05:15:19.210
<v Alice Friend (Google)>being interested

5301
05:15:19.371 --> 05:15:20.650
<v Alice Friend (Google)>and committed

5302
05:15:20.890 --> 05:15:21.531
<v Alice Friend (Google)>to

5303
05:15:21.930 --> 05:15:27.210
<v Alice Friend (Google)>an Internet where people can get information
that is useful to them, that can help

5304
05:15:27.210 --> 05:15:28.410
<v Alice Friend (Google)>them live better lives.

5305
05:15:29.625 --> 05:15:31.705
<v Alice Friend (Google)>And finally, we invest in,

5306
05:15:32.266 --> 05:15:36.586
<v Alice Friend (Google)>the safety of our tools so that we can
deploy them to solve some of the world's

5307
05:15:36.586 --> 05:15:37.785
<v Alice Friend (Google)>biggest challenges.

5308
05:15:38.586 --> 05:15:39.785
<v Alice Friend (Google)>Everything from,

5309
05:15:40.744 --> 05:15:41.945
<v Alice Friend (Google)>natural disasters,

5310
05:15:42.270 --> 05:15:45.629
<v Alice Friend (Google)>warning people ahead of time
about floods and fires,

5311
05:15:46.510 --> 05:15:48.430
<v Alice Friend (Google)>to breakthroughs in diseases,

5312
05:15:48.510 --> 05:15:51.070
<v Alice Friend (Google)>to unlocking 200,000,000,

5313
05:15:51.309 --> 05:15:53.870
<v Alice Friend (Google)>all of the world's known proteins

5314
05:15:53.950 --> 05:15:56.270
<v Alice Friend (Google)>so that we can someday cure all disease.

5315
05:15:57.346 --> 05:16:00.545
<v Alice Friend (Google)>To us, safety is just a part
of this broader mission,

5316
05:16:01.105 --> 05:16:05.426
<v Alice Friend (Google)>to use technology and to use AI, in
particular, to make people's lives better.

5317
05:16:06.705 --> 05:16:07.586
<v Logan Graham (Anthropic)>And

5318
05:16:07.586 --> 05:16:10.225
<v Logan Graham (Anthropic)>if if I can share a
personal example of this,

5319
05:16:11.550 --> 05:16:15.630
<v Logan Graham (Anthropic)>I joined Anthropic when we fit around two
lunch tables. And the reason I did was

5320
05:16:15.630 --> 05:16:19.630
<v Logan Graham (Anthropic)>because these were the people that I saw
were taking it most seriously, that the

5321
05:16:19.630 --> 05:16:24.227
<v Logan Graham (Anthropic)>most important component of getting this
technology right was making it safe. And I

5322
05:16:24.227 --> 05:16:28.146
<v Logan Graham (Anthropic)>can say that, but let me demonstrate it.
From its outset, we are a public benefit

5323
05:16:28.146 --> 05:16:32.865
<v Logan Graham (Anthropic)>corporation legally, so we are legally
obligated to weigh the mission of SafeAI,

5324
05:16:33.025 --> 05:16:33.986
<v Logan Graham (Anthropic)>not just profit.

5325
05:16:34.625 --> 05:16:38.600
<v Logan Graham (Anthropic)>And at every step of the way, frequently
at meaningful cost to ourselves,

5326
05:16:39.000 --> 05:16:43.080
<v Logan Graham (Anthropic)>we have pioneered or tried to pioneer
the science of safety. Everything from

5327
05:16:43.080 --> 05:16:44.200
<v Logan Graham (Anthropic)>training itself,

5328
05:16:44.680 --> 05:16:48.360
<v Logan Graham (Anthropic)>the fundamental research, came
from us, the evaluations,

5329
05:16:48.600 --> 05:16:50.120
<v Logan Graham (Anthropic)>the predeployment

5330
05:16:50.120 --> 05:16:50.920
<v Logan Graham (Anthropic)>testing regime,

5331
05:16:51.336 --> 05:16:56.855
<v Logan Graham (Anthropic)>the monitoring and control, and just this
year, withholding our most frontier model

5332
05:16:56.855 --> 05:17:02.057
<v Logan Graham (Anthropic)>at the time at great cost in order to deploy
with defenders to ensure the world is

5333
05:17:02.057 --> 05:17:02.615
<v Logan Graham (Anthropic)>safe.

5334
05:17:03.176 --> 05:17:07.790
<v Logan Graham (Anthropic)>This is literally our our DNA. This is why
I'm here. This is why my team and I wake

5335
05:17:07.790 --> 05:17:08.830
<v Logan Graham (Anthropic)>up every morning.

5336
05:17:09.148 --> 05:17:12.830
<v Logan Graham (Anthropic)>And I I could tell you for a long time
about the... All the other parts of the all

5337
05:17:12.830 --> 05:17:14.830
<v Logan Graham (Anthropic)>the other parts of the stack,
but I hope that resonates.

5338
05:17:18.266 --> 05:17:18.906
<v Speaker Julie Menin>Yep.

5339
05:17:19.305 --> 05:17:20.346
<v Speaker Julie Menin>Okay.

5340
05:17:20.666 --> 05:17:23.785
<v Speaker Julie Menin>I'm gonna thank this panel for testifying.

5341
05:17:23.865 --> 05:17:29.625
<v Speaker Julie Menin>The council still has many questions that
we feel we did not get answers to today.

5342
05:17:29.625 --> 05:17:32.500
<v Speaker Julie Menin>We will be sending each of the companies a

5343
05:17:32.740 --> 05:17:36.740
<v Speaker Julie Menin>list of these questions that we
would like answers back in writing.

5344
05:17:37.300 --> 05:17:41.781
<v Speaker Julie Menin>But I thank the companies for being here
today, answering the members' questions.

5345
05:17:42.340 --> 05:17:43.300
<v Speaker Julie Menin>And thanks,

5346
05:17:43.541 --> 05:17:46.420
<v Speaker Julie Menin>to the public for their patience.
We're now gonna transition

5347
05:17:46.885 --> 05:17:50.484
<v Speaker Julie Menin>to the next panel, which is
the city administration,

5348
05:17:50.885 --> 05:17:53.686
<v Speaker Julie Menin>and we'll call on them
right now. Thank you.

5349
05:17:54.965 --> 05:17:55.525
<v Logan Graham (Anthropic)>Thank you.

5350
05:18:00.210 --> 05:18:04.451
<v CM Carmen De La Rosa [?]>Okay. Please remove them from
the chamber. Thank you. We

5351
05:18:04.770 --> 05:18:07.730
<v Unidentified [?]>know that the companies
are $80,000,000,000

5352
05:18:07.730 --> 05:18:09.810
<v Unidentified [?]>with our iPhone. Okay.

5353
05:18:10.451 --> 05:18:15.264
<v CM Carmen De La Rosa>Okay. You're in the phone. Now we will
move on to hear from the administration.

5354
05:18:15.664 --> 05:18:17.424
<v CM Carmen De La Rosa>And today, we will hear

5355
05:18:17.664 --> 05:18:21.424
<v CM Carmen De La Rosa>from Sarah Milstein, deputy
commissioner of strategic advisory

5356
05:18:21.424 --> 05:18:26.705
<v CM Carmen De La Rosa>services at the... At OTI CJ Dixon,
head of the New York City Cyber Command

5357
05:18:27.041 --> 05:18:31.682
<v CM Carmen De La Rosa>Benjamin Krakauer, first deputy commissioner
of New York City Emergency Management

5358
05:18:31.682 --> 05:18:35.920
<v CM Carmen De La Rosa>Department and Sam Levine, commissioner of
the New York City Department of Consumer

5359
05:18:35.920 --> 05:18:36.960
<v CM Carmen De La Rosa>and Worker Protection.

5360
05:18:48.254 --> 05:18:49.375
<v Committee Counsel>Thank you, chair.

5361
05:18:49.855 --> 05:18:50.814
<v Committee Counsel>Please

5362
05:18:50.895 --> 05:18:52.174
<v Committee Counsel>raise your right hands.

5363
05:18:53.360 --> 05:18:55.520
<v Committee Counsel>Thank you. Deputy... Do

5364
05:18:55.680 --> 05:18:57.120
<v Committee Counsel>you swear or affirm

5365
05:18:57.280 --> 05:19:01.280
<v Committee Counsel>to tell the truth and respond
honestly to council member questions?

5366
05:19:02.078 --> 05:19:03.840
<v Sarah Milstein (OTI) [?]>I do. Commissioner Sarah

5367
05:19:04.479 --> 05:19:05.120
<v Committee Counsel>Milstein?

5368
05:19:05.805 --> 05:19:10.445
<v Committee Counsel>Yeah. Yes. Thank you. Head
of Cyber Command, CJ Dixon?

5369
05:19:10.445 --> 05:19:13.244
<v Committee Counsel>Yes. Thank you. Deputy commissioner

5370
05:19:13.324 --> 05:19:14.285
<v Committee Counsel>Benjamin

5371
05:19:14.686 --> 05:19:15.725
<v Committee Counsel>Krakauer?

5372
05:19:16.924 --> 05:19:17.805
<v Benjamin Krakauer (NYC Emergency Management) [?]>I think so.

5373
05:19:21.330 --> 05:19:24.930
<v Committee Counsel>Thank you. I just wanna,
acknowledge that it was yes.

5374
05:19:25.250 --> 05:19:29.330
<v Committee Counsel>And commissioner Samuel Levine? Yes. Thank
you very much. You can begin with your

5375
05:19:29.330 --> 05:19:31.568
<v Committee Counsel>testimony. And deputy
commissioner Carlos Ortiz.

5376
05:19:36.225 --> 05:19:41.664
<v Committee Counsel>And, I apologize. And deputy commissioner
Carlos Ortiz. I do. Thank you. You may

5377
05:19:41.664 --> 05:19:42.225
<v Committee Counsel>begin.

5378
05:19:42.945 --> 05:19:43.664
<v Sarah Milstein (OTI)>Great.

5379
05:19:44.305 --> 05:19:49.210
<v Sarah Milstein (OTI)>Good morning, speaker Menon, chair De La
Rosa, and council members. My name is Sarah

5380
05:19:49.210 --> 05:19:53.210
<v Sarah Milstein (OTI)>Milstein, and I'm the deputy commissioner
for strategic advisory services

5381
05:19:53.291 --> 05:19:56.412
<v Sarah Milstein (OTI)>for the Office of Technology
and Innovation, OTI.

5382
05:19:56.730 --> 05:20:00.656
<v Sarah Milstein (OTI)>I'm joined here today by my
colleague, CJ Dixon, OTI's

5383
05:20:00.656 --> 05:20:03.535
<v Sarah Milstein (OTI)>deputy commissioner of New
York City Cyber Command

5384
05:20:03.615 --> 05:20:06.576
<v Sarah Milstein (OTI)>and New York City's chief
information security officer.

5385
05:20:07.055 --> 05:20:11.295
<v Sarah Milstein (OTI)>I'm also joined by Samuel Levine, commissioner
of the Department of Consumer Affairs

5386
05:20:11.295 --> 05:20:12.656
<v Sarah Milstein (OTI)>and Worker Protection,

5387
05:20:12.975 --> 05:20:14.256
<v Sarah Milstein (OTI)>DCWP

5388
05:20:14.256 --> 05:20:15.455
<v Sarah Milstein (OTI)>Carlos Ortiz,

5389
05:20:15.720 --> 05:20:16.840
<v Sarah Milstein (OTI)>DCWP's

5390
05:20:16.840 --> 05:20:19.880
<v Sarah Milstein (OTI)>chief of staff and deputy
commissioner for external affairs

5391
05:20:20.121 --> 05:20:23.002
<v Sarah Milstein (OTI)>and Benjamin Krakauer,
first deputy commissioner

5392
05:20:23.160 --> 05:20:25.480
<v Sarah Milstein (OTI)>for New York City emergency management,

5393
05:20:25.640 --> 05:20:26.361
<v Sarah Milstein (OTI)>Nyssen.

5394
05:20:27.320 --> 05:20:32.525
<v Sarah Milstein (OTI)>In my role at OTI, I oversee both both
the Office of Algorithmic Accountability,

5395
05:20:32.525 --> 05:20:33.404
<v Sarah Milstein (OTI)>OAA,

5396
05:20:33.725 --> 05:20:37.965
<v Sarah Milstein (OTI)>and New York City Cyber Command. We
appreciate that the council is holding this

5397
05:20:37.965 --> 05:20:40.926
<v Sarah Milstein (OTI)>timely hearing about the
risks of AI technology.

5398
05:20:41.084 --> 05:20:44.932
<v Sarah Milstein (OTI)>Thank you, speaker, for getting
representatives of the companies that develop,

5399
05:20:44.932 --> 05:20:45.650
<v Sarah Milstein (OTI)>deploy,

5400
05:20:45.810 --> 05:20:49.730
<v Sarah Milstein (OTI)>and sell AI tools to the general
public to speak to the capabilities,

5401
05:20:49.890 --> 05:20:51.650
<v Sarah Milstein (OTI)>limits, and risks associated

5402
05:20:51.970 --> 05:20:53.810
<v Sarah Milstein (OTI)>with these systems at this hearing.

5403
05:20:54.705 --> 05:20:56.785
<v Sarah Milstein (OTI)>As a representative of OTI,

5404
05:20:56.785 --> 05:21:01.025
<v Sarah Milstein (OTI)>I'm here today to provide information to
the Committee of the Whole about how we

5405
05:21:01.025 --> 05:21:05.344
<v Sarah Milstein (OTI)>serve the city's technology and cyber
defense needs and to address the proposed

5406
05:21:05.344 --> 05:21:06.145
<v Sarah Milstein (OTI)>legislation

5407
05:21:06.145 --> 05:21:08.305
<v Sarah Milstein (OTI)>that potentially impacts
the administration.

5408
05:21:12.330 --> 05:21:17.450
<v Sarah Milstein (OTI)>For those unfamiliar with our work, OTI's
core mission is to use technology, data,

5409
05:21:17.450 --> 05:21:22.250
<v Sarah Milstein (OTI)>and design to make the city work better
for New Yorkers. We achieved this through a

5410
05:21:22.250 --> 05:21:24.170
<v Sarah Milstein (OTI)>wide range of digital
products, infrastructure,

5411
05:21:24.766 --> 05:21:26.926
<v Sarah Milstein (OTI)>data systems, and shared expertise.

5412
05:21:27.246 --> 05:21:31.086
<v Sarah Milstein (OTI)>In other words, we use technology to
help agencies make every New Yorker's

5413
05:21:31.086 --> 05:21:32.846
<v Sarah Milstein (OTI)>experience of city government better.

5414
05:21:33.246 --> 05:21:37.566
<v Sarah Milstein (OTI)>For our discussion today, we can speak to
the work we're doing around agency use of

5415
05:21:37.566 --> 05:21:39.727
<v Sarah Milstein (OTI)>algorithmic tools and our cybersecurity

5416
05:21:39.727 --> 05:21:40.125
<v Sarah Milstein (OTI)>program.

5417
05:21:41.460 --> 05:21:44.260
<v Sarah Milstein (OTI)>To touch briefly on the AI arm of OTI,

5418
05:21:44.340 --> 05:21:48.180
<v Sarah Milstein (OTI)>I'm pleased to share that the office the
office of algorithmic accountability,

5419
05:21:48.260 --> 05:21:49.140
<v Sarah Milstein (OTI)>OAA,

5420
05:21:49.380 --> 05:21:53.621
<v Sarah Milstein (OTI)>has been formally established and six new
positions have been created in the f y

5421
05:21:53.621 --> 05:21:58.574
<v Sarah Milstein (OTI)>twenty seven adopted budget to staff it
up. The hiring process for those roles is

5422
05:21:58.574 --> 05:21:59.695
<v Sarah Milstein (OTI)>actively underway.

5423
05:22:00.414 --> 05:22:02.334
<v Sarah Milstein (OTI)>The OAA's duties include

5424
05:22:02.414 --> 05:22:07.135
<v Sarah Milstein (OTI)>analyzing algorithmic tools submitted by
agencies to determine whether there is risk

5425
05:22:07.135 --> 05:22:11.140
<v Sarah Milstein (OTI)>that the proposed tool could result
in discriminatory decision making,

5426
05:22:11.860 --> 05:22:15.300
<v Sarah Milstein (OTI)>conducting and publicly reporting
on predeployment assessments,

5427
05:22:15.860 --> 05:22:20.100
<v Sarah Milstein (OTI)>creating and maintaining a public facing
platform for submission of comments,

5428
05:22:20.420 --> 05:22:24.395
<v Sarah Milstein (OTI)>establishing a protocol with the
Department of Investigation for receiving

5429
05:22:24.395 --> 05:22:25.754
<v Sarah Milstein (OTI)>complaints from the public,

5430
05:22:26.314 --> 05:22:30.635
<v Sarah Milstein (OTI)>promulgating rules establishing basic
compliance standards that all agencies must

5431
05:22:30.635 --> 05:22:35.033
<v Sarah Milstein (OTI)>meet in developing, procuring, deploying,
and use a public impacting artificial

5432
05:22:35.033 --> 05:22:35.594
<v Sarah Milstein (OTI)>intelligence,

5433
05:22:36.541 --> 05:22:40.861
<v Sarah Milstein (OTI)>and reporting the results of predeployment
assessments and audits conducted by the

5434
05:22:40.861 --> 05:22:41.500
<v Sarah Milstein (OTI)>office.

5435
05:22:42.220 --> 05:22:46.140
<v Sarah Milstein (OTI)>Overall, the office is progressing as planned,
and we remain focused on building the

5436
05:22:46.140 --> 05:22:48.701
<v Sarah Milstein (OTI)>capacity needed to carry
out this work effectively.

5437
05:22:49.686 --> 05:22:54.166
<v Sarah Milstein (OTI)>I will now turn to the legislation on
today's docket that would impact OTI.

5438
05:22:55.285 --> 05:22:58.086
<v Sarah Milstein (OTI)>Introduction one six one of 2026

5439
05:22:58.244 --> 05:23:03.060
<v Sarah Milstein (OTI)>would amend annual reporting on algorithmic
tools to include their impact on city

5440
05:23:03.060 --> 05:23:08.182
<v Sarah Milstein (OTI)>employment with a focus on potential changes
to funded agency positions, salaries,

5441
05:23:08.182 --> 05:23:08.900
<v Sarah Milstein (OTI)>and duties.

5442
05:23:10.021 --> 05:23:15.140
<v Sarah Milstein (OTI)>OTI is currently undertaking a study on the
impact of algorithmic tools on municipal

5443
05:23:15.140 --> 05:23:18.637
<v Sarah Milstein (OTI)>employees pursuant to
local law '25 of 2026.

5444
05:23:19.275 --> 05:23:23.035
<v Sarah Milstein (OTI)>This study will provide insights into
the extent to which algorithmic tools,

5445
05:23:23.275 --> 05:23:24.875
<v Sarah Milstein (OTI)>including artificial intelligence,

5446
05:23:25.115 --> 05:23:28.316
<v Sarah Milstein (OTI)>impacts city employees and the
administration of their duties,

5447
05:23:28.635 --> 05:23:30.156
<v Sarah Milstein (OTI)>including their hiring and work functions.

5448
05:23:30.820 --> 05:23:32.580
<v Sarah Milstein (OTI)>While we agree that it is important

5449
05:23:32.740 --> 05:23:36.740
<v Sarah Milstein (OTI)>to examine the impact of AI on our workforce,
the bill as written would not produce

5450
05:23:36.740 --> 05:23:41.621
<v Sarah Milstein (OTI)>new insights beyond those in the larger,
more expansive directive of local law 25.

5451
05:23:43.385 --> 05:23:46.424
<v Sarah Milstein (OTI)>The next three bills involve
New York City Cyber Command.

5452
05:23:46.664 --> 05:23:50.744
<v Sarah Milstein (OTI)>This office plays a vital role in
protecting and defending the city and its

5453
05:23:50.744 --> 05:23:52.984
<v Sarah Milstein (OTI)>residents from the
impacts of cyber attacks.

5454
05:23:53.305 --> 05:23:57.305
<v Sarah Milstein (OTI)>On a day to day basis, we provide
twenty four seven security services

5455
05:23:57.510 --> 05:24:00.630
<v Sarah Milstein (OTI)>and assist agencies in
bolstering their cyber maturity.

5456
05:24:01.110 --> 05:24:05.350
<v Sarah Milstein (OTI)>We work collaboratively with agency partners
as well as state, federal, and private

5457
05:24:05.350 --> 05:24:09.910
<v Sarah Milstein (OTI)>entities to safeguard the essential services
and data New Yorkers depend on daily.

5458
05:24:10.965 --> 05:24:14.645
<v Sarah Milstein (OTI)>Our core mission is to make New York City
the most cyber resilient city in the

5459
05:24:14.645 --> 05:24:17.205
<v Sarah Milstein (OTI)>world, and we take this
extremely seriously.

5460
05:24:17.523 --> 05:24:22.164
<v Sarah Milstein (OTI)>New York City is a target for cyber
attacks with a technology landscape that is

5461
05:24:22.164 --> 05:24:24.244
<v Sarah Milstein (OTI)>unparalleled among
other cities and states.

5462
05:24:24.810 --> 05:24:29.371
<v Sarah Milstein (OTI)>This requires a unified comprehensive
defense against constant cyber threats.

5463
05:24:29.610 --> 05:24:34.090
<v Sarah Milstein (OTI)>Our key duties include setting information
security policies and standards for the

5464
05:24:34.090 --> 05:24:34.650
<v Sarah Milstein (OTI)>city,

5465
05:24:35.050 --> 05:24:38.250
<v Sarah Milstein (OTI)>directing the city's citywide cyber
defense and incident response,

5466
05:24:38.975 --> 05:24:42.016
<v Sarah Milstein (OTI)>deploying defensive technical
and administrative controls,

5467
05:24:42.174 --> 05:24:45.615
<v Sarah Milstein (OTI)>and providing guidance to city hall
and agencies on cyber defense.

5468
05:24:46.094 --> 05:24:49.934
<v Sarah Milstein (OTI)>In my fairly new role at OTI, it is a
privilege to work with our staff and our

5469
05:24:49.934 --> 05:24:52.654
<v Sarah Milstein (OTI)>agency partners in furtherance
of this critical mission.

5470
05:24:53.690 --> 05:24:57.850
<v Sarah Milstein (OTI)>New York City Cyber Command's alignment
within OTI has placed the team in a strong

5471
05:24:57.850 --> 05:25:01.129
<v Sarah Milstein (OTI)>position to monitor and respond
to wide ranging cyber threats.

5472
05:25:01.690 --> 05:25:02.490
<v Sarah Milstein (OTI)>Cybersecurity

5473
05:25:02.490 --> 05:25:06.889
<v Sarah Milstein (OTI)>is continuous work. In my role, I'm
confident we can continue to adapt to a

5474
05:25:06.889 --> 05:25:11.035
<v Sarah Milstein (OTI)>constantly evolving threat landscape. Simply
put, New York City Cyber Command has a

5475
05:25:11.035 --> 05:25:14.715
<v Sarah Milstein (OTI)>twenty four seven apparatus in place to
defend the city systems from malicious

5476
05:25:14.715 --> 05:25:16.715
<v Sarah Milstein (OTI)>attacks regardless of the origin.

5477
05:25:17.195 --> 05:25:21.754
<v Sarah Milstein (OTI)>With that in mind, preconsidered introduction
of twenty twenty six two six zero two

5478
05:25:21.754 --> 05:25:26.270
<v Sarah Milstein (OTI)>would make it unlawful to market, offer
for sale, sell, or deploy an artificial

5479
05:25:26.270 --> 05:25:26.910
<v Sarah Milstein (OTI)>model...

5480
05:25:27.390 --> 05:25:31.230
<v Sarah Milstein (OTI)>An artificial intelligence model in the
New York... In New York City that has not

5481
05:25:31.230 --> 05:25:35.549
<v Sarah Milstein (OTI)>received third party validation or does
not have a technical capability to be shut

5482
05:25:35.549 --> 05:25:36.830
<v Sarah Milstein (OTI)>down by a human operator.

5483
05:25:38.166 --> 05:25:41.846
<v Sarah Milstein (OTI)>The bill would require the director of
the Office of Cyber Command to promulgate

5484
05:25:41.846 --> 05:25:45.605
<v Sarah Milstein (OTI)>rules regarding what the third party
valid... What the third party validation,

5485
05:25:45.605 --> 05:25:49.926
<v Sarah Milstein (OTI)>assessments, certifications, and disclosure
would entail and the qualifications for

5486
05:25:49.926 --> 05:25:51.047
<v Sarah Milstein (OTI)>third party validators.

5487
05:25:51.550 --> 05:25:55.791
<v Sarah Milstein (OTI)>To be clear, Cyber Command's core mission
is to defend the cities of New York's

5488
05:25:55.791 --> 05:25:57.630
<v Sarah Milstein (OTI)>infrastructure from cyber attacks.

5489
05:25:58.111 --> 05:26:02.270
<v Sarah Milstein (OTI)>We are not subject matter experts on the
qualifications for an entity to be to be

5490
05:26:02.270 --> 05:26:03.951
<v Sarah Milstein (OTI)>considered a third party validator.

5491
05:26:04.426 --> 05:26:07.146
<v Sarah Milstein (OTI)>Further, nor... Neither an OT... OTI

5492
05:26:07.146 --> 05:26:10.186
<v Sarah Milstein (OTI)>nor Cyber Command have
experience relegating...

5493
05:26:10.186 --> 05:26:11.146
<v Sarah Milstein (OTI)>Regulating

5494
05:26:11.225 --> 05:26:15.465
<v Sarah Milstein (OTI)>the private sector in the way that this
legislation would require and neither would

5495
05:26:15.465 --> 05:26:19.691
<v Sarah Milstein (OTI)>be the best choice to lead this regulatory
regime for consumer products. We defer to

5496
05:26:19.691 --> 05:26:20.650
<v Sarah Milstein (OTI)>DCWP

5497
05:26:20.650 --> 05:26:24.170
<v Sarah Milstein (OTI)>to speak about the private sector
policy and enforcement piece.

5498
05:26:24.970 --> 05:26:26.650
<v Sarah Milstein (OTI)>To that end, DCWP

5499
05:26:26.650 --> 05:26:28.730
<v Sarah Milstein (OTI)>supports the intent of this legislation.

5500
05:26:28.810 --> 05:26:31.611
<v Sarah Milstein (OTI)>They recommend including robust
record keeping requirements

5501
05:26:31.744 --> 05:26:34.945
<v Sarah Milstein (OTI)>to better allow the agency to
conduct affirmative enforcement.

5502
05:26:35.184 --> 05:26:39.744
<v Sarah Milstein (OTI)>They also note based on conversations with
advocates and sister agencies that the

5503
05:26:39.744 --> 05:26:43.504
<v Sarah Milstein (OTI)>administration is not certain about the
availability or ability of such businesses.

5504
05:26:44.020 --> 05:26:46.340
<v Sarah Milstein (OTI)>We look forward to hearing
from those stakeholders

5505
05:26:46.420 --> 05:26:47.140
<v Sarah Milstein (OTI)>today,

5506
05:26:47.459 --> 05:26:51.459
<v Sarah Milstein (OTI)>including on whether initiatives like this
one could help spur a market for such

5507
05:26:51.459 --> 05:26:52.180
<v Sarah Milstein (OTI)>services.

5508
05:26:52.979 --> 05:26:56.500
<v Sarah Milstein (OTI)>Preconsidered introduction of
twenty twenty six two six zero one

5509
05:26:57.145 --> 05:27:00.984
<v Sarah Milstein (OTI)>require the Office of Cyber Command to
establish standards and procedures for

5510
05:27:00.984 --> 05:27:05.385
<v Sarah Milstein (OTI)>contractors to identify the occurrence
of a reportable AI safety incident.

5511
05:27:06.023 --> 05:27:10.340
<v Sarah Milstein (OTI)>To take a step back, I'd like to lay out
how AI risk and incidents are categorized

5512
05:27:10.340 --> 05:27:14.100
<v Sarah Milstein (OTI)>according to the National Institute
for Standards and Technology, NIST.

5513
05:27:14.738 --> 05:27:20.180
<v Sarah Milstein (OTI)>Under a... AI under attack, AI cybersecurity
incidents are traditional cybersecurity

5514
05:27:20.180 --> 05:27:22.340
<v Sarah Milstein (OTI)>compromises where the AI system

5515
05:27:22.340 --> 05:27:25.459
<v Sarah Milstein (OTI)>is the target. For example,
data poisoning or jailbreaks.

5516
05:27:26.365 --> 05:27:27.885
<v Sarah Milstein (OTI)>Misuse or malfunction,

5517
05:27:27.965 --> 05:27:29.484
<v Sarah Milstein (OTI)>AI induced incidents,

5518
05:27:29.885 --> 05:27:33.885
<v Sarah Milstein (OTI)>would be harms resulting from the use
or failure of AI aside from a cyber

5519
05:27:33.885 --> 05:27:35.885
<v Sarah Milstein (OTI)>compromise. In this category,

5520
05:27:36.125 --> 05:27:40.951
<v Sarah Milstein (OTI)>the AI system itself behaves exactly as it
was built or trained, but the real world

5521
05:27:40.951 --> 05:27:42.470
<v Sarah Milstein (OTI)>outcome causes harm.

5522
05:27:42.710 --> 05:27:45.031
<v Sarah Milstein (OTI)>These incidents are not cyberattacks.

5523
05:27:45.670 --> 05:27:49.510
<v Sarah Milstein (OTI)>We understand the council's intent
here given the publicized incidents...

5524
05:27:49.670 --> 05:27:53.590
<v Sarah Milstein (OTI)>Instances of AI systems operating
outside the bounds of their intended use.

5525
05:27:54.484 --> 05:27:58.084
<v Sarah Milstein (OTI)>As I noted earlier, the risk
landscape is constantly evolving,

5526
05:27:58.244 --> 05:28:01.443
<v Sarah Milstein (OTI)>and we are continuing to evolve our
strategies to keep the city's infrastructure

5527
05:28:01.443 --> 05:28:02.164
<v Sarah Milstein (OTI)>safe.

5528
05:28:02.564 --> 05:28:06.965
<v Sarah Milstein (OTI)>But not all threats related to ace... AI
systems are necessarily cyber attacks,

5529
05:28:07.285 --> 05:28:09.684
<v Sarah Milstein (OTI)>and the legislation as
written does not address that.

5530
05:28:10.310 --> 05:28:13.910
<v Sarah Milstein (OTI)>Cyber command would be a critical
stakeholder in this discussion, but it is

5531
05:28:13.910 --> 05:28:17.590
<v Sarah Milstein (OTI)>important to emphasize that we need to work
with our state and federal partners to

5532
05:28:17.590 --> 05:28:19.352
<v Sarah Milstein (OTI)>address these issues holistically.

5533
05:28:20.070 --> 05:28:23.990
<v Sarah Milstein (OTI)>I also wanna provide further context
on breaches of security generally.

5534
05:28:24.436 --> 05:28:29.475
<v Sarah Milstein (OTI)>Agency disclosure of a breach of security
in any form is governed by section ten

5535
05:28:29.475 --> 05:28:32.035
<v Sarah Milstein (OTI)>five zero two of the administrative code.

5536
05:28:32.355 --> 05:28:36.516
<v Sarah Milstein (OTI)>There are extensive protocols for
interagency coordination in the event of any

5537
05:28:36.516 --> 05:28:37.395
<v Sarah Milstein (OTI)>breach of security.

5538
05:28:38.670 --> 05:28:42.910
<v Sarah Milstein (OTI)>These protocols are aligned with the
New York State Stop Hacks and Improve

5539
05:28:42.910 --> 05:28:45.709
<v Sarah Milstein (OTI)>Electronic Data Security Shield Act.

5540
05:28:46.110 --> 05:28:50.910
<v Sarah Milstein (OTI)>Current law policies and protocols are
in place for incidents described in the

5541
05:28:50.910 --> 05:28:51.389
<v Sarah Milstein (OTI)>legislation.

5542
05:28:52.715 --> 05:28:57.115
<v Sarah Milstein (OTI)>Further, the threshold of twenty four
hour public notice is inconsistent with

5543
05:28:57.115 --> 05:29:01.516
<v Sarah Milstein (OTI)>current law and would compromise the
city's investigation of a cyber incident.

5544
05:29:03.035 --> 05:29:08.030
<v Sarah Milstein (OTI)>Preconsidered introduction of twenty twenty
six two six zero six would require the

5545
05:29:08.030 --> 05:29:09.470
<v Sarah Milstein (OTI)>office of cyber command

5546
05:29:09.630 --> 05:29:11.470
<v Sarah Milstein (OTI)>in coordination with NISM

5547
05:29:11.550 --> 05:29:16.750
<v Sarah Milstein (OTI)>to develop or update an emergency response
plan concerning AI system related threats

5548
05:29:16.750 --> 05:29:18.430
<v Sarah Milstein (OTI)>to New York City's infrastructure,

5549
05:29:18.510 --> 05:29:19.310
<v Sarah Milstein (OTI)>operations,

5550
05:29:19.310 --> 05:29:20.670
<v Sarah Milstein (OTI)>public health, and welfare.

5551
05:29:21.494 --> 05:29:25.814
<v Sarah Milstein (OTI)>I'd once again like to emphasize that
there are laws, policies, and protocols in

5552
05:29:25.814 --> 05:29:29.174
<v Sarah Milstein (OTI)>place for breaches and
security regardless of actor.

5553
05:29:29.494 --> 05:29:33.895
<v Sarah Milstein (OTI)>However, we are happy to discuss more broadly
with council how we already coordinate

5554
05:29:33.895 --> 05:29:37.334
<v Sarah Milstein (OTI)>with NISM to coordinate and engage
with partners to restore services.

5555
05:29:38.590 --> 05:29:39.390
<v Sarah Milstein (OTI)>Finally,

5556
05:29:39.470 --> 05:29:41.150
<v Sarah Milstein (OTI)>I will provide DCWP's

5557
05:29:41.150 --> 05:29:45.070
<v Sarah Milstein (OTI)>feedback on the remaining legislation
that would impact the administration.

5558
05:29:46.111 --> 05:29:49.150
<v Sarah Milstein (OTI)>Preconsidered introduction to 2026,

5559
05:29:49.150 --> 05:29:54.295
<v Sarah Milstein (OTI)>two six zero five, would allow any
person to submit a complaint to DCWP

5560
05:29:54.295 --> 05:29:59.016
<v Sarah Milstein (OTI)>alleging that a person or entity has violated
certain provisions of law related to

5561
05:29:59.016 --> 05:30:00.375
<v Sarah Milstein (OTI)>artificial intelligence.

5562
05:30:00.855 --> 05:30:02.215
<v Sarah Milstein (OTI)>DCWP

5563
05:30:02.215 --> 05:30:06.690
<v Sarah Milstein (OTI)>strongly supports the intent of the... This
bill to establish a civilian enforcement

5564
05:30:06.690 --> 05:30:08.530
<v Sarah Milstein (OTI)>structure for AI violations,

5565
05:30:08.770 --> 05:30:13.170
<v Sarah Milstein (OTI)>especially given widespread reporting that
concerns being raised by AI companies

5566
05:30:13.170 --> 05:30:15.090
<v Sarah Milstein (OTI)>employees are being ignored.

5567
05:30:15.889 --> 05:30:19.570
<v Sarah Milstein (OTI)>However, there are fiscal and operational
concerns with the volume of complaints

5568
05:30:19.570 --> 05:30:23.705
<v Sarah Milstein (OTI)>that the agency would receive and the
resources and expertise required to handle

5569
05:30:23.705 --> 05:30:25.145
<v Sarah Milstein (OTI)>each individual complaint.

5570
05:30:25.783 --> 05:30:29.705
<v Sarah Milstein (OTI)>Therefore, the administration believes it
would be important to coordinate closely

5571
05:30:29.705 --> 05:30:33.682
<v Sarah Milstein (OTI)>with mission aligned enforcement agencies
at the state level. The New York state

5572
05:30:33.682 --> 05:30:37.842
<v Sarah Milstein (OTI)>attorney general just recently announced
its secure whistleblower portal for New

5573
05:30:37.842 --> 05:30:41.842
<v Sarah Milstein (OTI)>Yorkers to report unsafe and illegal
conduct related to AI technology.

5574
05:30:42.320 --> 05:30:46.240
<v Sarah Milstein (OTI)>This could be an opportunity to partner on
this existing program to more effectively

5575
05:30:46.240 --> 05:30:48.560
<v Sarah Milstein (OTI)>enforce complaints on
behalf of New Yorkers.

5576
05:30:49.656 --> 05:30:54.295
<v Sarah Milstein (OTI)>Preconsidered introduction twenty twenty
six two five nine nine would impose

5577
05:30:54.295 --> 05:30:57.656
<v Sarah Milstein (OTI)>requirements and restrictions on
companies that provide chatbots,

5578
05:30:57.895 --> 05:31:00.535
<v Sarah Milstein (OTI)>including requirements related
to data security and privacy,

5579
05:31:01.030 --> 05:31:05.670
<v Sarah Milstein (OTI)>the right of chatbot users to access their
own data, and restrictions on how chatbot

5580
05:31:05.670 --> 05:31:08.070
<v Sarah Milstein (OTI)>providers can use chatbot user data.

5581
05:31:08.709 --> 05:31:09.750
<v Sarah Milstein (OTI)>DCWP

5582
05:31:09.750 --> 05:31:13.988
<v Sarah Milstein (OTI)>has concerns about this bill. While we
support council's goal to better regulate

5583
05:31:13.988 --> 05:31:16.229
<v Sarah Milstein (OTI)>companies deployment of AI chatbots

5584
05:31:16.576 --> 05:31:19.217
<v Sarah Milstein (OTI)>and believe the approach
taken by this bill is sound,

5585
05:31:19.535 --> 05:31:23.217
<v Sarah Milstein (OTI)>the implementation and enforcement of
this bill's data security and privacy

5586
05:31:23.217 --> 05:31:26.336
<v Sarah Milstein (OTI)>requirements would require a
deep technical understanding

5587
05:31:26.416 --> 05:31:29.695
<v Sarah Milstein (OTI)>and investigative ability that is
outside of the agency's scope.

5588
05:31:30.710 --> 05:31:32.710
<v Sarah Milstein (OTI)>To be clear, DCWP

5589
05:31:32.710 --> 05:31:35.191
<v Sarah Milstein (OTI)>is committed to regulating
harmful practices,

5590
05:31:35.430 --> 05:31:39.750
<v Sarah Milstein (OTI)>including relating to chatbots that fall
within the agency's investigative ability.

5591
05:31:40.230 --> 05:31:44.244
<v Sarah Milstein (OTI)>To that end, the administration is working
on rules related... On rules related to

5592
05:31:44.244 --> 05:31:46.084
<v Sarah Milstein (OTI)>the use of AI chatbots

5593
05:31:46.084 --> 05:31:48.404
<v Sarah Milstein (OTI)>to better regulate the
space for consumers.

5594
05:31:49.045 --> 05:31:51.924
<v Sarah Milstein (OTI)>The administration has also
invested in supporting

5595
05:31:51.924 --> 05:31:55.920
<v Sarah Milstein (OTI)>the attorney general... The New York
attorney general in their efforts to rein in

5596
05:31:55.920 --> 05:31:59.121
<v Sarah Milstein (OTI)>data abuses for which this
bill may serve as a model.

5597
05:32:00.240 --> 05:32:04.480
<v Sarah Milstein (OTI)>Preconsidered introduction of twenty twenty
six two six zero three would require any

5598
05:32:04.480 --> 05:32:07.205
<v Sarah Milstein (OTI)>advertisement that promotes
an artificial intelligence

5599
05:32:07.205 --> 05:32:12.484
<v Sarah Milstein (OTI)>AI model in the city to disclose whether
such a model has undergone validation by a

5600
05:32:12.484 --> 05:32:13.285
<v Sarah Milstein (OTI)>third party.

5601
05:32:14.006 --> 05:32:18.645
<v Sarah Milstein (OTI)>It also would prohibit any such advertisement
from including any material... Any

5602
05:32:18.645 --> 05:32:20.965
<v Sarah Milstein (OTI)>materially false or misleading statement

5603
05:32:21.170 --> 05:32:23.330
<v Sarah Milstein (OTI)>regarding the safety of an AI model.

5604
05:32:23.570 --> 05:32:24.690
<v Sarah Milstein (OTI)>DCWP

5605
05:32:24.690 --> 05:32:27.090
<v Sarah Milstein (OTI)>supports this bill to require disclosure

5606
05:32:27.410 --> 05:32:29.010
<v Sarah Milstein (OTI)>of third party validation

5607
05:32:29.170 --> 05:32:31.570
<v Sarah Milstein (OTI)>of any advertisement of an AI model.

5608
05:32:31.810 --> 05:32:35.250
<v Sarah Milstein (OTI)>However, as mentioned above, the
administration is not certain about the

5609
05:32:35.250 --> 05:32:35.810
<v Sarah Milstein (OTI)>availability

5610
05:32:36.125 --> 05:32:37.965
<v Sarah Milstein (OTI)>or abilities of such businesses,

5611
05:32:38.205 --> 05:32:41.484
<v Sarah Milstein (OTI)>and we want to be clear that any
advertising that uses false or misleading

5612
05:32:41.484 --> 05:32:46.125
<v Sarah Milstein (OTI)>statements is currently prohibited under
the city's consumer protection law and will

5613
05:32:46.125 --> 05:32:48.365
<v Sarah Milstein (OTI)>be enforced with the full scope of DCWP's

5614
05:32:48.365 --> 05:32:48.686
<v Sarah Milstein (OTI)>authority.

5615
05:32:49.550 --> 05:32:53.310
<v Sarah Milstein (OTI)>We look forward to working with counsel
to ensure the law includes effective

5616
05:32:53.310 --> 05:32:57.550
<v Sarah Milstein (OTI)>enforcement mechanisms and to continue
addressing issues within DCWP's

5617
05:32:57.550 --> 05:32:59.871
<v Sarah Milstein (OTI)>scope of work as it relates to AI.

5618
05:33:00.111 --> 05:33:03.070
<v Sarah Milstein (OTI)>Thank you for the time. This
panel will now take questions.

5619
05:33:03.230 --> 05:33:07.814
<v Speaker Julie Menin>Great. Thank you so much for your testimony.
We appreciate it. I've got a number of

5620
05:33:07.814 --> 05:33:13.416
<v Speaker Julie Menin>questions. So OpenAI sent the city
council a letter on October 1 recommending

5621
05:33:13.416 --> 05:33:15.736
<v Speaker Julie Menin>that New York City better connect

5622
05:33:15.895 --> 05:33:19.975
<v Speaker Julie Menin>its existing cyber and emergency
management capabilities,

5623
05:33:19.975 --> 05:33:23.840
<v Speaker Julie Menin>and I'm gonna... Just reading from the
letter, with the escalation, containment, and

5624
05:33:23.840 --> 05:33:27.201
<v Speaker Julie Menin>recovery capabilities of
frontier AI companies.

5625
05:33:27.520 --> 05:33:30.800
<v Speaker Julie Menin>So has the city of New York
this year met with OpenAI,

5626
05:33:30.800 --> 05:33:35.426
<v Speaker Julie Menin>Anthropic, or any other frontier AI company
to discuss how the city would coordinate

5627
05:33:35.426 --> 05:33:40.225
<v Speaker Julie Menin>with them in the event of a serious AI
related cyber or public safety incident?

5628
05:33:47.960 --> 05:33:51.960
<v Sarah Milstein (OTI)>Thank you for the question. I'm going
to defer to my colleague, CJ Dixon.

5629
05:33:52.600 --> 05:33:55.960
<v CJ Dixon (NYC Cyber Command / OTI)>Thank you. So I would like to note
that New York City Cyber Command

5630
05:33:56.520 --> 05:34:01.174
<v CJ Dixon (NYC Cyber Command / OTI)>follows the NIST cybersecurity framework
for all of our cyber programming that is

5631
05:34:01.174 --> 05:34:04.615
<v CJ Dixon (NYC Cyber Command / OTI)>everything from identify, protect,
detect, respond, and recover.

5632
05:34:04.775 --> 05:34:08.615
<v CJ Dixon (NYC Cyber Command / OTI)>The basic first principles in cybersecurity
have not changed even with the advent of

5633
05:34:08.615 --> 05:34:09.895
<v CJ Dixon (NYC Cyber Command / OTI)>artificial intelligence.

5634
05:34:10.375 --> 05:34:14.055
<v CJ Dixon (NYC Cyber Command / OTI)>And we have been in conversations with
various frontier models about how we may

5635
05:34:14.055 --> 05:34:17.920
<v CJ Dixon (NYC Cyber Command / OTI)>integrate their tools and software to
to better enhance the functions that we

5636
05:34:17.920 --> 05:34:22.640
<v CJ Dixon (NYC Cyber Command / OTI)>already deliver to the city. So in reference
to your question, yes, we have been in

5637
05:34:22.640 --> 05:34:25.121
<v CJ Dixon (NYC Cyber Command / OTI)>conversation with the
frontier AI companies

5638
05:34:25.360 --> 05:34:27.760
<v CJ Dixon (NYC Cyber Command / OTI)>to enhance the way that
we perform cybersecurity.

5639
05:34:27.760 --> 05:34:31.975
<v CJ Dixon (NYC Cyber Command / OTI)>However, those basic cybersecurity controls
have functionally remained the same. So

5640
05:34:31.975 --> 05:34:37.174
<v CJ Dixon (NYC Cyber Command / OTI)>you've met with OpenAI and Anthropic in
particular? We have met with Anthropic. I

5641
05:34:37.174 --> 05:34:40.375
<v CJ Dixon (NYC Cyber Command / OTI)>have not met with OpenAI since
coming on board on May 5.

5642
05:34:41.094 --> 05:34:41.654
<v Speaker Julie Menin>Okay.

5643
05:34:42.375 --> 05:34:44.295
<v Speaker Julie Menin>Are you planning on meeting with OpenAI?

5644
05:34:45.701 --> 05:34:47.861
<v CJ Dixon (NYC Cyber Command / OTI)>Only based on a... Only

5645
05:34:48.420 --> 05:34:53.621
<v CJ Dixon (NYC Cyber Command / OTI)>dependent on our needs right now. We do
not know definitively if there is a valid

5646
05:34:53.621 --> 05:34:56.422
<v CJ Dixon (NYC Cyber Command / OTI)>reason for us to meet with any frontier

5647
05:34:56.500 --> 05:35:01.576
<v Speaker Julie Menin>AI company in the service concerned of by
the statement that nothing has changed in

5648
05:35:01.576 --> 05:35:06.055
<v Speaker Julie Menin>your protocols given that we're now dealing
with a rapidly escalating technology

5649
05:35:06.055 --> 05:35:09.256
<v Speaker Julie Menin>that as we certainly heard today
and we've been hearing now for

5650
05:35:10.215 --> 05:35:12.695
<v Speaker Julie Menin>the last couple weeks to
months that there are

5651
05:35:13.510 --> 05:35:19.750
<v Speaker Julie Menin>serious safety concerns. So I just wanna
get an answer in terms of what proactive

5652
05:35:19.750 --> 05:35:21.750
<v Speaker Julie Menin>steps is cybersecurity

5653
05:35:21.750 --> 05:35:22.549
<v Speaker Julie Menin>taking

5654
05:35:22.709 --> 05:35:27.385
<v Speaker Julie Menin>to meet with some of these frontier companies
and ensure that our systems are safe.

5655
05:35:27.785 --> 05:35:32.824
<v CJ Dixon (NYC Cyber Command / OTI)>Acknowledged. And we... It does not mean
that we have not changed our protocols.

5656
05:35:32.824 --> 05:35:36.744
<v CJ Dixon (NYC Cyber Command / OTI)>Some of the risks in hand with
artificial intelligence are unrelated to

5657
05:35:36.744 --> 05:35:37.625
<v CJ Dixon (NYC Cyber Command / OTI)>cybersecurity.

5658
05:35:38.344 --> 05:35:41.945
<v CJ Dixon (NYC Cyber Command / OTI)>In those situations where an AI
model presents a cybersecurity risk,

5659
05:35:42.371 --> 05:35:46.771
<v CJ Dixon (NYC Cyber Command / OTI)>those AI models will still exploit the
exact same software vulnerabilities

5660
05:35:46.771 --> 05:35:48.932
<v CJ Dixon (NYC Cyber Command / OTI)>that are inherent in computer technology

5661
05:35:48.932 --> 05:35:54.050
<v CJ Dixon (NYC Cyber Command / OTI)>as they've always been. So our protocols
have not had to change to address the

5662
05:35:54.050 --> 05:35:55.410
<v CJ Dixon (NYC Cyber Command / OTI)>fundamental issues

5663
05:35:55.725 --> 05:36:00.125
<v Speaker Julie Menin>inherent in computer technology, only the
speed by which we must react in a cyber

5664
05:36:00.125 --> 05:36:06.523
<v Speaker Julie Menin>attack? Okay. I I am concerned by news that
city and state reported on Friday that

5665
05:36:06.523 --> 05:36:11.951
<v Speaker Julie Menin>the AI czar, Jaihao Chen, who is our
chief technology officer, resigned,

5666
05:36:11.951 --> 05:36:13.951
<v Speaker Julie Menin>I think, a week to week and a half ago

5667
05:36:14.510 --> 05:36:18.352
<v Speaker Julie Menin>and reports saying that he
resigned in frustration that

5668
05:36:18.750 --> 05:36:20.191
<v Speaker Julie Menin>a lot of positions

5669
05:36:20.350 --> 05:36:23.951
<v Speaker Julie Menin>were being not hired or
being held up by OMB.

5670
05:36:24.190 --> 05:36:28.145
<v Speaker Julie Menin>And that relates to my second question,
which is council member Gutierrez.

5671
05:36:28.145 --> 05:36:30.225
<v Speaker Julie Menin>I think she's still here. Yes. And I,

5672
05:36:30.705 --> 05:36:31.904
<v Speaker Julie Menin>passed bills

5673
05:36:32.064 --> 05:36:33.186
<v Speaker Julie Menin>last year

5674
05:36:33.584 --> 05:36:37.105
<v Speaker Julie Menin>to create the office of
algorithmic accountability.

5675
05:36:37.824 --> 05:36:42.291
<v Speaker Julie Menin>And so it appears that mister Chen's
resignation from what we're reading in the

5676
05:36:42.291 --> 05:36:46.691
<v Speaker Julie Menin>press was related to that and the fact
that that office was supposed to be up and

5677
05:36:46.691 --> 05:36:48.371
<v Speaker Julie Menin>running by the city in June,

5678
05:36:48.531 --> 05:36:50.371
<v Speaker Julie Menin>but it is not up and running.

5679
05:36:50.611 --> 05:36:54.050
<v Speaker Julie Menin>And in fact, in your testimony, you said
you're now posting for the positions. It's

5680
05:36:54.050 --> 05:36:54.691
<v Speaker Julie Menin>now October.

5681
05:36:56.254 --> 05:36:58.574
<v Sarah Milstein (OTI)>Thank you for the the question,

5682
05:36:59.135 --> 05:36:59.775
<v Sarah Milstein (OTI)>speaker.

5683
05:37:00.734 --> 05:37:06.494
<v Sarah Milstein (OTI)>Ji Hao Chen served as director of AI and
ML at OTA... At OTI for two and a half

5684
05:37:06.494 --> 05:37:10.190
<v Sarah Milstein (OTI)>years. He did important work
in the previous administration

5685
05:37:10.190 --> 05:37:14.670
<v Sarah Milstein (OTI)>to help define and understand
AI use throughout the city.

5686
05:37:15.070 --> 05:37:19.710
<v Sarah Milstein (OTI)>He's been very helpful in getting new OTI
administration up to speed, and we wish

5687
05:37:19.710 --> 05:37:20.350
<v Sarah Milstein (OTI)>him the best.

5688
05:37:21.396 --> 05:37:25.477
<v Sarah Milstein (OTI)>We are very excited,
to be staffing up OAA.

5689
05:37:25.955 --> 05:37:28.916
<v Sarah Milstein (OTI)>I... The the law went
into effect in December.

5690
05:37:29.477 --> 05:37:34.916
<v Sarah Milstein (OTI)>I was brought in this summer to help
stand up the office. We got funding

5691
05:37:35.076 --> 05:37:35.717
<v Sarah Milstein (OTI)>in

5692
05:37:36.030 --> 05:37:36.830
<v Sarah Milstein (OTI)>June

5693
05:37:37.068 --> 05:37:39.870
<v Sarah Milstein (OTI)>for... In the f y twenty seven budget,

5694
05:37:40.270 --> 05:37:43.230
<v Sarah Milstein (OTI)>and we've been actively
working to hire for the roles,

5695
05:37:43.709 --> 05:37:47.950
<v Sarah Milstein (OTI)>for the director role for OAA and look
forward to announcing something quite soon.

5696
05:37:47.950 --> 05:37:48.670
<v Sarah Milstein (OTI)>We're really,

5697
05:37:49.068 --> 05:37:53.514
<v Sarah Milstein (OTI)>grateful to the council for having stood
up, for having helped us stand this up. I

5698
05:37:53.514 --> 05:37:55.275
<v Sarah Milstein (OTI)>think the work is gonna be excellent.

5699
05:37:55.436 --> 05:37:59.834
<v Speaker Julie Menin>It just seems that this is an
inopportune time for his resignation,

5700
05:37:59.834 --> 05:38:04.580
<v Speaker Julie Menin>and I am concerned that the office
was under council member Gutierrez's

5701
05:38:04.580 --> 05:38:08.262
<v Speaker Julie Menin>bill, supposed to be up and
running in June. In addition,

5702
05:38:08.740 --> 05:38:14.340
<v Speaker Julie Menin>the companion bill, which was my bill,
which we passed a year ago, required OTI to

5703
05:38:14.340 --> 05:38:18.564
<v Speaker Julie Menin>engage in rule making around
this office of algorithmic

5704
05:38:18.564 --> 05:38:19.525
<v Speaker Julie Menin>accountability.

5705
05:38:20.006 --> 05:38:24.326
<v Speaker Julie Menin>But no... OTI did no rule making whatsoever,
so it did not comply with that bill.

5706
05:38:24.326 --> 05:38:26.006
<v Speaker Julie Menin>Instead, OTI issued

5707
05:38:26.166 --> 05:38:27.445
<v Speaker Julie Menin>updated guidance,

5708
05:38:27.445 --> 05:38:28.885
<v Speaker Julie Menin>ignored the rule making process.

5709
05:38:29.320 --> 05:38:33.640
<v Speaker Julie Menin>There's been no proposed rule, no public
hearing, no public comment period, and no

5710
05:38:33.640 --> 05:38:36.041
<v Speaker Julie Menin>adoption into the rules
of the city Of New York.

5711
05:38:37.480 --> 05:38:38.441
<v Sarah Milstein (OTI)>Thank

5712
05:38:39.080 --> 05:38:39.560
<v Sarah Milstein (OTI)>you.

5713
05:38:39.880 --> 05:38:41.720
<v Sarah Milstein (OTI)>Thank you for the the question,

5714
05:38:41.720 --> 05:38:42.121
<v Sarah Milstein (OTI)>speaker.

5715
05:38:42.846 --> 05:38:47.086
<v Sarah Milstein (OTI)>So OTI is fully compliant
with the law. The

5716
05:38:48.605 --> 05:38:49.406
<v Sarah Milstein (OTI)>rulemaking

5717
05:38:49.406 --> 05:38:50.766
<v Sarah Milstein (OTI)>piece of the law,

5718
05:38:51.564 --> 05:38:52.525
<v Sarah Milstein (OTI)>had us,

5719
05:38:53.246 --> 05:38:54.285
<v Sarah Milstein (OTI)>this

5720
05:38:54.365 --> 05:38:57.086
<v Sarah Milstein (OTI)>year, submit the initial

5721
05:38:58.630 --> 05:38:59.830
<v Sarah Milstein (OTI)>the initial,

5722
05:39:01.432 --> 05:39:06.230
<v Sarah Milstein (OTI)>I don't wanna say paperwork, but the
initial points for our rulemaking

5723
05:39:06.230 --> 05:39:09.432
<v Sarah Milstein (OTI)>and the actual rulemaking will
happen in this fiscal year.

5724
05:39:09.830 --> 05:39:10.951
<v Sarah Milstein (OTI)>We are also,

5725
05:39:11.271 --> 05:39:13.670
<v Sarah Milstein (OTI)>fully compliant with local law 35,

5726
05:39:13.955 --> 05:39:18.475
<v Sarah Milstein (OTI)>which was brought under the office
of OAA as part of the December

5727
05:39:18.475 --> 05:39:19.395
<v Sarah Milstein (OTI)>25

5728
05:39:19.395 --> 05:39:20.195
<v Sarah Milstein (OTI)>law.

5729
05:39:20.836 --> 05:39:23.635
<v Sarah Milstein (OTI)>We've been in the process of,

5730
05:39:24.115 --> 05:39:28.940
<v Sarah Milstein (OTI)>fulfilling that report for five years.
We've done it on time for five years in a

5731
05:39:28.940 --> 05:39:32.621
<v Sarah Milstein (OTI)>row. That's the law that requires agencies

5732
05:39:32.621 --> 05:39:33.180
<v Sarah Milstein (OTI)>to,

5733
05:39:34.300 --> 05:39:37.660
<v Sarah Milstein (OTI)>disclose their use of public
impacting algorithmic tools.

5734
05:39:38.860 --> 05:39:40.380
<v Sarah Milstein (OTI)>And we're on track now

5735
05:39:40.541 --> 05:39:41.100
<v Sarah Milstein (OTI)>for,

5736
05:39:41.500 --> 05:39:43.420
<v Sarah Milstein (OTI)>on time reporting for

5737
05:39:44.035 --> 05:39:44.914
<v Sarah Milstein (OTI)>March

5738
05:39:44.914 --> 05:39:45.793
<v Sarah Milstein (OTI)>2027.

5739
05:39:45.795 --> 05:39:48.914
<v Sarah Milstein (OTI)>So we're really excited actually with
doing that work. Very proud of it. We just

5740
05:39:48.914 --> 05:39:53.635
<v Sarah Milstein (OTI)>last week sent out the email
to, all of the city agencies

5741
05:39:53.635 --> 05:39:58.291
<v Sarah Milstein (OTI)>to kick off this reporting cycle and are
holding on this Thursday the orientation

5742
05:39:58.291 --> 05:40:00.451
<v Sarah Milstein (OTI)>meeting that gets everybody rolling.

5743
05:40:00.611 --> 05:40:04.850
<v Speaker Julie Menin>Okay. I mean, just to clarify for the
record, that... We don't believe it... That's

5744
05:40:04.850 --> 05:40:08.291
<v Speaker Julie Menin>in compliance at all. The office was supposed
to be up and running in June. I'm sure

5745
05:40:08.291 --> 05:40:12.064
<v Speaker Julie Menin>council member Gutierrez will talk about
her bill, the office was not up and running

5746
05:40:12.064 --> 05:40:14.385
<v Speaker Julie Menin>in June and was supposed
to do the rulemaking.

5747
05:40:14.705 --> 05:40:17.506
<v Speaker Julie Menin>When do you feel then the
law department will be...

5748
05:40:17.904 --> 05:40:21.346
<v Speaker Julie Menin>Will... You've mentioned law department.
When will... Can we expect to actually see

5749
05:40:21.346 --> 05:40:22.385
<v Speaker Julie Menin>the rulemaking?

5750
05:40:22.465 --> 05:40:25.266
<v Speaker Julie Menin>This is in the March 27 deadline?

5751
05:40:25.266 --> 05:40:26.660
<v Sarah Milstein (OTI)>Yeah. Okay.

5752
05:40:27.139 --> 05:40:30.180
<v Speaker Julie Menin>So what safeguards are
currently in place to prevent,

5753
05:40:30.740 --> 05:40:32.660
<v Speaker Julie Menin>city employees from entering,

5754
05:40:33.700 --> 05:40:38.820
<v Speaker Julie Menin>confidential private city data, whether
authorized or unauthorized, into AI systems

5755
05:40:38.820 --> 05:40:43.174
<v Speaker Julie Menin>such as Co Pilot that are being
controlled by outside corporations?

5756
05:40:45.014 --> 05:40:46.455
<v Sarah Milstein (OTI)>Thank you for the question.

5757
05:40:47.334 --> 05:40:47.975
<v Sarah Milstein (OTI)>Our

5758
05:40:48.215 --> 05:40:50.055
<v Sarah Milstein (OTI)>contract with Microsoft,

5759
05:40:50.375 --> 05:40:52.375
<v Sarah Milstein (OTI)>which runs Co Pilot Chat,

5760
05:40:53.930 --> 05:40:55.932
<v Sarah Milstein (OTI)>prevents them from

5761
05:40:56.570 --> 05:40:57.371
<v Sarah Milstein (OTI)>using

5762
05:40:57.531 --> 05:41:01.531
<v Sarah Milstein (OTI)>city worker data in...
To train their models.

5763
05:41:01.850 --> 05:41:03.210
<v Sarah Milstein (OTI)>And the

5764
05:41:03.690 --> 05:41:08.016
<v Sarah Milstein (OTI)>the way they host data for us also is

5765
05:41:08.016 --> 05:41:09.936
<v Sarah Milstein (OTI)>ensures that we're HIPAA compliant,

5766
05:41:10.814 --> 05:41:12.256
<v Sarah Milstein (OTI)>CRISP

5767
05:41:12.256 --> 05:41:13.055
<v Sarah Milstein (OTI)>compliant,

5768
05:41:13.295 --> 05:41:15.615
<v Sarah Milstein (OTI)>that the the data is

5769
05:41:15.936 --> 05:41:17.055
<v Sarah Milstein (OTI)>is kept,

5770
05:41:18.096 --> 05:41:19.615
<v Sarah Milstein (OTI)>carefully and securely,

5771
05:41:19.855 --> 05:41:22.656
<v Sarah Milstein (OTI)>that there isn't risk of the data being

5772
05:41:22.910 --> 05:41:28.510
<v Speaker Julie Menin>leaked either to Microsoft or to the
public. So you can confidently say that no

5773
05:41:28.510 --> 05:41:32.590
<v Speaker Julie Menin>private or confidential Citi data has ever
been obtained by an AI company and then

5774
05:41:32.590 --> 05:41:34.670
<v Speaker Julie Menin>used to train an AI model?

5775
05:41:36.535 --> 05:41:40.535
<v Sarah Milstein (OTI)>I'm... Thank you for the
question. I'm confident that our

5776
05:41:40.615 --> 05:41:42.215
<v Sarah Milstein (OTI)>agreements with Microsoft,

5777
05:41:42.934 --> 05:41:44.375
<v Sarah Milstein (OTI)>prevent them from

5778
05:41:44.775 --> 05:41:46.535
<v Sarah Milstein (OTI)>from taking such actions.

5779
05:41:46.695 --> 05:41:51.771
<v Speaker Julie Menin>And does... Is OTI currently measure the
impact that AI tools are having on the

5780
05:41:51.771 --> 05:41:52.970
<v Speaker Julie Menin>city's workforce?

5781
05:41:53.611 --> 05:42:00.012
<v Sarah Milstein (OTI)>Yeah. Thank you for the question.
We... Per local law 12/25/2025,

5782
05:42:00.012 --> 05:42:02.890
<v Sarah Milstein (OTI)>we are engaged in a a workforce study

5783
05:42:03.130 --> 05:42:07.996
<v Sarah Milstein (OTI)>to understand the impact of algorithmic
tools on city workers and automated,

5784
05:42:08.395 --> 05:42:09.996
<v Sarah Milstein (OTI)>employment decision tools.

5785
05:42:10.316 --> 05:42:11.115
<v Sarah Milstein (OTI)>And we,

5786
05:42:11.516 --> 05:42:15.756
<v Sarah Milstein (OTI)>will be having... We will have
results from that to share next year.

5787
05:42:16.076 --> 05:42:21.250
<v Speaker Julie Menin>Okay. Alright. I'm gonna pass it over to
chair De La Rosa. Thank you. Thank you so

5788
05:42:21.250 --> 05:42:24.770
<v Speaker Julie Menin>much. And just continuing
on on that same question,

5789
05:42:25.010 --> 05:42:27.568
<v CM Carmen De La Rosa>you all testified that my legislation...

5790
05:42:27.568 --> 05:42:31.730
<v CM Carmen De La Rosa>My bill one six one basically does
the same thing as local law 25.

5791
05:42:32.530 --> 05:42:36.135
<v CM Carmen De La Rosa>One, we look forward to seeing
the report. Local law 25,

5792
05:42:36.135 --> 05:42:40.695
<v CM Carmen De La Rosa>to my knowledge, requires a onetime report
that's gonna come out of that study.

5793
05:42:40.936 --> 05:42:46.375
<v CM Carmen De La Rosa>One sixty one is more comprehensive and
is requiring that report to be updated

5794
05:42:47.111 --> 05:42:47.990
<v CM Carmen De La Rosa>yearly.

5795
05:42:48.070 --> 05:42:49.990
<v CM Carmen De La Rosa>And as this and as this

5796
05:42:50.310 --> 05:42:51.910
<v CM Carmen De La Rosa>technology is evolving,

5797
05:42:52.150 --> 05:42:54.871
<v CM Carmen De La Rosa>in my opinion, hence why
I put the bill forward,

5798
05:42:55.191 --> 05:42:56.230
<v CM Carmen De La Rosa>you know, more

5799
05:42:56.470 --> 05:42:57.510
<v CM Carmen De La Rosa>consistent

5800
05:42:57.510 --> 05:43:03.445
<v CM Carmen De La Rosa>reporting is going to be needed, especially
as we begin to see how the workforce is

5801
05:43:03.445 --> 05:43:04.166
<v CM Carmen De La Rosa>impacted.

5802
05:43:04.566 --> 05:43:07.846
<v CM Carmen De La Rosa>Before I was a technology chair, I was
the labor chair. We have a wonderful labor

5803
05:43:07.846 --> 05:43:09.445
<v CM Carmen De La Rosa>chair in the house,

5804
05:43:10.006 --> 05:43:14.898
<v CM Carmen De La Rosa>and I'm grateful to her leadership. But
this is an issue that impacts not only

5805
05:43:14.898 --> 05:43:16.340
<v CM Carmen De La Rosa>organized labor,

5806
05:43:16.340 --> 05:43:19.379
<v CM Carmen De La Rosa>but I also see it as an
issue where we're leading,

5807
05:43:20.340 --> 05:43:24.580
<v CM Carmen De La Rosa>the nation, really, in
how the workforce will,

5808
05:43:25.459 --> 05:43:27.459
<v CM Carmen De La Rosa>react to this evolving technology.

5809
05:43:27.916 --> 05:43:32.795
<v CM Carmen De La Rosa>And so is there anything that you can share
with us right now on where the workforce

5810
05:43:32.795 --> 05:43:35.436
<v CM Carmen De La Rosa>is and what those impacts will look like?

5811
05:43:35.756 --> 05:43:38.396
<v CM Carmen De La Rosa>I know you're working on the study,
but is there anything you can give us

5812
05:43:38.396 --> 05:43:38.955
<v CM Carmen De La Rosa>forthcoming?

5813
05:43:40.621 --> 05:43:42.701
<v Sarah Milstein (OTI)>Thank you for the
question, council member.

5814
05:43:43.180 --> 05:43:47.820
<v Sarah Milstein (OTI)>So the initial stages of
the report are in part about

5815
05:43:48.220 --> 05:43:53.420
<v Sarah Milstein (OTI)>determining how we even define impact.
It's a it's a pretty broad term and could

5816
05:43:53.420 --> 05:43:55.100
<v Sarah Milstein (OTI)>mean many different things. So

5817
05:43:55.666 --> 05:44:00.545
<v Sarah Milstein (OTI)>we're in this... The phase of determining
that now so that we're able to produce a

5818
05:44:00.545 --> 05:44:03.266
<v Sarah Milstein (OTI)>report that's really
meaningful for city workers.

5819
05:44:03.826 --> 05:44:05.346
<v Sarah Milstein (OTI)>I... We certainly

5820
05:44:06.146 --> 05:44:07.426
<v Sarah Milstein (OTI)>support the intent

5821
05:44:07.586 --> 05:44:09.666
<v Sarah Milstein (OTI)>of your bill, and

5822
05:44:10.190 --> 05:44:14.510
<v Sarah Milstein (OTI)>I would be interested in further
conversation about the annual nature of the

5823
05:44:14.510 --> 05:44:15.230
<v Sarah Milstein (OTI)>reporting.

5824
05:44:15.709 --> 05:44:20.190
<v CM Carmen De La Rosa>I appreciate that. Some of the information
that we're looking for, for example, is,

5825
05:44:21.070 --> 05:44:21.469
<v CM Carmen De La Rosa>I think,

5826
05:44:22.125 --> 05:44:26.525
<v CM Carmen De La Rosa>an important question. How many funded
agency positions, for example, will be

5827
05:44:26.525 --> 05:44:29.006
<v CM Carmen De La Rosa>eliminated due to the
use of of these tools?

5828
05:44:30.205 --> 05:44:33.885
<v CM Carmen De La Rosa>The agency vacancies has been something
that this council has really been looking

5829
05:44:33.885 --> 05:44:38.219
<v CM Carmen De La Rosa>at, especially as we look at civil service
reform across the board and bringing

5830
05:44:38.219 --> 05:44:42.540
<v CM Carmen De La Rosa>people into municipal service. So,
you know, we do wanna understand

5831
05:44:42.780 --> 05:44:44.379
<v CM Carmen De La Rosa>sort of displacement

5832
05:44:44.379 --> 05:44:46.620
<v CM Carmen De La Rosa>and what that looks
like for the workforce.

5833
05:44:46.700 --> 05:44:49.818
<v CM Carmen De La Rosa>We wanna also understand
reduction of hours

5834
05:44:50.045 --> 05:44:53.484
<v CM Carmen De La Rosa>over time, which this council has
not been shy about questioning the

5835
05:44:53.484 --> 05:44:54.285
<v CM Carmen De La Rosa>administrations...

5836
05:44:54.285 --> 05:44:56.926
<v CM Carmen De La Rosa>Past administrations and
present administrations about.

5837
05:44:57.564 --> 05:45:00.686
<v CM Carmen De La Rosa>We also wanna know about subcontractors.

5838
05:45:00.766 --> 05:45:02.285
<v CM Carmen De La Rosa>Right? There's a ton of

5839
05:45:03.280 --> 05:45:05.200
<v CM Carmen De La Rosa>city contracts that are

5840
05:45:06.158 --> 05:45:09.120
<v CM Carmen De La Rosa>in play right now and wanna
know what the impacts are.

5841
05:45:09.360 --> 05:45:14.639
<v CM Carmen De La Rosa>So I look forward to continuing
this conversation with you all and

5842
05:45:14.799 --> 05:45:19.326
<v CM Carmen De La Rosa>building on, you know, what we will
see from the report of local law 25.

5843
05:45:19.326 --> 05:45:22.686
<v CM Carmen De La Rosa>I also wanna say that while
my bill doesn't specifically

5844
05:45:23.326 --> 05:45:25.727
<v CM Carmen De La Rosa>speak to worker protections,

5845
05:45:26.125 --> 05:45:26.766
<v CM Carmen De La Rosa>this

5846
05:45:26.926 --> 05:45:29.965
<v CM Carmen De La Rosa>council does believe that worker
protections are important,

5847
05:45:30.205 --> 05:45:31.006
<v CM Carmen De La Rosa>especially

5848
05:45:31.006 --> 05:45:35.310
<v CM Carmen De La Rosa>as we look at that displacement and and
what could happen. Right? We we just

5849
05:45:35.310 --> 05:45:36.670
<v CM Carmen De La Rosa>questioned the companies

5850
05:45:36.670 --> 05:45:39.150
<v CM Carmen De La Rosa>for the better part of the day about

5851
05:45:39.310 --> 05:45:43.310
<v CM Carmen De La Rosa>whistleblower protections and
what happens when this technology

5852
05:45:44.110 --> 05:45:46.750
<v CM Carmen De La Rosa>gets out of control.
So we wanna understand

5853
05:45:46.750 --> 05:45:51.016
<v CM Carmen De La Rosa>how those protections are also extended to
municipal workers. Do you have anything

5854
05:45:51.016 --> 05:45:54.215
<v CM Carmen De La Rosa>in place right now that
protects our city's workforce?

5855
05:45:54.775 --> 05:45:55.414
<v Sarah Milstein (OTI)>Yes.

5856
05:45:55.814 --> 05:45:57.096
<v Sarah Milstein (OTI)>Thank you for the question.

5857
05:46:01.820 --> 05:46:05.740
<v Sarah Milstein (OTI)>Because this is a little technical, I
wanna make sure I get exactly the right,

5858
05:46:06.140 --> 05:46:07.100
<v Sarah Milstein (OTI)>information,

5859
05:46:07.340 --> 05:46:08.460
<v Sarah Milstein (OTI)>which is that the,

5860
05:46:09.260 --> 05:46:10.700
<v Sarah Milstein (OTI)>New York State civil,

5861
05:46:11.740 --> 05:46:15.180
<v Sarah Milstein (OTI)>civil service law section
eighty ten establishes

5862
05:46:15.180 --> 05:46:17.826
<v Sarah Milstein (OTI)>a moratorium on discharge, displacement,

5863
05:46:18.145 --> 05:46:22.385
<v Sarah Milstein (OTI)>or the transfer of existing duties and
functions currently performed by employees of

5864
05:46:22.385 --> 05:46:25.984
<v Sarah Milstein (OTI)>the New York City and the New York
City school district to an artificial

5865
05:46:25.984 --> 05:46:28.385
<v Sarah Milstein (OTI)>intelligence system until 2028.

5866
05:46:28.705 --> 05:46:30.830
<v Sarah Milstein (OTI)>The civil service law also

5867
05:46:30.830 --> 05:46:35.310
<v Sarah Milstein (OTI)>stipulates that the use of AI shall not
alter existing collective bargaining rights,

5868
05:46:35.629 --> 05:46:38.190
<v Sarah Milstein (OTI)>terms of employment, or
civil service status.

5869
05:46:38.750 --> 05:46:39.390
<v Sarah Milstein (OTI)>Further,

5870
05:46:39.629 --> 05:46:43.870
<v Sarah Milstein (OTI)>collective bargaining agreements may
offer employment protections beyond 2028.

5871
05:46:44.395 --> 05:46:48.154
<v Sarah Milstein (OTI)>So with that law in place, we have a little
bit of time, and we really appreciate

5872
05:46:48.154 --> 05:46:52.234
<v Sarah Milstein (OTI)>the intent of this law and the fact that
it gives us in New York City a little bit

5873
05:46:52.234 --> 05:46:53.594
<v Sarah Milstein (OTI)>of time to figure out,

5874
05:46:54.234 --> 05:46:55.275
<v Sarah Milstein (OTI)>how we

5875
05:46:55.434 --> 05:46:58.234
<v Sarah Milstein (OTI)>how we understand the
potential for displacements,

5876
05:46:58.234 --> 05:46:59.035
<v Sarah Milstein (OTI)>for deskilling,

5877
05:46:59.510 --> 05:47:01.670
<v Sarah Milstein (OTI)>for the use of subcontractors

5878
05:47:01.670 --> 05:47:05.590
<v CM Carmen De La Rosa>that might be, displacing other city
workers and the sorts of things you were

5879
05:47:05.590 --> 05:47:09.270
<v CM Carmen De La Rosa>raising. So I really appreciate that
question. Great. So we'll continue that

5880
05:47:09.270 --> 05:47:13.430
<v CM Carmen De La Rosa>conversation. I also wanna give you a
preview. I have a bill that creates a civil

5881
05:47:13.430 --> 05:47:15.030
<v CM Carmen De La Rosa>service title for AI

5882
05:47:15.346 --> 05:47:19.025
<v CM Carmen De La Rosa>within our city agencies. And one of the
things that I'm curious about is, like,

5883
05:47:19.025 --> 05:47:20.225
<v CM Carmen De La Rosa>this interagency

5884
05:47:20.225 --> 05:47:21.586
<v CM Carmen De La Rosa>communication

5885
05:47:21.586 --> 05:47:26.465
<v CM Carmen De La Rosa>around AI. Can you speak to us where
are you right now when it comes to

5886
05:47:27.025 --> 05:47:31.440
<v CM Carmen De La Rosa>each of the city agencies and evaluating
where the agencies are in terms of the use

5887
05:47:31.440 --> 05:47:32.320
<v CM Carmen De La Rosa>of AI,

5888
05:47:32.640 --> 05:47:33.520
<v CM Carmen De La Rosa>internally?

5889
05:47:34.400 --> 05:47:36.080
<v Sarah Milstein (OTI)>Yes. Thank you for the question.

5890
05:47:36.720 --> 05:47:40.160
<v Sarah Milstein (OTI)>So one of the privileges
that we have at OTI as,

5891
05:47:40.799 --> 05:47:41.600
<v Sarah Milstein (OTI)>the

5892
05:47:41.760 --> 05:47:44.879
<v Sarah Milstein (OTI)>agency that runs the
reporting for local law 35

5893
05:47:45.115 --> 05:47:49.836
<v Sarah Milstein (OTI)>is that we learn from agencies throughout
the city how they are using algorithmic

5894
05:47:49.836 --> 05:47:52.875
<v Sarah Milstein (OTI)>tools that have, direct public impact.

5895
05:47:53.436 --> 05:47:54.955
<v Sarah Milstein (OTI)>And in the process

5896
05:47:55.035 --> 05:47:55.596
<v Sarah Milstein (OTI)>of

5897
05:47:56.074 --> 05:48:00.780
<v Sarah Milstein (OTI)>collecting that data, we have a fair amount
of contact with agencies to be able to

5898
05:48:00.780 --> 05:48:01.580
<v Sarah Milstein (OTI)>learn more,

5899
05:48:01.980 --> 05:48:05.740
<v Sarah Milstein (OTI)>and to establish relationships with them
so that we can also serve as advisers to

5900
05:48:05.740 --> 05:48:06.700
<v Sarah Milstein (OTI)>them and,

5901
05:48:07.100 --> 05:48:08.700
<v Sarah Milstein (OTI)>understand their challenges

5902
05:48:09.020 --> 05:48:10.700
<v Sarah Milstein (OTI)>as this technology unfolds.

5903
05:48:11.355 --> 05:48:16.475
<v Sarah Milstein (OTI)>In the last cycle last year, we
had 56 agencies participating.

5904
05:48:17.596 --> 05:48:21.596
<v CM Carmen De La Rosa>Our first hearing as technology
chair was around surveillance.

5905
05:48:21.596 --> 05:48:25.990
<v CM Carmen De La Rosa>And one of the things that we pretty much
grew these companies about was about self

5906
05:48:25.990 --> 05:48:26.870
<v CM Carmen De La Rosa>regulation.

5907
05:48:26.950 --> 05:48:30.470
<v CM Carmen De La Rosa>And I think from the tone
and tenure of this council,

5908
05:48:30.709 --> 05:48:35.590
<v CM Carmen De La Rosa>we don't believe in self regulation.
However, we've given the grace to our city

5909
05:48:35.590 --> 05:48:37.670
<v CM Carmen De La Rosa>agencies to self report

5910
05:48:37.830 --> 05:48:38.790
<v CM Carmen De La Rosa>the use of AI.

5911
05:48:39.266 --> 05:48:44.785
<v CM Carmen De La Rosa>I wanna ask you, and I understand that
OTI is not an agency that's gonna impose

5912
05:48:44.785 --> 05:48:47.986
<v CM Carmen De La Rosa>penalties on other agencies
for not reporting.

5913
05:48:48.225 --> 05:48:53.059
<v CM Carmen De La Rosa>But there is a specific concern that my
colleagues and I have around self reporting,

5914
05:48:53.059 --> 05:48:55.139
<v CM Carmen De La Rosa>especially when it comes to the NYPD's

5915
05:48:55.139 --> 05:48:57.059
<v CM Carmen De La Rosa>reporting of surveillance technologies.

5916
05:48:57.540 --> 05:49:00.100
<v CM Carmen De La Rosa>This council has had
many hearings on that.

5917
05:49:00.420 --> 05:49:01.540
<v CM Carmen De La Rosa>I I wonder,

5918
05:49:02.260 --> 05:49:04.635
<v CM Carmen De La Rosa>you know, under the office of algorithmic

5919
05:49:04.635 --> 05:49:05.756
<v CM Carmen De La Rosa>accountability,

5920
05:49:06.154 --> 05:49:07.914
<v CM Carmen De La Rosa>what accountability tools

5921
05:49:07.914 --> 05:49:09.275
<v CM Carmen De La Rosa>will be put in place

5922
05:49:09.436 --> 05:49:10.715
<v CM Carmen De La Rosa>to make sure

5923
05:49:10.955 --> 05:49:12.955
<v CM Carmen De La Rosa>that we are properly regulating

5924
05:49:12.955 --> 05:49:17.516
<v CM Carmen De La Rosa>the use of surveillance technology in the
NYPD and other technology. You know, this

5925
05:49:17.516 --> 05:49:22.531
<v CM Carmen De La Rosa>council held a hearing just in June around
the Department of Education and the use

5926
05:49:22.531 --> 05:49:26.611
<v CM Carmen De La Rosa>of AI. So what are some of those tools
that we can look forward to seeing? Yeah.

5927
05:49:26.611 --> 05:49:28.932
<v CM Carmen De La Rosa>Thank you for that
question, council member.

5928
05:49:29.810 --> 05:49:30.371
<v Sarah Milstein (OTI)>So

5929
05:49:31.730 --> 05:49:35.010
<v Sarah Milstein (OTI)>in advance of having
full time staff at OAA,

5930
05:49:35.785 --> 05:49:39.305
<v Sarah Milstein (OTI)>we have been doing research
to understand some of the,

5931
05:49:39.945 --> 05:49:41.865
<v Sarah Milstein (OTI)>types of risk

5932
05:49:41.945 --> 05:49:43.064
<v Sarah Milstein (OTI)>assess... Risk,

5933
05:49:43.305 --> 05:49:44.984
<v Sarah Milstein (OTI)>assessment and risk management

5934
05:49:45.064 --> 05:49:49.945
<v Sarah Milstein (OTI)>that organizations like ours can do in a
situation where self reporting is what we

5935
05:49:49.945 --> 05:49:50.984
<v Sarah Milstein (OTI)>have been relying on.

5936
05:49:51.630 --> 05:49:56.590
<v Sarah Milstein (OTI)>We're a little bit early to say how we
would implement that in the city, but,

5937
05:49:57.630 --> 05:49:59.230
<v Sarah Milstein (OTI)>we are very actively

5938
05:49:59.791 --> 05:50:04.750
<v Sarah Milstein (OTI)>hiring for a director who understands that
set of problems and can help us figure

5939
05:50:04.750 --> 05:50:10.137
<v Sarah Milstein (OTI)>out how we come up with effective risk
assessment and management that includes,

5940
05:50:10.535 --> 05:50:12.775
<v Sarah Milstein (OTI)>a view into city agencies

5941
05:50:13.336 --> 05:50:15.016
<v Sarah Milstein (OTI)>perhaps beyond self reporting.

5942
05:50:15.576 --> 05:50:19.550
<v CM Carmen De La Rosa>Thank you for that answer. And, of course,
we will continue that conversation.

5943
05:50:19.550 --> 05:50:24.352
<v CM Carmen De La Rosa>Absolutely. I did wanna ask you, how,
how prepared would you say we are,

5944
05:50:26.271 --> 05:50:30.352
<v CM Carmen De La Rosa>for enterprise security system
for autonomous AI driven attacks?

5945
05:50:30.775 --> 05:50:35.336
<v CM Carmen De La Rosa>Not just attacks by human, but AI autonomous
attacks. How prepared do you all feel

5946
05:50:35.336 --> 05:50:36.775
<v CM Carmen De La Rosa>we are as a city?

5947
05:50:37.016 --> 05:50:40.215
<v Sarah Milstein (OTI)>Yeah. Thank you for the question.
I'll let CJ Dixon answer that.

5948
05:50:41.096 --> 05:50:44.775
<v CJ Dixon (NYC Cyber Command / OTI)>And that is a great question. And and I will
acknowledge that this is a very dynamic

5949
05:50:44.775 --> 05:50:48.430
<v CJ Dixon (NYC Cyber Command / OTI)>environment. It is always changing that's
just the nature of technology. New York

5950
05:50:48.430 --> 05:50:52.590
<v CJ Dixon (NYC Cyber Command / OTI)>City Cyber Command, however, provides
24 by seven detection and response

5951
05:50:52.590 --> 05:50:56.190
<v CJ Dixon (NYC Cyber Command / OTI)>capabilities through our 24 by
seven security operation center.

5952
05:50:56.590 --> 05:50:59.950
<v CJ Dixon (NYC Cyber Command / OTI)>We have provided that service to the city
since the establishment of New York City

5953
05:50:59.950 --> 05:51:04.514
<v CJ Dixon (NYC Cyber Command / OTI)>Cyber Command. We provide that service
whether or not the attacker is human or an

5954
05:51:04.514 --> 05:51:06.594
<v CJ Dixon (NYC Cyber Command / OTI)>artificial intelligence based

5955
05:51:06.674 --> 05:51:07.475
<v CJ Dixon (NYC Cyber Command / OTI)>attacker.

5956
05:51:07.875 --> 05:51:11.395
<v CJ Dixon (NYC Cyber Command / OTI)>That attacker would still attack the exact
same vulnerabilities in the system, and

5957
05:51:11.395 --> 05:51:14.834
<v CJ Dixon (NYC Cyber Command / OTI)>therefore, we would still respond functionally
in the same manner. We would just

5958
05:51:14.834 --> 05:51:15.715
<v CJ Dixon (NYC Cyber Command / OTI)>have to do it faster.

5959
05:51:16.500 --> 05:51:20.660
<v CJ Dixon (NYC Cyber Command / OTI)>We continue to evolve with the technology,
and we will continue to do so with the

5960
05:51:20.660 --> 05:51:23.860
<v CJ Dixon (NYC Cyber Command / OTI)>best of our ability as the
technology continues to evolve.

5961
05:51:24.980 --> 05:51:26.100
<v CM Carmen De La Rosa>The speaker,

5962
05:51:26.740 --> 05:51:30.820
<v CM Carmen De La Rosa>rightfully so, in my opinion, asked the
companies about the possibility of a kill

5963
05:51:30.820 --> 05:51:35.414
<v CM Carmen De La Rosa>switch and where that kill switch should
live. Do you have any opinions if the kill

5964
05:51:35.414 --> 05:51:37.895
<v CM Carmen De La Rosa>switch should should
live in Cyber Command?

5965
05:51:38.455 --> 05:51:39.574
<v CJ Dixon (NYC Cyber Command / OTI)>I I agree

5966
05:51:39.975 --> 05:51:44.615
<v CJ Dixon (NYC Cyber Command / OTI)>in principle with the idea of a kill
switch. I will say there are some technical,

5967
05:51:45.574 --> 05:51:46.215
<v CJ Dixon (NYC Cyber Command / OTI)>barriers

5968
05:51:46.500 --> 05:51:48.500
<v CJ Dixon (NYC Cyber Command / OTI)>that would prevent a

5969
05:51:49.220 --> 05:51:50.980
<v CJ Dixon (NYC Cyber Command / OTI)>kill switch as conceptualized

5970
05:51:50.980 --> 05:51:52.500
<v CJ Dixon (NYC Cyber Command / OTI)>living in any

5971
05:51:52.660 --> 05:51:57.220
<v CJ Dixon (NYC Cyber Command / OTI)>agency within New York City, whether that
is New York City Cyber Command, OTI, or

5972
05:51:57.220 --> 05:51:58.180
<v CJ Dixon (NYC Cyber Command / OTI)>anywhere else.

5973
05:51:58.420 --> 05:52:02.756
<v CJ Dixon (NYC Cyber Command / OTI)>The reason being is the technical reality
for these models is they are deployed on

5974
05:52:02.756 --> 05:52:04.035
<v CJ Dixon (NYC Cyber Command / OTI)>distributed systems,

5975
05:52:04.916 --> 05:52:06.516
<v CJ Dixon (NYC Cyber Command / OTI)>on distributed architecture.

5976
05:52:06.516 --> 05:52:07.717
<v CJ Dixon (NYC Cyber Command / OTI)>In order to truly...

5977
05:52:10.676 --> 05:52:14.676
<v CJ Dixon (NYC Cyber Command / OTI)>With a singular button or with a kill switch
or anything of that concept for that to

5978
05:52:14.676 --> 05:52:15.980
<v CJ Dixon (NYC Cyber Command / OTI)>actually work work

5979
05:52:15.980 --> 05:52:16.940
<v CJ Dixon (NYC Cyber Command / OTI)>functionally.

5980
05:52:16.940 --> 05:52:20.621
<v CJ Dixon (NYC Cyber Command / OTI)>We would need to be able to coordinate
across multiple stakeholders that are not

5981
05:52:20.621 --> 05:52:22.621
<v CJ Dixon (NYC Cyber Command / OTI)>actually regionally colocated

5982
05:52:22.621 --> 05:52:26.621
<v CJ Dixon (NYC Cyber Command / OTI)>to be able to turn off everything at
the exact same time. New York City Cyber

5983
05:52:26.621 --> 05:52:30.781
<v CJ Dixon (NYC Cyber Command / OTI)>Command or OTI does not have the authorities
or the technical capability to do that.

5984
05:52:31.695 --> 05:52:32.975
<v CJ Dixon (NYC Cyber Command / OTI)>And this runs up...

5985
05:52:33.295 --> 05:52:37.535
<v CJ Dixon (NYC Cyber Command / OTI)>Us up against a similar challenge that
we've had with other kill switch bills as it

5986
05:52:37.535 --> 05:52:41.615
<v CJ Dixon (NYC Cyber Command / OTI)>relates to computer technology in general.
Mhmm. As this technology gets more

5987
05:52:41.615 --> 05:52:44.494
<v CJ Dixon (NYC Cyber Command / OTI)>distributed, it gets harder
and harder for any singular

5988
05:52:44.580 --> 05:52:47.701
<v CJ Dixon (NYC Cyber Command / OTI)>solitary entity to turn
it off specifically.

5989
05:52:47.860 --> 05:52:52.980
<v CJ Dixon (NYC Cyber Command / OTI)>That is not to say that there is not
technical means by which a group of entities

5990
05:52:52.980 --> 05:52:55.220
<v CJ Dixon (NYC Cyber Command / OTI)>with the legal authority to do so

5991
05:52:55.460 --> 05:52:56.980
<v CJ Dixon (NYC Cyber Command / OTI)>could not disconnect

5992
05:52:56.980 --> 05:53:01.355
<v CJ Dixon (NYC Cyber Command / OTI)>a given model from a given function.
But in this particular case, we agree on

5993
05:53:01.355 --> 05:53:05.436
<v CJ Dixon (NYC Cyber Command / OTI)>principle, but it would not be able to
be implemented at New York State Cyber

5994
05:53:05.436 --> 05:53:08.715
<v CJ Dixon (NYC Cyber Command / OTI)>Command in practice, not with the
technical barriers that are in place.

5995
05:53:09.115 --> 05:53:11.195
<v CM Carmen De La Rosa>Okay. I'm sure there'll be more on that.

5996
05:53:13.355 --> 05:53:13.994
<v CM Carmen De La Rosa>Let's see.

5997
05:53:17.568 --> 05:53:21.090
<v CM Carmen De La Rosa>In your opinion, should AI
control be recognized...

5998
05:53:21.250 --> 05:53:22.930
<v CM Carmen De La Rosa>A recognized cybersecurity

5999
05:53:22.930 --> 05:53:27.568
<v CM Carmen De La Rosa>specialty? If so, what should be the
training and curriculum included in that?

6000
05:53:29.465 --> 05:53:31.545
<v CJ Dixon (NYC Cyber Command / OTI)>I do not believe every

6001
05:53:32.824 --> 05:53:36.424
<v CJ Dixon (NYC Cyber Command / OTI)>function related to artificial
intelligence is necessarily going to be a

6002
05:53:36.424 --> 05:53:37.225
<v CJ Dixon (NYC Cyber Command / OTI)>cybersecurity

6003
05:53:37.225 --> 05:53:41.305
<v CJ Dixon (NYC Cyber Command / OTI)>specific challenge. There are some
intersections between artificial intelligence

6004
05:53:41.305 --> 05:53:42.664
<v CJ Dixon (NYC Cyber Command / OTI)>and cybersecurity

6005
05:53:42.664 --> 05:53:44.424
<v CJ Dixon (NYC Cyber Command / OTI)>as a particular domain.

6006
05:53:45.050 --> 05:53:46.250
<v CJ Dixon (NYC Cyber Command / OTI)>And when those,

6007
05:53:47.130 --> 05:53:48.890
<v CJ Dixon (NYC Cyber Command / OTI)>two domains interact,

6008
05:53:49.050 --> 05:53:52.490
<v CJ Dixon (NYC Cyber Command / OTI)>that is where certain things should fall
within cyber command and a cybersecurity

6009
05:53:52.490 --> 05:53:53.371
<v CJ Dixon (NYC Cyber Command / OTI)>professional's

6010
05:53:53.371 --> 05:53:54.250
<v CJ Dixon (NYC Cyber Command / OTI)>purview.

6011
05:53:54.330 --> 05:53:58.890
<v CJ Dixon (NYC Cyber Command / OTI)>And that is any instance of a cyber attack
or cybersecurity regardless of if the

6012
05:53:58.890 --> 05:54:02.125
<v CJ Dixon (NYC Cyber Command / OTI)>perpetrator is a human or
an artificial intelligence.

6013
05:54:02.766 --> 05:54:07.645
<v Speaker Julie Menin>I just wanna ask one follow-up. How many
people are working in cyber command right

6014
05:54:07.645 --> 05:54:09.246
<v Speaker Julie Menin>now? A 121.

6015
05:54:09.246 --> 05:54:11.645
<v Speaker Julie Menin>And how many vacant positions do you have?

6016
05:54:12.205 --> 05:54:15.166
<v CJ Dixon (NYC Cyber Command / OTI)>I will need to go back and find
that exact number for you, speaker.

6017
05:54:15.880 --> 05:54:19.320
<v Speaker Julie Menin>Okay. Thank you. Please provide... If
you could please provide that. Thank you.

6018
05:54:20.280 --> 05:54:21.160
<v CM Carmen De La Rosa>Thank you.

6019
05:54:21.880 --> 05:54:25.240
<v CM Carmen De La Rosa>I have a question for our DCWP,

6020
05:54:25.720 --> 05:54:27.240
<v CM Carmen De La Rosa>commissioner and and team.

6021
05:54:28.596 --> 05:54:33.715
<v CM Carmen De La Rosa>What types of AI misuse are
you seeing that harm consumers

6022
05:54:33.715 --> 05:54:35.555
<v CM Carmen De La Rosa>that you're most concerned about?

6023
05:54:36.994 --> 05:54:40.676
<v Samuel Levine (DCWP)>Well, thank you, council member. Thank
you, chair. And it's great to have the

6024
05:54:40.676 --> 05:54:44.541
<v Samuel Levine (DCWP)>opportunity to be here. We're concerned
about a wide range of misuse, whether it's

6025
05:54:44.541 --> 05:54:49.020
<v Samuel Levine (DCWP)>people using voice cloning technology that
these AI companies make fully available

6026
05:54:49.020 --> 05:54:50.460
<v Samuel Levine (DCWP)>to scanner seniors,

6027
05:54:50.701 --> 05:54:55.820
<v Samuel Levine (DCWP)>whether it's grocery stores using AI
technology to fix prices or to personalize

6028
05:54:55.820 --> 05:54:59.994
<v Samuel Levine (DCWP)>prices, something that the speaker has
introduced to build or prohibit. We're seeing

6029
05:54:59.994 --> 05:55:00.795
<v Samuel Levine (DCWP)>widespread

6030
05:55:00.795 --> 05:55:04.395
<v Samuel Levine (DCWP)>misuse of these tools. And the other
thing we're seeing across the board is an

6031
05:55:04.395 --> 05:55:08.074
<v Samuel Levine (DCWP)>attitude by these AI companies
that they bear no responsibility

6032
05:55:08.234 --> 05:55:09.360
<v Samuel Levine (DCWP)>for the downstream

6033
05:55:09.360 --> 05:55:13.600
<v Samuel Levine (DCWP)>consequences. I have to just say, chair, I
was struck by the final answer I heard by

6034
05:55:13.600 --> 05:55:15.200
<v Samuel Levine (DCWP)>the witness for OpenAI

6035
05:55:15.200 --> 05:55:19.760
<v Samuel Levine (DCWP)>who, when asked a basic question about
AI safety, responded that she works in

6036
05:55:19.760 --> 05:55:20.480
<v Samuel Levine (DCWP)>policy,

6037
05:55:20.799 --> 05:55:21.360
<v Samuel Levine (DCWP)>not safety.

6038
05:55:21.846 --> 05:55:24.566
<v Samuel Levine (DCWP)>How are you running a company like OpenAI

6039
05:55:24.566 --> 05:55:26.406
<v Samuel Levine (DCWP)>set to go public introducing

6040
05:55:26.406 --> 05:55:28.086
<v Samuel Levine (DCWP)>trillion dollar tools

6041
05:55:28.246 --> 05:55:32.406
<v Samuel Levine (DCWP)>and you don't have safety as part of your
policy division? Deeply disheartened by

6042
05:55:32.406 --> 05:55:35.605
<v Samuel Levine (DCWP)>what we heard at the hearing earlier today.
Mhmm. Absolutely. You should've heard

6043
05:55:35.605 --> 05:55:38.590
<v Samuel Levine (DCWP)>their first answer. It went
something like, I don't know.

6044
05:55:39.070 --> 05:55:42.830
<v Samuel Levine (DCWP)>There were a lot of I there were a lot of
I don't knows. And on behalf of the admin,

6045
05:55:42.830 --> 05:55:46.110
<v Samuel Levine (DCWP)>we hope to give you more fulsome answers.
Thank you. And we look forward to engaging

6046
05:55:46.110 --> 05:55:50.645
<v Samuel Levine (DCWP)>with you in that more broad conversation
as well. In your opinion, is DCWP

6047
05:55:50.645 --> 05:55:53.926
<v CM Carmen De La Rosa>sufficiently staffed to
combat this AI misuse?

6048
05:55:54.244 --> 05:55:58.725
<v CM Carmen De La Rosa>And, would adding, in your opinion, a
private way of action... A private right of

6049
05:55:58.725 --> 05:56:02.941
<v CM Carmen De La Rosa>action strengthen the ability to enforce
against violators? Thank you, sir. I'll be

6050
05:56:02.941 --> 05:56:04.941
<v CM Carmen De La Rosa>candid. I feel that even as we've...

6051
05:56:05.660 --> 05:56:09.740
<v Samuel Levine (DCWP)>As under the Mamdani administration, we
are adding to our ranks, I believe we are

6052
05:56:09.740 --> 05:56:14.541
<v Samuel Levine (DCWP)>capable of taking on many of the misuses
of AI, whether that's DoorDash using

6053
05:56:14.541 --> 05:56:18.574
<v Samuel Levine (DCWP)>algorithms to underpay its workers,
something we remedied in the largest labor

6054
05:56:18.574 --> 05:56:22.334
<v Samuel Levine (DCWP)>enforcement action in history,
whether it's companies using AI to,

6055
05:56:22.654 --> 05:56:26.814
<v Samuel Levine (DCWP)>harm consumers through surveillance pricing,
we're strongly supporting the speaker's

6056
05:56:26.814 --> 05:56:32.281
<v Samuel Levine (DCWP)>bill on this. I think we have strong tools.
That said, no enforcement agency, and

6057
05:56:32.281 --> 05:56:36.602
<v Samuel Levine (DCWP)>I've I've worked at the state, federal,
and municipal level, will ever have the

6058
05:56:36.602 --> 05:56:42.121
<v Samuel Levine (DCWP)>resources to take on every possible misuse
of this technology. That is why, chair, I

6059
05:56:42.121 --> 05:56:45.715
<v Samuel Levine (DCWP)>would strongly support a private right of
action as as contemplated by a number of

6060
05:56:45.715 --> 05:56:49.236
<v Samuel Levine (DCWP)>these bills to give ordinary
people, workers, and businesses

6061
05:56:49.316 --> 05:56:52.836
<v Samuel Levine (DCWP)>the opportunity to hold these companies
accountable. I can guarantee you, though,

6062
05:56:52.836 --> 05:56:56.930
<v Samuel Levine (DCWP)>these companies, as soon as the city does
that, are gonna run to Washington asking

6063
05:56:56.930 --> 05:56:57.969
<v Samuel Levine (DCWP)>for preemption.

6064
05:56:58.129 --> 05:57:01.889
<v Samuel Levine (DCWP)>Regardless of what they might be saying at
this hearing, what really speaks is where

6065
05:57:01.889 --> 05:57:07.809
<v Samuel Levine (DCWP)>their lobbying dollars are going, and that
is to DC congress asking to hit delete on

6066
05:57:07.809 --> 05:57:10.690
<v Samuel Levine (DCWP)>all of these city and
state, AI safety laws.

6067
05:57:11.904 --> 05:57:16.785
<v CM Carmen De La Rosa>Thank you, commissioner and and team.
I'm gonna pass it on to my colleagues,

6068
05:57:17.025 --> 05:57:18.785
<v CM Carmen De La Rosa>council member Gutierrez,

6069
05:57:18.785 --> 05:57:21.426
<v CM Carmen De La Rosa>followed by Aldeball, followed by Maloney.

6070
05:57:22.145 --> 05:57:23.025
<v CM Jennifer Gutiérrez>Thank you.

6071
05:57:23.346 --> 05:57:24.625
<v CM Jennifer Gutiérrez>Good to see everyone.

6072
05:57:25.586 --> 05:57:27.549
<v CM Jennifer Gutiérrez>My first question is

6073
05:57:27.709 --> 05:57:28.350
<v CM Jennifer Gutiérrez>for,

6074
05:57:28.590 --> 05:57:30.430
<v CM Jennifer Gutiérrez>deputy commissioner. Good to see you.

6075
05:57:32.110 --> 05:57:36.830
<v CM Jennifer Gutiérrez>Going off of, the speaker's line of
questioning regarding the the office of

6076
05:57:36.830 --> 05:57:37.709
<v CM Jennifer Gutiérrez>algorithmic

6077
05:57:37.709 --> 05:57:38.270
<v CM Jennifer Gutiérrez>accountability,

6078
05:57:38.975 --> 05:57:42.654
<v CM Jennifer Gutiérrez>Can you share... I know I know you said
you're excited about this report. You're

6079
05:57:42.654 --> 05:57:44.414
<v CM Jennifer Gutiérrez>setting up an orientation meeting.

6080
05:57:44.494 --> 05:57:46.975
<v CM Jennifer Gutiérrez>It it feels like not a lot lot of time.

6081
05:57:47.295 --> 05:57:50.814
<v CM Jennifer Gutiérrez>I guess, can you share... Before the report
is scheduled to be released, what can

6082
05:57:50.814 --> 05:57:54.950
<v CM Jennifer Gutiérrez>you share about the cadence of how
often you're gonna be able to meet?

6083
05:57:55.350 --> 05:57:58.230
<v CM Jennifer Gutiérrez>Do you expect the report
to be released on time?

6084
05:57:59.350 --> 05:58:03.270
<v CM Jennifer Gutiérrez>This is not a reflection of you, but I
really felt like the previous administration

6085
05:58:03.270 --> 05:58:05.190
<v CM Jennifer Gutiérrez>under the previous OTI

6086
05:58:05.190 --> 05:58:05.990
<v CM Jennifer Gutiérrez>leadership

6087
05:58:06.266 --> 05:58:11.785
<v CM Jennifer Gutiérrez>was delayed in in in many things. And we
certainly don't want compliance to be one

6088
05:58:11.785 --> 05:58:14.506
<v CM Jennifer Gutiérrez>of those things. So if you can just share
with me a little bit about the cadence,

6089
05:58:14.586 --> 05:58:15.625
<v CM Jennifer Gutiérrez>how many people,

6090
05:58:16.105 --> 05:58:19.865
<v CM Jennifer Gutiérrez>you said the funding you got in June, how
many people will the office be able to

6091
05:58:19.865 --> 05:58:20.266
<v CM Jennifer Gutiérrez>employ,

6092
05:58:20.810 --> 05:58:23.611
<v CM Jennifer Gutiérrez>Where you are in that process currently?

6093
05:58:23.691 --> 05:58:24.090
<v CM Jennifer Gutiérrez>Oh,

6094
05:58:24.650 --> 05:58:27.531
<v CM Jennifer Gutiérrez>okay. That was it for you. I had a question
for DCW. I can come back. Go ahead. Go

6095
05:58:27.531 --> 05:58:29.291
<v CM Jennifer Gutiérrez>ahead. Okay. And then for DCWP,

6096
05:58:29.291 --> 05:58:30.650
<v CM Jennifer Gutiérrez>I just wanted to ask,

6097
05:58:30.890 --> 05:58:34.010
<v CM Jennifer Gutiérrez>if you can share a
little bit about how the

6098
05:58:34.795 --> 05:58:36.074
<v CM Jennifer Gutiérrez>the implementation

6099
05:58:36.074 --> 05:58:41.035
<v CM Jennifer Gutiérrez>of local law one forty four is going, the
automated deployment decision tools. Just

6100
05:58:41.035 --> 05:58:41.916
<v CM Jennifer Gutiérrez>wanted to see

6101
05:58:42.795 --> 05:58:45.994
<v CM Jennifer Gutiérrez>how the implementation of that law is going,
and that's it. Fair enough. Thank you.

6102
05:58:47.441 --> 05:58:51.201
<v Samuel Levine (DCWP)>Well, thank you councilmember. Let me start
by 01:44. I think this is a really well

6103
05:58:51.201 --> 05:58:55.041
<v Samuel Levine (DCWP)>intentioned bill to regulate the use of
automated decision making systems for

6104
05:58:55.041 --> 05:58:58.720
<v Samuel Levine (DCWP)>hiring. I think it does have some flaw...
Those... Flaws, though, especially

6105
05:58:59.766 --> 05:59:05.205
<v Samuel Levine (DCWP)>in that it requires companies that use AI
systems for hiring to conduct audits. It

6106
05:59:05.205 --> 05:59:08.965
<v Samuel Levine (DCWP)>doesn't actually require that those audits
come up clean. That's really a civil

6107
05:59:08.965 --> 05:59:12.670
<v Samuel Levine (DCWP)>rights question. That is why our administration
put out a policy statement last week

6108
05:59:12.670 --> 05:59:14.590
<v Samuel Levine (DCWP)>on behalf of DCWP

6109
05:59:14.590 --> 05:59:18.590
<v Samuel Levine (DCWP)>Commission on Human Rights and Tatsium
Limousine, making clear that existing laws

6110
05:59:18.590 --> 05:59:22.590
<v Samuel Levine (DCWP)>apply. But I do wanna circle back to a
broader point that this hearing reveals,

6111
05:59:22.590 --> 05:59:23.229
<v Samuel Levine (DCWP)>which is that

6112
05:59:24.615 --> 05:59:28.775
<v Samuel Levine (DCWP)>bias in decision making produced by AI
systems is a very real concern you're

6113
05:59:28.775 --> 05:59:32.775
<v Samuel Levine (DCWP)>addressing. These companies would like an
exemption from the law. The companies that

6114
05:59:32.775 --> 05:59:37.574
<v Samuel Levine (DCWP)>testified earlier want to assert that if
decisions are made in a black box, nobody

6115
05:59:37.574 --> 05:59:41.951
<v Samuel Levine (DCWP)>can held... Be held accountable for I
applaud this council's effort to actually

6116
05:59:42.191 --> 05:59:43.630
<v Samuel Levine (DCWP)>impute responsibility

6117
05:59:43.710 --> 05:59:47.951
<v Samuel Levine (DCWP)>for these harms to the companies making
billions of dollars on it, and we would be

6118
05:59:47.951 --> 05:59:51.230
<v Samuel Levine (DCWP)>very happy to work with the
council to work on a revised,

6119
05:59:51.230 --> 05:59:52.031
<v Samuel Levine (DCWP)>expanded,

6120
05:59:52.031 --> 05:59:55.727
<v Samuel Levine (DCWP)>strengthened automated decision making
bill that places the responsibility

6121
05:59:55.727 --> 06:00:00.207
<v Samuel Levine (DCWP)>to remove bias from hiring where it
belongs. And do you think that the...

6122
06:00:01.807 --> 06:00:07.246
<v Samuel Levine (DCWP)>Is there... Are there instances where the
the harm is happening amongst the city

6123
06:00:07.246 --> 06:00:12.299
<v Samuel Levine (DCWP)>agencies using these tools? I don't believe
that... I I would defer to OTI and to

6124
06:00:12.299 --> 06:00:16.940
<v Samuel Levine (DCWP)>those in hiring. I don't believe
that the city is using AI tools

6125
06:00:17.100 --> 06:00:21.180
<v Samuel Levine (DCWP)>to sort resumes, but I would defer to others
in the admin... I can tell you it in my

6126
06:00:21.180 --> 06:00:23.625
<v Samuel Levine (DCWP)>agency, we have people reviewing

6127
06:00:23.625 --> 06:00:25.465
<v Samuel Levine (DCWP)>people reviewing applications.

6128
06:00:25.705 --> 06:00:29.705
<v Samuel Levine (DCWP)>Just as starting in January when delivery
workers get deactivated, we have people

6129
06:00:29.705 --> 06:00:33.705
<v Samuel Levine (DCWP)>reviewing that. When consumers file
complaints, we have people reviewing it. We are

6130
06:00:33.705 --> 06:00:37.800
<v Samuel Levine (DCWP)>a big believer in our administration that
the people of New York should be served by

6131
06:00:37.800 --> 06:00:40.920
<v Samuel Levine (DCWP)>the people of the city government. Thank
you. Part of the package of bills that the

6132
06:00:40.920 --> 06:00:43.640
<v Samuel Levine (DCWP)>speaker mentioned that
we passed last December

6133
06:00:44.041 --> 06:00:49.880
<v CM Jennifer Gutiérrez>on on Christmas was, to get the city and
to get OTI to report on all those AI tools

6134
06:00:49.880 --> 06:00:53.924
<v CM Jennifer Gutiérrez>that every city agency is reporting on.
So we'll know soon enough. But please,

6135
06:00:53.924 --> 06:00:55.605
<v CM Jennifer Gutiérrez>deputy commissioner, thank you.

6136
06:00:56.244 --> 06:00:58.805
<v Sarah Milstein (OTI)>Yeah. Thank you for the
question, council member.

6137
06:00:59.045 --> 06:01:02.805
<v Sarah Milstein (OTI)>So I think there's a couple of reports we're
talking about here. Wanna give a little

6138
06:01:02.805 --> 06:01:03.443
<v Sarah Milstein (OTI)>bit of context.

6139
06:01:03.940 --> 06:01:08.340
<v Sarah Milstein (OTI)>Local law 35 is the report
that, requires agencies

6140
06:01:08.340 --> 06:01:10.500
<v Sarah Milstein (OTI)>to tell us about the algorithmic

6141
06:01:10.578 --> 06:01:12.260
<v Sarah Milstein (OTI)>tools they're using that have public,

6142
06:01:12.500 --> 06:01:13.219
<v Sarah Milstein (OTI)>impact.

6143
06:01:13.540 --> 06:01:15.059
<v Sarah Milstein (OTI)>And that report,

6144
06:01:15.139 --> 06:01:18.180
<v Sarah Milstein (OTI)>we have, delivered on time every
year for the past five years.

6145
06:01:18.775 --> 06:01:19.975
<v Sarah Milstein (OTI)>The the,

6146
06:01:20.775 --> 06:01:24.535
<v Sarah Milstein (OTI)>process court kicked off last week
Mhmm. And we'll be collecting

6147
06:01:24.615 --> 06:01:29.135
<v Sarah Milstein (OTI)>data over the course of the next few months
with the plan to publish it in March

6148
06:01:29.135 --> 06:01:29.975
<v Sarah Milstein (OTI)>2027.

6149
06:01:30.535 --> 06:01:35.270
<v Sarah Milstein (OTI)>The... We have all the indications that
that will come in on time as it has in each

6150
06:01:35.270 --> 06:01:36.549
<v Sarah Milstein (OTI)>of the past five years.

6151
06:01:37.110 --> 06:01:40.870
<v Sarah Milstein (OTI)>So we're feeling good about that. I'm just
referring to the report back from, intro

6152
06:01:40.870 --> 06:01:41.830
<v Sarah Milstein (OTI)>199.

6153
06:01:41.908 --> 06:01:44.469
<v CM Jennifer Gutiérrez>I'm sorry. Could you I was
just referring to the report,

6154
06:01:44.950 --> 06:01:46.870
<v CM Jennifer Gutiérrez>required by intro 199.

6155
06:01:46.870 --> 06:01:49.844
<v CM Jennifer Gutiérrez>So that's the package of bills from last
fall. Yeah. Thank you for clarifying.

6156
06:01:50.004 --> 06:01:51.523
<v CM Jennifer Gutiérrez>So we

6157
06:01:53.283 --> 06:01:54.004
<v Sarah Milstein (OTI)>are...

6158
06:01:54.404 --> 06:01:59.125
<v Sarah Milstein (OTI)>The director of OAA, which we hope
to be able to announce quite soon,

6159
06:01:59.523 --> 06:02:02.164
<v Sarah Milstein (OTI)>will be, primarily responsible for

6160
06:02:02.750 --> 06:02:06.830
<v Sarah Milstein (OTI)>the additional reporting that
will be required in March 2027.

6161
06:02:06.990 --> 06:02:09.790
<v Sarah Milstein (OTI)>And in part because we
have this good experience

6162
06:02:10.030 --> 06:02:14.350
<v Sarah Milstein (OTI)>internally understanding process with
local law 35, we think we're in a good

6163
06:02:14.350 --> 06:02:19.686
<v Sarah Milstein (OTI)>position to move quickly on the additional
reporting that will be due in March 2027.

6164
06:02:20.164 --> 06:02:23.045
<v CM Jennifer Gutiérrez>Okay. And you said announcement.
Do you have someone

6165
06:02:23.205 --> 06:02:24.805
<v CM Jennifer Gutiérrez>already for the role?

6166
06:02:26.244 --> 06:02:30.004
<v CM Carmen De La Rosa>We will have an announcement, and I hope
that we'll be about the announcement. Okay.

6167
06:02:30.004 --> 06:02:34.380
<v CM Carmen De La Rosa>Thank you. Thank you, chair. Thank you. We
have, council member Audubaugh, followed

6168
06:02:34.380 --> 06:02:36.220
<v CM Carmen De La Rosa>by Maloney, followed by Juan.

6169
06:02:38.220 --> 06:02:40.701
<v CM Shirley Aldebol>Thank you, chair and and speaker Manon.

6170
06:02:41.900 --> 06:02:45.020
<v CM Shirley Aldebol>So you said the hiring process for the six

6171
06:02:46.057 --> 06:02:49.977
<v CM Shirley Aldebol>new positions that were
created in the adopted budget,

6172
06:02:51.176 --> 06:02:53.176
<v CM Shirley Aldebol>are actively underway.

6173
06:02:53.176 --> 06:02:54.457
<v CM Shirley Aldebol>Like, specifically,

6174
06:02:54.695 --> 06:02:56.297
<v CM Shirley Aldebol>where are you in the process?

6175
06:02:56.695 --> 06:02:58.695
<v CM Shirley Aldebol>How many people have already been hired?

6176
06:03:00.410 --> 06:03:04.012
<v CM Shirley Aldebol>How many people are left to be hired? And
what are... What have been the challenges

6177
06:03:04.012 --> 06:03:07.291
<v CM Shirley Aldebol>in in being able to hire people quickly?

6178
06:03:08.330 --> 06:03:10.730
<v Sarah Milstein (OTI)>Yeah. Thank you for the
question, council member.

6179
06:03:11.625 --> 06:03:12.266
<v Sarah Milstein (OTI)>So

6180
06:03:12.504 --> 06:03:13.545
<v Sarah Milstein (OTI)>we

6181
06:03:13.545 --> 06:03:16.824
<v Sarah Milstein (OTI)>hope to be announcing a director
for the office quite soon,

6182
06:03:17.865 --> 06:03:19.225
<v Sarah Milstein (OTI)>in the next week or two.

6183
06:03:19.945 --> 06:03:20.584
<v Sarah Milstein (OTI)>And

6184
06:03:21.064 --> 06:03:22.984
<v Sarah Milstein (OTI)>we are in the process of

6185
06:03:23.630 --> 06:03:28.031
<v Sarah Milstein (OTI)>defining the additional roles so that
they can be brought on quite quickly.

6186
06:03:28.510 --> 06:03:33.390
<v Sarah Milstein (OTI)>We certainly are challenged to find
people who have the right combination

6187
06:03:33.630 --> 06:03:38.430
<v Sarah Milstein (OTI)>of AI expertise and understanding of city
systems to be able to, like, hit the

6188
06:03:38.430 --> 06:03:41.445
<v Sarah Milstein (OTI)>ground running. That has been
a little bit of a challenge,

6189
06:03:42.006 --> 06:03:46.725
<v Sarah Milstein (OTI)>but we do have... I think bringing in a
strong leader is gonna help us draw good

6190
06:03:46.725 --> 06:03:49.445
<v Sarah Milstein (OTI)>people who are gonna wanna
do this work with us.

6191
06:03:50.166 --> 06:03:51.686
<v CM Shirley Aldebol>If I may, chair.

6192
06:03:55.570 --> 06:03:57.890
<v CM Shirley Aldebol>Part of the role is to to analyze

6193
06:03:58.932 --> 06:04:02.932
<v CM Shirley Aldebol>algorithmic tools submitted by the agencies
to determine whether there's risk that

6194
06:04:02.932 --> 06:04:06.050
<v CM Shirley Aldebol>the proposal could
result in discriminatory

6195
06:04:06.210 --> 06:04:07.170
<v CM Shirley Aldebol>decision making.

6196
06:04:10.695 --> 06:04:14.936
<v CM Shirley Aldebol>What is your office what
is your office doing or

6197
06:04:15.256 --> 06:04:20.137
<v CM Shirley Aldebol>or thinking about in terms of
protecting the city's workforce

6198
06:04:20.855 --> 06:04:23.656
<v CM Shirley Aldebol>as AI develops and evolves

6199
06:04:23.656 --> 06:04:24.215
<v CM Shirley Aldebol>and

6200
06:04:25.160 --> 06:04:27.080
<v CM Shirley Aldebol>you're kind of like the gatekeeper

6201
06:04:27.800 --> 06:04:29.000
<v Unidentified [?]>Yeah. AI.

6202
06:04:29.160 --> 06:04:30.840
<v Sarah Milstein (OTI)>Yeah. Thank you for the question.

6203
06:04:31.720 --> 06:04:33.800
<v Sarah Milstein (OTI)>So in parallel

6204
06:04:33.800 --> 06:04:38.360
<v Sarah Milstein (OTI)>with some of the risk assessment that we
can do of the software, we are doing the

6205
06:04:38.725 --> 06:04:39.605
<v Sarah Milstein (OTI)>research

6206
06:04:40.324 --> 06:04:43.285
<v Sarah Milstein (OTI)>associated with local law 25, which has us

6207
06:04:43.445 --> 06:04:47.445
<v Sarah Milstein (OTI)>researching how algorithmic tools are
impacting... Algorithmic tools and,

6208
06:04:48.484 --> 06:04:52.324
<v Sarah Milstein (OTI)>automated employment decision tools
might be impacting city workers.

6209
06:04:53.220 --> 06:04:53.861
<v Sarah Milstein (OTI)>And

6210
06:04:54.260 --> 06:04:55.380
<v Sarah Milstein (OTI)>believe that

6211
06:04:55.780 --> 06:04:58.820
<v Sarah Milstein (OTI)>we should have results from that in 2027,

6212
06:04:58.820 --> 06:05:02.260
<v Sarah Milstein (OTI)>and we'll have a much better
sense of the landscape of how

6213
06:05:02.740 --> 06:05:04.820
<v Sarah Milstein (OTI)>how workers in the city are impacted.

6214
06:05:05.885 --> 06:05:09.244
<v CM Carmen De La Rosa>K. Thank you. Thank you.
Council member Maloney,

6215
06:05:09.244 --> 06:05:10.285
<v CM Carmen De La Rosa>followed by Juan,

6216
06:05:10.525 --> 06:05:12.045
<v CM Carmen De La Rosa>followed by Brewer.

6217
06:05:12.605 --> 06:05:13.725
<v CM Virginia Maloney>Thank you, chair.

6218
06:05:14.125 --> 06:05:19.830
<v CM Virginia Maloney>As we all know, generative AI technology
is changing how we communicate, work, and

6219
06:05:19.830 --> 06:05:24.310
<v CM Virginia Maloney>how businesses operate. And the same
could be said of our government offices,

6220
06:05:24.390 --> 06:05:27.830
<v CM Virginia Maloney>both how we work and operate
either officially or unofficially.

6221
06:05:28.070 --> 06:05:31.990
<v CM Virginia Maloney>So I wanna know how New York City
agencies are adapting to the new reality.

6222
06:05:32.494 --> 06:05:34.975
<v CM Virginia Maloney>The first on the opportunity side,

6223
06:05:34.975 --> 06:05:39.215
<v CM Virginia Maloney>there are plenty of examples
of government wins of adopting

6224
06:05:39.215 --> 06:05:41.535
<v CM Virginia Maloney>useful technologies for the public good.

6225
06:05:42.014 --> 06:05:44.814
<v CM Virginia Maloney>In Connecticut, they have
real time translation

6226
06:05:44.814 --> 06:05:49.182
<v CM Virginia Maloney>of 80... In 80 languages for these
kinds of public meetings to increase

6227
06:05:49.182 --> 06:05:51.342
<v CM Virginia Maloney>accessibility in California.

6228
06:05:51.500 --> 06:05:54.380
<v CM Virginia Maloney>They automatically check
permit applications

6229
06:05:54.380 --> 06:05:55.182
<v CM Virginia Maloney>against

6230
06:05:55.340 --> 06:05:56.380
<v CM Virginia Maloney>18,000

6231
06:05:56.380 --> 06:05:58.861
<v CM Virginia Maloney>pages of of of

6232
06:05:59.102 --> 06:06:01.820
<v CM Virginia Maloney>code... Building code in
order to speed up reviews.

6233
06:06:02.234 --> 06:06:06.955
<v CM Virginia Maloney>So my first question is, what is our
equivalent win in New York? And if we don't

6234
06:06:06.955 --> 06:06:10.234
<v CM Virginia Maloney>have one yet, what's the plan to
identify those opportunities?

6235
06:06:10.635 --> 06:06:13.115
<v CM Virginia Maloney>And the second question
is around oversight.

6236
06:06:13.834 --> 06:06:17.650
<v CM Virginia Maloney>If I may wrap up, OTI published generative

6237
06:06:17.650 --> 06:06:20.611
<v CM Virginia Maloney>AI guidance for agencies last December.

6238
06:06:21.090 --> 06:06:23.250
<v CM Virginia Maloney>So as we're exploring these opportunities,

6239
06:06:23.650 --> 06:06:26.770
<v CM Virginia Maloney>is that guidance binding or is it advice?

6240
06:06:26.930 --> 06:06:29.650
<v CM Virginia Maloney>What happens if agencies
ignore that guidance?

6241
06:06:30.244 --> 06:06:32.885
<v CM Virginia Maloney>Has OTI been prescriptive

6242
06:06:32.885 --> 06:06:38.645
<v CM Virginia Maloney>about a list of generative AI or agentic
AI tools that have been reviewed and

6243
06:06:38.645 --> 06:06:42.805
<v CM Virginia Maloney>approved for city use that are use
cases that the city officially supports?

6244
06:06:43.340 --> 06:06:48.540
<v CM Virginia Maloney>And then, how are we thinking about
procurement when these models are constantly

6245
06:06:48.540 --> 06:06:53.580
<v CM Virginia Maloney>iterating and changing every few months
and when new... Every few days, truly, and

6246
06:06:53.580 --> 06:06:56.459
<v CM Virginia Maloney>when new technologies are
being introduced? Thank you.

6247
06:06:58.414 --> 06:07:00.895
<v Sarah Milstein (OTI)>Thank you for those
questions, council member.

6248
06:07:01.375 --> 06:07:05.855
<v Sarah Milstein (OTI)>I'm so glad you brought up the question
about wins because we're really interested

6249
06:07:05.855 --> 06:07:10.256
<v Sarah Milstein (OTI)>in both effective and responsible
use of AI, and effective

6250
06:07:10.256 --> 06:07:13.080
<v Sarah Milstein (OTI)>would cover the wins. I don't have,

6251
06:07:14.201 --> 06:07:17.560
<v Sarah Milstein (OTI)>a perfect list here, and I'd like to follow-up
with that because we'd really like to

6252
06:07:17.560 --> 06:07:19.320
<v Sarah Milstein (OTI)>represent the agency as well.

6253
06:07:20.760 --> 06:07:23.480
<v Sarah Milstein (OTI)>In terms of tools that we have approved,

6254
06:07:24.256 --> 06:07:25.615
<v Sarah Milstein (OTI)>broadly,

6255
06:07:25.775 --> 06:07:27.775
<v Sarah Milstein (OTI)>there are two sets of tools,

6256
06:07:28.176 --> 06:07:28.977
<v Sarah Milstein (OTI)>that

6257
06:07:28.977 --> 06:07:30.977
<v Sarah Milstein (OTI)>OTI approves centrally,

6258
06:07:31.535 --> 06:07:32.656
<v Sarah Milstein (OTI)>Microsoft

6259
06:07:32.656 --> 06:07:38.090
<v Sarah Milstein (OTI)>Office Suite and other Microsoft tools,
and that includes their Copilot chat. And

6260
06:07:38.090 --> 06:07:38.970
<v Sarah Milstein (OTI)>some of their,

6261
06:07:39.291 --> 06:07:42.250
<v Sarah Milstein (OTI)>some of their other tools
have AI features built in.

6262
06:07:42.810 --> 06:07:44.650
<v Sarah Milstein (OTI)>And we also have a central,

6263
06:07:45.291 --> 06:07:49.130
<v Sarah Milstein (OTI)>license for Adobe Creative
Cloud, and many of those tools

6264
06:07:49.610 --> 06:07:51.450
<v Sarah Milstein (OTI)>have AI features built in.

6265
06:07:52.596 --> 06:07:56.436
<v Sarah Milstein (OTI)>Other tools that... Other software
tools that are used by agencies are,

6266
06:07:56.836 --> 06:07:57.715
<v Sarah Milstein (OTI)>reviewed

6267
06:07:57.795 --> 06:08:00.916
<v Sarah Milstein (OTI)>for the most part on a
one off basis by OTI,

6268
06:08:01.395 --> 06:08:01.955
<v Sarah Milstein (OTI)>and

6269
06:08:02.195 --> 06:08:07.236
<v Sarah Milstein (OTI)>we are still determining the best way to
move forward with that process to make sure

6270
06:08:07.236 --> 06:08:10.730
<v Sarah Milstein (OTI)>that we are managing...
Identifying and managing risk

6271
06:08:10.810 --> 06:08:11.770
<v Sarah Milstein (OTI)>appropriately.

6272
06:08:13.290 --> 06:08:18.010
<v Sarah Milstein (OTI)>I think your question about procurement
is an excellent one. And part of what we

6273
06:08:18.010 --> 06:08:18.650
<v Sarah Milstein (OTI)>have

6274
06:08:18.970 --> 06:08:20.170
<v Sarah Milstein (OTI)>started to research

6275
06:08:20.746 --> 06:08:25.547
<v Sarah Milstein (OTI)>ahead of having staff at OAA.
We started to, talk with

6276
06:08:25.945 --> 06:08:26.746
<v Sarah Milstein (OTI)>various

6277
06:08:27.227 --> 06:08:31.307
<v Sarah Milstein (OTI)>experts in risk management to understand
some of the ways that we might think about

6278
06:08:31.307 --> 06:08:34.252
<v Sarah Milstein (OTI)>risk in an ongoing way when traditionally

6279
06:08:34.890 --> 06:08:39.451
<v Sarah Milstein (OTI)>all kinds of procurement has... Like,
there's a a gate at the beginning for various

6280
06:08:39.451 --> 06:08:40.252
<v Sarah Milstein (OTI)>concerns,

6281
06:08:40.730 --> 06:08:42.410
<v Sarah Milstein (OTI)>and then then

6282
06:08:42.410 --> 06:08:43.531
<v Sarah Milstein (OTI)>then it's rolling.

6283
06:08:44.010 --> 06:08:46.970
<v Sarah Milstein (OTI)>I think you're right. That's not
appropriate for this situation,

6284
06:08:47.210 --> 06:08:47.691
<v Sarah Milstein (OTI)>and

6285
06:08:48.115 --> 06:08:52.355
<v Sarah Milstein (OTI)>we should have more to discuss on that
in the coming months. But it's an active

6286
06:08:52.355 --> 06:08:53.795
<v Sarah Milstein (OTI)>question that we have as well.

6287
06:08:55.074 --> 06:08:57.236
<v CM Carmen De La Rosa>Thank you so much, council member,

6288
06:08:57.875 --> 06:08:58.676
<v CM Carmen De La Rosa>Juan,

6289
06:08:58.836 --> 06:09:00.355
<v CM Carmen De La Rosa>followed by Hanks,

6290
06:09:00.836 --> 06:09:06.470
<v CM Julie Won>followed by Brewer. Thank you so much,
chair De La Rosa. So for the panel, what I

6291
06:09:06.470 --> 06:09:10.710
<v CM Julie Won>have questions about is building off of
what council member Maloney just said. The

6292
06:09:10.710 --> 06:09:15.270
<v CM Julie Won>one... The agencies are allowed to currently
use Microsoft products and also Adobe.

6293
06:09:15.605 --> 06:09:19.686
<v CM Julie Won>But it's clear to me and the public
that they are using other AI tools.

6294
06:09:19.926 --> 06:09:24.166
<v CM Julie Won>So not just the passive tools like using
Zoom to transcribe their notes and for

6295
06:09:24.166 --> 06:09:27.365
<v CM Julie Won>follow ups, but in addition to
that, they're using ChatGPT,

6296
06:09:27.365 --> 06:09:31.110
<v CM Julie Won>they're using Claw, they're using Nano
Banana Pro because we're seeing it for

6297
06:09:31.110 --> 06:09:34.070
<v CM Julie Won>agencies like HPD, DOE, DEP.

6298
06:09:34.070 --> 06:09:36.791
<v CM Julie Won>They're creating flyers
for public consumption.

6299
06:09:37.030 --> 06:09:40.871
<v CM Julie Won>They're using it for their public emails.
They're using it for public messaging. So

6300
06:09:40.871 --> 06:09:45.465
<v CM Julie Won>what are you actually doing? Because I want
to know what the public policy is from

6301
06:09:45.465 --> 06:09:49.627
<v CM Julie Won>the mayor on how the agencies
are using AI and how you are

6302
06:09:50.105 --> 06:09:54.266
<v CM Julie Won>regulating it internally as well. And,
also, I would like to build off of that for

6303
06:09:54.266 --> 06:09:54.986
<v CM Julie Won>city contracts.

6304
06:09:56.520 --> 06:10:01.080
<v CM Julie Won>What are you doing currently for data
privacy writers and city contracts for all

6305
06:10:01.080 --> 06:10:02.760
<v CM Julie Won>procurement of AI companies

6306
06:10:02.840 --> 06:10:07.080
<v CM Julie Won>for making sure that our constituent
data is not being used to train their AI

6307
06:10:07.080 --> 06:10:07.479
<v CM Julie Won>models?

6308
06:10:08.516 --> 06:10:12.676
<v CM Julie Won>And is there a clause at all or a rider
in the city contract, especially with

6309
06:10:12.676 --> 06:10:16.436
<v CM Julie Won>copilot since that's the only real one
that you have? And people are just using

6310
06:10:16.436 --> 06:10:19.236
<v CM Julie Won>their own personal accounts or free
accounts for everything else that they're

6311
06:10:19.236 --> 06:10:21.076
<v CM Julie Won>currently using for their day to day jobs.

6312
06:10:22.318 --> 06:10:26.959
<v Sarah Milstein (OTI)>Thank you for the question, council member.
So let me first address the question

6313
06:10:26.959 --> 06:10:32.318
<v Sarah Milstein (OTI)>about the AI rider. We shared, that with
general councils across the city this

6314
06:10:32.318 --> 06:10:36.238
<v Sarah Milstein (OTI)>summer with an updated AI rider, and I
wanna just make sure I've got the right

6315
06:10:36.238 --> 06:10:38.135
<v Sarah Milstein (OTI)>language here, so I'm gonna pull that up.

6316
06:10:40.295 --> 06:10:41.096
<v Sarah Milstein (OTI)>Let's see.

6317
06:10:42.936 --> 06:10:47.496
<v CM Julie Won>Yeah. Because my privacy concerns are not
just for the contracts that exist, but it

6318
06:10:47.496 --> 06:10:53.629
<v CM Julie Won>is for the free willing use of of
agency staff that are using non city

6319
06:10:54.590 --> 06:10:56.750
<v CM Julie Won>enterprise licensed AI.

6320
06:10:56.830 --> 06:10:57.389
<v Sarah Milstein (OTI)>Yeah.

6321
06:10:58.030 --> 06:11:01.549
<v Sarah Milstein (OTI)>Yeah. So I'll first address the writer and
then we'll talk about the the shadow use.

6322
06:11:03.295 --> 06:11:03.855
<v Sarah Milstein (OTI)>So,

6323
06:11:04.734 --> 06:11:08.734
<v Sarah Milstein (OTI)>the rider, as I said, was distributed
to city... State agencies citywide this

6324
06:11:08.734 --> 06:11:12.734
<v Sarah Milstein (OTI)>summer, and it's intended to mitigate
risks and apply the city's policies and

6325
06:11:12.734 --> 06:11:14.814
<v Sarah Milstein (OTI)>standards to AI products and features.

6326
06:11:14.975 --> 06:11:18.570
<v Sarah Milstein (OTI)>It's available for all city software
contracts, and our legal team recommends that

6327
06:11:18.570 --> 06:11:23.370
<v Sarah Milstein (OTI)>agencies use it even where no AI
component is currently contemplated.

6328
06:11:23.770 --> 06:11:27.450
<v Sarah Milstein (OTI)>The writer includes terms relating
to intellectual property,

6329
06:11:27.690 --> 06:11:28.250
<v Sarah Milstein (OTI)>security,

6330
06:11:28.715 --> 06:11:31.916
<v Sarah Milstein (OTI)>how the city's data may
be used, and privacy.

6331
06:11:32.715 --> 06:11:33.355
<v Sarah Milstein (OTI)>And,

6332
06:11:33.674 --> 06:11:36.074
<v Sarah Milstein (OTI)>so that is now active and in use,

6333
06:11:36.715 --> 06:11:40.875
<v Sarah Milstein (OTI)>and I think that... With also a proactive
component of recommending that it's used

6334
06:11:40.875 --> 06:11:42.395
<v Sarah Milstein (OTI)>for all manner of technology.

6335
06:11:44.280 --> 06:11:49.640
<v Sarah Milstein (OTI)>In terms of the the shadow use, which I
think is probably both at the individual and

6336
06:11:49.640 --> 06:11:50.840
<v Sarah Milstein (OTI)>at the agency

6337
06:11:50.840 --> 06:11:51.480
<v Sarah Milstein (OTI)>level,

6338
06:11:51.640 --> 06:11:53.880
<v Sarah Milstein (OTI)>we are not in a

6339
06:11:54.840 --> 06:11:57.160
<v Sarah Milstein (OTI)>technical position to monitor that.

6340
06:11:58.443 --> 06:12:03.164
<v Sarah Milstein (OTI)>But I certainly... We, as an agency, share
concerns about that, and part of what

6341
06:12:03.164 --> 06:12:07.164
<v Sarah Milstein (OTI)>we're interested in doing is making sure
that agencies have the tools that they need

6342
06:12:07.164 --> 06:12:09.484
<v Sarah Milstein (OTI)>that are properly regulated.

6343
06:12:10.205 --> 06:12:14.490
<v Sarah Milstein (OTI)>Microsoft tools, for example, that...
Where data can't be shared or used to train

6344
06:12:14.490 --> 06:12:18.250
<v Sarah Milstein (OTI)>their models, that agencies have the tools
that they need to do the jobs that they

6345
06:12:18.250 --> 06:12:22.730
<v Sarah Milstein (OTI)>have so that they are not turning to
outside tools. So who is responsible if you

6346
06:12:22.730 --> 06:12:27.555
<v Sarah Milstein (OTI)>aren't able to regulate shadow use of AI
tools for a day to day use, especially with

6347
06:12:27.555 --> 06:12:30.355
<v Sarah Milstein (OTI)>constituent data? Who is
willing to regulate that?

6348
06:12:32.195 --> 06:12:34.436
<v Sarah Milstein (OTI)>Like to follow-up with
you on that question.

6349
06:12:34.914 --> 06:12:39.395
<v CM Carmen De La Rosa>Okay. Thank thank you, council members.
So I know the admin has a hard stop at

6350
06:12:39.395 --> 06:12:42.195
<v CM Carmen De La Rosa>05:30. We have about six
more member questions.

6351
06:12:42.690 --> 06:12:47.490
<v CM Carmen De La Rosa>So if you all could just hang on tight,
we're gonna try to get through this members.

6352
06:12:47.650 --> 06:12:49.410
<v CM Carmen De La Rosa>Please consolidate the questions.

6353
06:12:50.530 --> 06:12:55.490
<v CM Kamillah Hanks>We have our majority whip, councilor Moranx.
Thank you so much, chair. I'll be very,

6354
06:12:55.490 --> 06:12:58.174
<v CM Kamillah Hanks>very brief. With so you testified,

6355
06:12:58.334 --> 06:13:01.135
<v CM Kamillah Hanks>with all the, technology
that's changing rapidly,

6356
06:13:01.693 --> 06:13:04.014
<v CM Kamillah Hanks>even the best can't, you know, keep up. So

6357
06:13:05.295 --> 06:13:09.135
<v CM Kamillah Hanks>you... In your testimony, you said the
administration is continuing to evolve your

6358
06:13:09.135 --> 06:13:14.390
<v CM Kamillah Hanks>strategies to keep our city's infrastructure
safe, But not all threats to AI systems

6359
06:13:14.390 --> 06:13:17.990
<v CM Kamillah Hanks>are necessarily cyber attacks.
Can you expand on that?

6360
06:13:20.550 --> 06:13:21.750
<v CJ Dixon (NYC Cyber Command / OTI)>Yes. So

6361
06:13:23.190 --> 06:13:27.443
<v CJ Dixon (NYC Cyber Command / OTI)>when looking at the way that it is currently
described by the National Institute of

6362
06:13:27.443 --> 06:13:30.645
<v CJ Dixon (NYC Cyber Command / OTI)>Science and Technology as well as some
of the universities and academics who've

6363
06:13:30.645 --> 06:13:36.164
<v CJ Dixon (NYC Cyber Command / OTI)>created some of these systems, there are
functionally two baskets of AI risk. There

6364
06:13:36.164 --> 06:13:36.725
<v CJ Dixon (NYC Cyber Command / OTI)>is

6365
06:13:36.965 --> 06:13:37.523
<v CJ Dixon (NYC Cyber Command / OTI)>risk

6366
06:13:38.210 --> 06:13:43.330
<v CJ Dixon (NYC Cyber Command / OTI)>related to AI when an AI is attacked as a
computer system just like any other piece

6367
06:13:43.330 --> 06:13:44.210
<v CJ Dixon (NYC Cyber Command / OTI)>of software.

6368
06:13:44.371 --> 06:13:48.451
<v CJ Dixon (NYC Cyber Command / OTI)>And then there are risks related to
artificial intelligence that are truly novel

6369
06:13:48.611 --> 06:13:53.004
<v CJ Dixon (NYC Cyber Command / OTI)>to this new technology inherent in the way
that these artificial intelligence models

6370
06:13:53.004 --> 06:13:54.125
<v CJ Dixon (NYC Cyber Command / OTI)>may interact

6371
06:13:54.285 --> 06:13:56.205
<v CJ Dixon (NYC Cyber Command / OTI)>with society when deployed.

6372
06:13:56.365 --> 06:14:00.846
<v CJ Dixon (NYC Cyber Command / OTI)>That second category of risk
is a function of often a...

6373
06:14:01.324 --> 06:14:05.980
<v CJ Dixon (NYC Cyber Command / OTI)>How much society may trust the artificial
intelligence and how reliable that

6374
06:14:05.980 --> 06:14:08.060
<v CJ Dixon (NYC Cyber Command / OTI)>artificial intelligence system may be.

6375
06:14:08.860 --> 06:14:13.660
<v CJ Dixon (NYC Cyber Command / OTI)>Those two things can run afoul of what we
may expect as consumers or as a government

6376
06:14:13.660 --> 06:14:17.980
<v CJ Dixon (NYC Cyber Command / OTI)>entity, completely agnostic of whether or
not the system for which those AIs are

6377
06:14:17.980 --> 06:14:18.540
<v CJ Dixon (NYC Cyber Command / OTI)>deployed

6378
06:14:18.795 --> 06:14:22.875
<v CJ Dixon (NYC Cyber Command / OTI)>are actually cyber secure. So a system
can be completely cyber secure,

6379
06:14:23.035 --> 06:14:28.555
<v CJ Dixon (NYC Cyber Command / OTI)>but the AI model itself in interactions
with other humans in society writ large may

6380
06:14:28.555 --> 06:14:32.875
<v CJ Dixon (NYC Cyber Command / OTI)>take an action that is unexpected and
therefore result in some sort of disruption,

6381
06:14:33.380 --> 06:14:35.861
<v CJ Dixon (NYC Cyber Command / OTI)>agnostic of a cybersecurity incident.

6382
06:14:35.861 --> 06:14:38.021
<v CJ Dixon (NYC Cyber Command / OTI)>That's functionally the two categories.

6383
06:14:38.021 --> 06:14:41.380
<v CJ Dixon (NYC Cyber Command / OTI)>That's a very broad breakdown of
those two categories of risk.

6384
06:14:42.262 --> 06:14:46.102
<v CM Kamillah Hanks>So I'm gonna keep my my question just to
that. I think that we should talk offline

6385
06:14:46.102 --> 06:14:50.006
<v CM Kamillah Hanks>because my legislation requires the twenty
four hour reporting requirement to Cyber

6386
06:14:50.006 --> 06:14:53.766
<v CM Kamillah Hanks>Command in which your testimony has
some issues, and you you talk about,

6387
06:14:54.406 --> 06:14:58.326
<v CM Kamillah Hanks>how we're gonna push this through
to to city... To federal and, state

6388
06:14:58.727 --> 06:15:03.125
<v CM Gale Brewer>and what specific role do they play in that,
how how can we make this better, but, I

6389
06:15:03.125 --> 06:15:07.420
<v CM Gale Brewer>will relinquish that. Thank you, Chen....
You so much. Up next, we have Brewer

6390
06:15:07.420 --> 06:15:11.260
<v CM Gale Brewer>followed by Jose. Thank you very much. I
know you talked about the Microsoft and the

6391
06:15:11.260 --> 06:15:16.140
<v CM Gale Brewer>Adobe, but if an agency wants to begin
using a new AI model, a, what... How do they

6392
06:15:16.140 --> 06:15:21.875
<v CM Gale Brewer>go about it? Number two, does the city run
these models on its own data centers or

6393
06:15:21.875 --> 06:15:25.475
<v CM Gale Brewer>only on the ones that are on
the control of vendor clouds?

6394
06:15:25.955 --> 06:15:30.959
<v CM Gale Brewer>And then if you have a problem, like there's
a criteria, what is it for testing to

6395
06:15:30.959 --> 06:15:32.880
<v CM Gale Brewer>identify the malicious,

6396
06:15:33.120 --> 06:15:35.200
<v CM Gale Brewer>misaligned AI product?

6397
06:15:35.600 --> 06:15:40.160
<v CM Gale Brewer>And then how much does the city spend
each year on outside AI consultants?

6398
06:15:40.560 --> 06:15:42.560
<v CM Gale Brewer>I could go on, but those
are three, four questions.

6399
06:15:43.535 --> 06:15:46.096
<v CM Gale Brewer>I have 10 pages here. Okay.

6400
06:15:47.055 --> 06:15:48.814
<v Sarah Milstein (OTI)>Thank you for the questions.

6401
06:15:49.135 --> 06:15:50.656
<v Sarah Milstein (OTI)>Let's see. So...

6402
06:15:52.096 --> 06:15:53.455
<v Sarah Milstein (OTI)>The process. Yeah.

6403
06:15:56.260 --> 06:15:56.979
<v Sarah Milstein (OTI)>Typically,

6404
06:15:57.540 --> 06:15:58.500
<v Sarah Milstein (OTI)>when agencies

6405
06:15:58.500 --> 06:16:00.500
<v Sarah Milstein (OTI)>want to buy software,

6406
06:16:00.500 --> 06:16:04.900
<v Sarah Milstein (OTI)>a part of the process
is discussing with OTI

6407
06:16:04.979 --> 06:16:09.965
<v Sarah Milstein (OTI)>what they are buying and what they will be
using it for. And we have a cybersecurity

6408
06:16:09.965 --> 06:16:14.684
<v Sarah Milstein (OTI)>review process that is part
of that to ensure that the

6409
06:16:15.164 --> 06:16:18.523
<v Sarah Milstein (OTI)>software that is either
going to be bought or built,

6410
06:16:19.004 --> 06:16:21.324
<v Sarah Milstein (OTI)>will be safe. And as,

6411
06:16:21.564 --> 06:16:26.490
<v Sarah Milstein (OTI)>my colleague has said, that would include...
Cybersecurity includes the typical AI,

6412
06:16:26.890 --> 06:16:27.611
<v Sarah Milstein (OTI)>concerns.

6413
06:16:28.330 --> 06:16:29.371
<v Sarah Milstein (OTI)>So that's

6414
06:16:29.451 --> 06:16:30.330
<v Sarah Milstein (OTI)>process.

6415
06:16:31.210 --> 06:16:32.250
<v Sarah Milstein (OTI)>You had...

6416
06:16:33.050 --> 06:16:37.210
<v CM Gale Brewer>Next question, who... The clouds. Who's
whose clouds are they on? Who... Who's whose

6417
06:16:37.210 --> 06:16:41.924
<v CM Gale Brewer>data centers? Who's you know, where are
they? Yeah. Are you running them or is it

6418
06:16:41.924 --> 06:16:42.965
<v CM Gale Brewer>outside vendors?

6419
06:16:45.764 --> 06:16:46.645
<v Sarah Milstein (OTI)>So

6420
06:16:47.764 --> 06:16:50.564
<v Sarah Milstein (OTI)>to the best of my knowledge, there are no

6421
06:16:51.690 --> 06:16:54.650
<v Sarah Milstein (OTI)>city agencies running their own models.

6422
06:16:55.450 --> 06:16:58.170
<v Sarah Milstein (OTI)>But I would like to... It's
vendor clouds, basically.

6423
06:16:58.570 --> 06:17:01.530
<v CM Gale Brewer>Vendor clouds. Vendor clouds. Yes.

6424
06:17:01.530 --> 06:17:02.970
<v Sarah Milstein (OTI)>So, for example,

6425
06:17:03.930 --> 06:17:04.650
<v Sarah Milstein (OTI)>Microsoft

6426
06:17:05.645 --> 06:17:08.764
<v Sarah Milstein (OTI)>host software that agencies use.

6427
06:17:09.965 --> 06:17:12.285
<v Sarah Milstein (OTI)>But I'd like to follow-up

6428
06:17:12.523 --> 06:17:16.363
<v Sarah Milstein (OTI)>after the hearing to make sure I've got
the exact right information for you. Okay.

6429
06:17:16.363 --> 06:17:18.045
<v Sarah Milstein (OTI)>And if there's something unexpected,

6430
06:17:18.045 --> 06:17:20.363
<v CM Gale Brewer>misaligned AI product
behavior, what do you do?

6431
06:17:21.680 --> 06:17:24.320
<v Sarah Milstein (OTI)>That is a question for
my colleague, CJ Dixon.

6432
06:17:25.760 --> 06:17:29.121
<v CJ Dixon (NYC Cyber Command / OTI)>Councilwoman, that depends on what the
misaligned behavior is. If that misaligned

6433
06:17:29.121 --> 06:17:33.281
<v CJ Dixon (NYC Cyber Command / OTI)>behavior, specifically a cyber incident,
we take appropriate action based on what

6434
06:17:33.281 --> 06:17:35.135
<v CJ Dixon (NYC Cyber Command / OTI)>any cyber professional may do.

6435
06:17:35.455 --> 06:17:36.895
<v CJ Dixon (NYC Cyber Command / OTI)>There are novel

6436
06:17:36.895 --> 06:17:37.775
<v CJ Dixon (NYC Cyber Command / OTI)>misalignments

6437
06:17:37.775 --> 06:17:41.695
<v CJ Dixon (NYC Cyber Command / OTI)>that are born of this technology that do
have to be unpacked, and we look forward to

6438
06:17:41.695 --> 06:17:46.016
<v CJ Dixon (NYC Cyber Command / OTI)>working with the council as well as our
partners across New York City to understand

6439
06:17:46.541 --> 06:17:49.021
<v CJ Dixon (NYC Cyber Command / OTI)>when those misalignments do arise

6440
06:17:49.102 --> 06:17:53.342
<v CJ Dixon (NYC Cyber Command / OTI)>unrelated to existing risks that the city
already deals with, how do we actually

6441
06:17:53.342 --> 06:17:57.420
<v CJ Dixon (NYC Cyber Command / OTI)>engage with those particular novel risks.
Have you had some in the last, since you

6442
06:17:57.420 --> 06:17:58.140
<v CJ Dixon (NYC Cyber Command / OTI)>started your job?

6443
06:18:00.355 --> 06:18:04.676
<v CJ Dixon (NYC Cyber Command / OTI)>None related to what cyber
command's purview is. What... And

6444
06:18:04.994 --> 06:18:09.635
<v CJ Dixon (NYC Cyber Command / OTI)>that is primarily cyber security issues.
We've had no misalignments related to cyber

6445
06:18:09.635 --> 06:18:14.979
<v CJ Dixon (NYC Cyber Command / OTI)>security that have resulted in negative
consequences for the city from an AI model

6446
06:18:14.979 --> 06:18:16.180
<v CJ Dixon (NYC Cyber Command / OTI)>very specifically.

6447
06:18:16.580 --> 06:18:17.940
<v CM Gale Brewer>Alright. That sounds very...

6448
06:18:18.500 --> 06:18:19.860
<v CM Gale Brewer>I don't know what it sounds.

6449
06:18:20.260 --> 06:18:24.420
<v CM Gale Brewer>How much outside AI consultant
dollars are being spent?

6450
06:18:29.273 --> 06:18:33.035
<v Sarah Milstein (OTI)>Council member, could you clarify what
you mean by AI consultants? Yeah.

6451
06:18:33.834 --> 06:18:37.434
<v CM Gale Brewer>How many people are you paying? I mean,
I've been doing this work a long time, so I

6452
06:18:37.434 --> 06:18:41.530
<v CM Gale Brewer>know sort of the answer.
But how much outside

6453
06:18:41.530 --> 06:18:42.410
<v CM Gale Brewer>consultants

6454
06:18:42.410 --> 06:18:43.690
<v CM Gale Brewer>are being paid

6455
06:18:43.930 --> 06:18:48.090
<v CM Gale Brewer>on terms of AI? In other words, how many
people are you paying to work with you on

6456
06:18:48.090 --> 06:18:50.410
<v CM Gale Brewer>AI issues outside vendors,

6457
06:18:51.370 --> 06:18:52.010
<v CM Gale Brewer>consultants?

6458
06:18:55.477 --> 06:18:59.076
<v Sarah Milstein (OTI)>None that I'm aware of, but I would like
to follow-up with you to make sure that we

6459
06:18:59.076 --> 06:18:59.877
<v Sarah Milstein (OTI)>can accurate information....

6460
06:19:00.596 --> 06:19:03.957
<v CM Gale Brewer>House. Is that what you're saying?
Everything to do in terms of strategy,

6461
06:19:04.195 --> 06:19:05.316
<v CM Gale Brewer>thought process,

6462
06:19:05.795 --> 06:19:06.516
<v CM Gale Brewer>future,

6463
06:19:06.596 --> 06:19:07.557
<v CM Gale Brewer>it's all in house?

6464
06:19:08.432 --> 06:19:13.710
<v CM Carmen De La Rosa>Yes. K. Thank you. Thank you. Council
member Osei, followed by deputy speaker

6465
06:19:13.710 --> 06:19:14.592
<v CM Carmen De La Rosa>Williams.

6466
06:19:14.830 --> 06:19:18.672
<v CM Chi Ossé>Thank you so much. I just
wanted to refer to the

6467
06:19:18.990 --> 06:19:21.390
<v CM Chi Ossé>recent hacking situation in Australia.

6468
06:19:22.236 --> 06:19:27.516
<v CM Chi Ossé>That was proof that AI agents do have
that ability to hack into our government

6469
06:19:27.516 --> 06:19:28.156
<v CM Chi Ossé>systems.

6470
06:19:28.557 --> 06:19:32.795
<v CM Chi Ossé>As I'm sure you know, in June, a rogue
OpenAI agent hacked into an Australian

6471
06:19:32.795 --> 06:19:35.195
<v CM Chi Ossé>government statistics portal portal

6472
06:19:35.390 --> 06:19:39.150
<v CM Chi Ossé>containing private data from Australia's
universal health care system.

6473
06:19:39.550 --> 06:19:43.791
<v CM Chi Ossé>At this current moment, if a similar hack
that happened to Australia happened here

6474
06:19:43.791 --> 06:19:48.510
<v CM Chi Ossé>in New York, do you feel that the city and
the admin has a sufficient plan to defend

6475
06:19:48.510 --> 06:19:53.064
<v CM Chi Ossé>against the hacks and protect any sensitive
data? And two, does the city currently

6476
06:19:53.064 --> 06:19:57.225
<v CM Chi Ossé>have a responsive liaison from each of
these companies to speak to in case of a

6477
06:19:57.225 --> 06:19:57.785
<v CM Chi Ossé>hack?

6478
06:19:58.824 --> 06:20:01.625
<v CJ Dixon (NYC Cyber Command / OTI)>To the first question, yes. We do have the

6479
06:20:01.900 --> 06:20:06.140
<v CJ Dixon (NYC Cyber Command / OTI)>people process technology and some of the
policies in place to deal with an incident

6480
06:20:06.140 --> 06:20:08.060
<v CJ Dixon (NYC Cyber Command / OTI)>similar to what happened in Australia.

6481
06:20:08.300 --> 06:20:11.820
<v CJ Dixon (NYC Cyber Command / OTI)>The reason being is even though
an AI was involved, the hacking

6482
06:20:11.820 --> 06:20:12.621
<v CJ Dixon (NYC Cyber Command / OTI)>process,

6483
06:20:12.621 --> 06:20:17.205
<v CJ Dixon (NYC Cyber Command / OTI)>the attack chain, if you will, happen in
the same way that if a human being were to

6484
06:20:17.205 --> 06:20:20.727
<v CJ Dixon (NYC Cyber Command / OTI)>execute that attack. At the end of the
day, the underlying infrastructure,

6485
06:20:20.805 --> 06:20:24.965
<v CJ Dixon (NYC Cyber Command / OTI)>I will use air quotes here, doesn't actually
care if the person at the end of the

6486
06:20:24.965 --> 06:20:29.205
<v CJ Dixon (NYC Cyber Command / OTI)>other end of the keyboard is
a human or a silicon based

6487
06:20:29.690 --> 06:20:30.490
<v CJ Dixon (NYC Cyber Command / OTI)>entity.

6488
06:20:30.570 --> 06:20:33.530
<v CJ Dixon (NYC Cyber Command / OTI)>The attack chain still remained
functionally the same.

6489
06:20:33.850 --> 06:20:38.729
<v CJ Dixon (NYC Cyber Command / OTI)>Therefore, our team is equipped and
trained very specifically to identify these

6490
06:20:38.729 --> 06:20:40.170
<v CJ Dixon (NYC Cyber Command / OTI)>indicators of compromise,

6491
06:20:40.490 --> 06:20:44.809
<v CJ Dixon (NYC Cyber Command / OTI)>identify the attack patterns, and then be
able to respond accordingly at multiple

6492
06:20:44.809 --> 06:20:45.209
<v CJ Dixon (NYC Cyber Command / OTI)>points.

6493
06:20:46.785 --> 06:20:49.025
<v CJ Dixon (NYC Cyber Command / OTI)>So we are prepared

6494
06:20:49.025 --> 06:20:51.266
<v CJ Dixon (NYC Cyber Command / OTI)>to address those types of issues,

6495
06:20:51.426 --> 06:20:53.666
<v CJ Dixon (NYC Cyber Command / OTI)>and we do have points of contacts

6496
06:20:53.666 --> 06:20:57.906
<v CJ Dixon (NYC Cyber Command / OTI)>when we are working as city agencies
with vendors who are under our contract

6497
06:20:59.230 --> 06:21:04.190
<v CJ Dixon (NYC Cyber Command / OTI)>to engage with third parties as they get
breached or if their product is the result

6498
06:21:04.190 --> 06:21:08.670
<v CJ Dixon (NYC Cyber Command / OTI)>of a breach. So if they are under contract
within the city by their service level

6499
06:21:08.670 --> 06:21:10.430
<v CJ Dixon (NYC Cyber Command / OTI)>agreements are required

6500
06:21:10.430 --> 06:21:15.344
<v CJ Dixon (NYC Cyber Command / OTI)>to keep us informed of when they have
downtimes or when their products are involved

6501
06:21:15.344 --> 06:21:16.305
<v CJ Dixon (NYC Cyber Command / OTI)>in a breach.

6502
06:21:16.625 --> 06:21:20.145
<v CM Chi Ossé>Thank you for that response. And in regards
to my legislation on an emergency plan,

6503
06:21:20.145 --> 06:21:24.225
<v CM Chi Ossé>and I know that you indicated that there's
already a plan in case, you know, an

6504
06:21:24.225 --> 06:21:27.264
<v CM Chi Ossé>emergency takes place. Am I
correct with that assumption?

6505
06:21:29.480 --> 06:21:33.400
<v CJ Dixon (NYC Cyber Command / OTI)>Yes. So there are cyber incident response
plans as well as my colleagues from NISM.

6506
06:21:33.400 --> 06:21:35.960
<v CJ Dixon (NYC Cyber Command / OTI)>There's also ongoing work for risk,

6507
06:21:36.840 --> 06:21:41.480
<v CJ Dixon (NYC Cyber Command / OTI)>response plans in general across the board,
and that is for every category of risk

6508
06:21:41.480 --> 06:21:44.096
<v CJ Dixon (NYC Cyber Command / OTI)>that the city may deal
with that might disrupt

6509
06:21:44.096 --> 06:21:49.135
<v CJ Dixon (NYC Cyber Command / OTI)>city infrastructure or city services. Is
that publicly available, or is that shared

6510
06:21:49.135 --> 06:21:53.215
<v CJ Dixon (NYC Cyber Command / OTI)>with the council? No. That is by design.
We do not want to share that publicly

6511
06:21:53.215 --> 06:21:57.950
<v CJ Dixon (NYC Cyber Command / OTI)>because if that is... The more public
Of course. Aware that someone is of an

6512
06:21:57.950 --> 06:22:00.590
<v CJ Dixon (NYC Cyber Command / OTI)>incident response plan, the more likely
they can circumvent it because that's

6513
06:22:00.590 --> 06:22:04.110
<v CJ Dixon (NYC Cyber Command / OTI)>something that you do wanna keep close
hold and classified. Okay. Thank you very

6514
06:22:04.110 --> 06:22:05.630
<v CJ Dixon (NYC Cyber Command / OTI)>much. Thank you so much.

6515
06:22:06.110 --> 06:22:08.590
<v CM Carmen De La Rosa>Council member Huni followed
by deputy speaker Williams.

6516
06:22:10.984 --> 06:22:14.664
<v CM Shahana Hanif>Oh, I thought the deputy speaker was going
before me, but I'm late. I messed up.

6517
06:22:14.664 --> 06:22:15.865
<v CM Shahana Hanif>Okay. No worries.

6518
06:22:17.064 --> 06:22:22.426
<v CM Shahana Hanif>I have a question first for deputy
commissioner Milstein. How does OTI

6519
06:22:22.504 --> 06:22:27.960
<v CM Shahana Hanif>ensure that the automated prescreening
tools and fraud detection algorithms used by

6520
06:22:27.960 --> 06:22:29.000
<v CM Shahana Hanif>the HRA

6521
06:22:29.000 --> 06:22:30.920
<v CM Shahana Hanif>do not create disparate

6522
06:22:30.920 --> 06:22:35.720
<v CM Shahana Hanif>impacts or algorithmic bias
against low income applicants?

6523
06:22:36.281 --> 06:22:39.000
<v CM Shahana Hanif>And then a question I have
for commissioner Levine,

6524
06:22:39.455 --> 06:22:41.936
<v CM Shahana Hanif>The testimony indicated that DCWP

6525
06:22:41.936 --> 06:22:43.695
<v CM Shahana Hanif>supports a complaint process,

6526
06:22:43.855 --> 06:22:48.414
<v CM Shahana Hanif>but you all raised concerns about staffing
and expertise. What specific resources

6527
06:22:48.414 --> 06:22:49.855
<v CM Shahana Hanif>would you need to investigate

6528
06:22:50.016 --> 06:22:51.615
<v CM Shahana Hanif>AI complaints effectively,

6529
06:22:52.190 --> 06:22:57.310
<v CM Shahana Hanif>and what protections can New Yorkers rely
on while that capacity is being built?

6530
06:23:00.030 --> 06:23:04.510
<v Samuel Levine (DCWP)>Sure. Well, thank you, council member.
I can start. And I wanna be clear. AI AI

6531
06:23:04.510 --> 06:23:07.494
<v Samuel Levine (DCWP)>complaints can can take
many different flavors.

6532
06:23:07.654 --> 06:23:12.455
<v Samuel Levine (DCWP)>If we hear from a worker who tells us
they believe their pay was incorrect,

6533
06:23:12.455 --> 06:23:15.654
<v Samuel Levine (DCWP)>maybe they didn't hit the minimum pay
rate because of an algorithm, that is

6534
06:23:15.654 --> 06:23:17.174
<v Samuel Levine (DCWP)>something we can fully address.

6535
06:23:17.414 --> 06:23:21.422
<v Samuel Levine (DCWP)>If we have someone like Jacob
Coxen come to us and say that,

6536
06:23:21.980 --> 06:23:24.701
<v Samuel Levine (DCWP)>Anthropic is developing models that could,

6537
06:23:26.060 --> 06:23:28.220
<v Samuel Levine (DCWP)>lead to catastrophic consequences,

6538
06:23:28.300 --> 06:23:32.541
<v Samuel Levine (DCWP)>that is not something we have the capacity
to do. So what we said in our testimony

6539
06:23:32.541 --> 06:23:36.285
<v Samuel Levine (DCWP)>is the New York attorney general
already has an Internet bureau

6540
06:23:36.285 --> 06:23:39.805
<v Samuel Levine (DCWP)>set up. They've already put out a call
for whistleblowers. I believe they already

6541
06:23:39.805 --> 06:23:43.004
<v Samuel Levine (DCWP)>have staff capable of
processing these complaints.

6542
06:23:43.965 --> 06:23:48.285
<v Samuel Levine (DCWP)>We'd be very happy to work with the council
to make sure complaint to get to the AG.

6543
06:23:48.285 --> 06:23:52.809
<v Samuel Levine (DCWP)>As the speaker has said, thousands of these
AI employees work right here in the five

6544
06:23:52.809 --> 06:23:57.370
<v Samuel Levine (DCWP)>boroughs. They know that these companies
are making reckless decisions in pursuit of

6545
06:23:57.370 --> 06:24:01.450
<v Samuel Levine (DCWP)>profits. We do wanna make sure that these
employees have a place to go. And if they

6546
06:24:01.450 --> 06:24:04.945
<v Samuel Levine (DCWP)>come to us and it's not something we can
handle, we'd wanna work with the state to

6547
06:24:04.945 --> 06:24:07.906
<v Samuel Levine (DCWP)>make sure those complaints can be addressed
adequately. But right now, is the city

6548
06:24:07.906 --> 06:24:12.227
<v Samuel Levine (DCWP)>working with the AG's office, or is that
a relationship that's not formalized?

6549
06:24:12.227 --> 06:24:16.307
<v Samuel Levine (DCWP)>We have a very close relationship with
the AG's office. We we speak constantly.

6550
06:24:20.059 --> 06:24:24.700
<v Sarah Milstein (OTI)>Can I ask you to repeat the question about
HRA? Absolutely. So it's basically trying

6551
06:24:24.700 --> 06:24:25.580
<v Sarah Milstein (OTI)>to understand,

6552
06:24:26.940 --> 06:24:27.660
<v CM Shahana Hanif>our

6553
06:24:27.740 --> 06:24:28.379
<v CM Shahana Hanif>auto...

6554
06:24:28.940 --> 06:24:31.898
<v CM Shahana Hanif>Automated prescreening tools
because I understand that not

6555
06:24:32.246 --> 06:24:36.086
<v CM Shahana Hanif>not... Humans are not checking all of
the applications. There's some level of

6556
06:24:36.086 --> 06:24:38.006
<v CM Shahana Hanif>automated algorithmic

6557
06:24:38.326 --> 06:24:42.885
<v CM Shahana Hanif>formula that is determining how applicants
get pooled when it comes to public

6558
06:24:42.885 --> 06:24:43.605
<v CM Shahana Hanif>benefits.

6559
06:24:43.766 --> 06:24:45.365
<v CM Shahana Hanif>So I'm curious if,

6560
06:24:46.660 --> 06:24:51.700
<v CM Shahana Hanif>how OTI ensures that automated prescreening
tools, fraud detection algorithms

6561
06:24:51.940 --> 06:24:54.740
<v CM Shahana Hanif>used by HRA are not creating disparate

6562
06:24:54.740 --> 06:24:56.340
<v CM Shahana Hanif>impacts or bias.

6563
06:24:56.980 --> 06:25:00.980
<v Sarah Milstein (OTI)>Thank you for the question. We
really share a concern about

6564
06:25:01.346 --> 06:25:02.146
<v Sarah Milstein (OTI)>the bias

6565
06:25:03.025 --> 06:25:04.305
<v Sarah Milstein (OTI)>and impacts

6566
06:25:04.545 --> 06:25:05.346
<v Sarah Milstein (OTI)>on

6567
06:25:05.586 --> 06:25:10.305
<v Sarah Milstein (OTI)>not only workers but on, New Yorkers.
Because this is really detailed and specific,

6568
06:25:10.305 --> 06:25:14.064
<v Sarah Milstein (OTI)>though, I do wanna make sure to check
back with HRA and make sure I've got the

6569
06:25:14.064 --> 06:25:17.871
<v Sarah Milstein (OTI)>details of this one correct. So we'll come
back to you on that for sure. Appreciate

6570
06:25:17.871 --> 06:25:18.432
<v Sarah Milstein (OTI)>it.

6571
06:25:19.150 --> 06:25:22.990
<v Unidentified [?]>Thank you so much. Council member, I just
wanted to follow-up on one point you had

6572
06:25:22.990 --> 06:25:27.310
<v Unidentified [?]>about potential resource impacts. I
think on that particular issue, we had

6573
06:25:27.310 --> 06:25:31.635
<v Unidentified [?]>identified very preliminarily of 14
lines for approximately 1,200,000.

6574
06:25:31.955 --> 06:25:35.074
<v Unidentified [?]>But I Wait. Could you repeat that one
more time? 14 lines for approximately

6575
06:25:35.074 --> 06:25:36.195
<v Unidentified [?]>1,200,000.

6576
06:25:36.354 --> 06:25:39.314
<v Unidentified [?]>But, again, that's a very preliminary
estimate, and we work on the council with the

6577
06:25:39.314 --> 06:25:40.594
<v Unidentified [?]>bill for a final version.

6578
06:25:42.514 --> 06:25:44.770
<v CM Carmen De La Rosa>Thank you. Deputy speaker Williams?

6579
06:25:44.930 --> 06:25:45.810
<v CM Nantasha Williams>Thank you, chair.

6580
06:25:46.690 --> 06:25:49.010
<v CM Nantasha Williams>I am looking at your testimony

6581
06:25:49.250 --> 06:25:52.451
<v CM Nantasha Williams>and your comments about the chair's bill.

6582
06:25:52.930 --> 06:25:55.410
<v CM Nantasha Williams>Was that local law 25,

6583
06:25:55.410 --> 06:25:56.610
<v CM Nantasha Williams>which was my bill,

6584
06:25:57.865 --> 06:26:02.824
<v CM Nantasha Williams>is... The study is currently underway. We
reached out... The compliance division of

6585
06:26:02.824 --> 06:26:06.744
<v CM Nantasha Williams>the council reached out, I think,
twice. I know it just went into

6586
06:26:07.064 --> 06:26:10.664
<v CM Nantasha Williams>effect in January, but would be
helpful to know if you have a timeline.

6587
06:26:11.420 --> 06:26:14.620
<v CM Nantasha Williams>So I know it might take
time, but how much time?

6588
06:26:15.100 --> 06:26:15.740
<v CM Nantasha Williams>Because,

6589
06:26:15.900 --> 06:26:19.740
<v CM Nantasha Williams>clearly, I think today's hearing is
evidence of the importance of really

6590
06:26:19.740 --> 06:26:21.420
<v CM Nantasha Williams>understanding AI's impact,

6591
06:26:21.740 --> 06:26:25.260
<v CM Nantasha Williams>on the municipal workforce. So just wanted
to know if you could share a timeline for

6592
06:26:25.260 --> 06:26:28.805
<v CM Nantasha Williams>us. I know the comments we
received back was just that

6593
06:26:29.285 --> 06:26:32.805
<v CM Nantasha Williams>there's no updates, but we
haven't received a timeline.

6594
06:26:33.363 --> 06:26:36.484
<v Sarah Milstein (OTI)>Yeah. Thank you for the question, and
thank you for the legislation. I think it's

6595
06:26:36.484 --> 06:26:37.764
<v Sarah Milstein (OTI)>really an important study.

6596
06:26:38.740 --> 06:26:43.300
<v Sarah Milstein (OTI)>I anticipate that we will have
pieces of it over the course of

6597
06:26:43.860 --> 06:26:44.980
<v Sarah Milstein (OTI)>2027,

6598
06:26:44.980 --> 06:26:49.379
<v Sarah Milstein (OTI)>and we will share updates as we have them.
I think rather... Because this is so,

6599
06:26:49.860 --> 06:26:54.086
<v Sarah Milstein (OTI)>timely and important, we don't have to
just wait till the end. We can be in

6600
06:26:54.086 --> 06:26:57.686
<v Sarah Milstein (OTI)>partnership with you and rolling
out information as we learn more.

6601
06:26:58.246 --> 06:27:02.246
<v CM Carmen De La Rosa>Thank you. I look forward to working with
you on that. Thank you. And the last

6602
06:27:02.246 --> 06:27:03.766
<v CM Carmen De La Rosa>question for this panel.

6603
06:27:04.246 --> 06:27:09.020
<v CM Carmen De La Rosa>This is from council member Selvina Brooks
Powers. She says my district is home to

6604
06:27:09.020 --> 06:27:14.060
<v CM Carmen De La Rosa>JFK Airport and to the coastal Rockaway
neighborhoods that depend on emergency

6605
06:27:14.060 --> 06:27:14.780
<v CM Carmen De La Rosa>alerts.

6606
06:27:15.100 --> 06:27:20.220
<v CM Carmen De La Rosa>What might an AI driven attack on
emergency systems or a major in... Or major

6607
06:27:20.220 --> 06:27:25.154
<v CM Carmen De La Rosa>infrastructure look like? Is there
currently a plan for an AI driven attack on

6608
06:27:25.154 --> 06:27:26.436
<v CM Carmen De La Rosa>critical infrastructure

6609
06:27:26.436 --> 06:27:27.875
<v CM Carmen De La Rosa>and relevant emergency

6610
06:27:27.875 --> 06:27:28.676
<v CM Carmen De La Rosa>systems?

6611
06:27:28.994 --> 06:27:30.355
<v CM Carmen De La Rosa>And then for NICEM,

6612
06:27:30.355 --> 06:27:32.516
<v CM Carmen De La Rosa>during storms and flooding emergencies,

6613
06:27:32.676 --> 06:27:37.050
<v CM Carmen De La Rosa>it can be hard for New Yorkers to know what
they're seeing on social media is real

6614
06:27:37.050 --> 06:27:38.490
<v CM Carmen De La Rosa>or AI generated.

6615
06:27:38.570 --> 06:27:40.092
<v CM Carmen De La Rosa>How is NICEM working

6616
06:27:40.250 --> 06:27:44.730
<v CM Carmen De La Rosa>to make their messaging clearly legitimate
and to cut through the noise of AI

6617
06:27:44.730 --> 06:27:45.611
<v CM Carmen De La Rosa>generated rumors?

6618
06:27:49.936 --> 06:27:52.256
<v Sarah Milstein (OTI)>For the question on notifications,

6619
06:27:53.055 --> 06:27:55.455
<v Sarah Milstein (OTI)>I'll defer to my colleague CJ Dixon.

6620
06:27:57.135 --> 06:28:01.979
<v CJ Dixon (NYC Cyber Command / OTI)>Yes. So for an AI based attack that is
a... Specifically a cyber security attack,

6621
06:28:01.979 --> 06:28:06.620
<v CJ Dixon (NYC Cyber Command / OTI)>there are notification requirements that
already exist under law. Seventy two hours

6622
06:28:06.620 --> 06:28:10.860
<v CJ Dixon (NYC Cyber Command / OTI)>for a confirmed breach, twenty four hours
for a breach related with extortion,

6623
06:28:10.940 --> 06:28:15.055
<v CJ Dixon (NYC Cyber Command / OTI)>and then thirty days for victim
notification. So there are plans in place for

6624
06:28:15.055 --> 06:28:18.975
<v CJ Dixon (NYC Cyber Command / OTI)>incident response in the event of a cyber
attack against any critical infrastructure

6625
06:28:18.975 --> 06:28:23.375
<v CJ Dixon (NYC Cyber Command / OTI)>owned by the city, and that includes if
that attack is perpetrated by an artificial

6626
06:28:23.375 --> 06:28:24.654
<v CJ Dixon (NYC Cyber Command / OTI)>intelligence system.

6627
06:28:26.895 --> 06:28:30.160
<v Benjamin Krakauer (NYC Emergency Management)>Thanks. And on the, New York City emergency
management front, I'll make two two

6628
06:28:30.160 --> 06:28:31.360
<v Benjamin Krakauer (NYC Emergency Management)>quick points. One,

6629
06:28:32.400 --> 06:28:36.000
<v Benjamin Krakauer (NYC Emergency Management)>I think that is a real risk. I appreciate
the council member raising it. It's one of

6630
06:28:36.000 --> 06:28:41.115
<v Benjamin Krakauer (NYC Emergency Management)>the reasons that we invest very heavily in
backup systems. So several months ago, a

6631
06:28:41.115 --> 06:28:43.596
<v Benjamin Krakauer (NYC Emergency Management)>very, well known and established,

6632
06:28:43.834 --> 06:28:48.314
<v Benjamin Krakauer (NYC Emergency Management)>notification provider had a seven hour
global outage at which... Where where

6633
06:28:48.314 --> 06:28:52.234
<v Benjamin Krakauer (NYC Emergency Management)>notifications were not able to be sent
out. That's something that we take very

6634
06:28:52.234 --> 06:28:56.170
<v Benjamin Krakauer (NYC Emergency Management)>seriously and which is why for over a decade
now, New York City Emergency Management

6635
06:28:56.170 --> 06:28:57.770
<v Benjamin Krakauer (NYC Emergency Management)>has invested in the backup system.

6636
06:28:58.170 --> 06:29:02.730
<v Benjamin Krakauer (NYC Emergency Management)>So as you went across the country, you saw
many large jurisdictions saying to their

6637
06:29:02.730 --> 06:29:04.330
<v Benjamin Krakauer (NYC Emergency Management)>to their constituents,

6638
06:29:04.730 --> 06:29:08.410
<v Benjamin Krakauer (NYC Emergency Management)>you know, sign up for this WhatsApp group
or follow us on social media if you want

6639
06:29:08.410 --> 06:29:12.273
<v Benjamin Krakauer (NYC Emergency Management)>alerts tonight. In New York City, we did
not have to do that because we had a backup

6640
06:29:12.273 --> 06:29:16.754
<v Benjamin Krakauer (NYC Emergency Management)>system that continued to work, that
disruption was averted. As far as,

6641
06:29:18.273 --> 06:29:20.434
<v Benjamin Krakauer (NYC Emergency Management)>you know, misdis or malinformation,

6642
06:29:20.434 --> 06:29:25.172
<v Benjamin Krakauer (NYC Emergency Management)>you know, generated by AI, that is that is
a valid concern. That's why we recommend

6643
06:29:25.172 --> 06:29:28.371
<v Benjamin Krakauer (NYC Emergency Management)>that New Yorkers follow us on our
official channels, which are verified,

6644
06:29:28.771 --> 06:29:32.932
<v Benjamin Krakauer (NYC Emergency Management)>through Notify NYC. You can
sign up at notify at nyc.gov/

6645
06:29:32.932 --> 06:29:33.650
<v Benjamin Krakauer (NYC Emergency Management)>notify.

6646
06:29:33.970 --> 06:29:37.730
<v Benjamin Krakauer (NYC Emergency Management)>You can sign up by short code or download
the Notify NYC mobile app that we operate

6647
06:29:37.730 --> 06:29:39.172
<v Benjamin Krakauer (NYC Emergency Management)>in concert with OTI.

6648
06:29:39.895 --> 06:29:40.773
<v CM Carmen De La Rosa>Thank

6649
06:29:40.773 --> 06:29:41.574
<v CM Carmen De La Rosa>you so much.

6650
06:29:42.215 --> 06:29:45.895
<v CM Carmen De La Rosa>Council member Brooks Powers wanted me
to note that, for example, during the

6651
06:29:45.895 --> 06:29:48.855
<v CM Carmen De La Rosa>nor'easter, there was a video that
was going around about a tornado,

6652
06:29:49.334 --> 06:29:53.654
<v CM Carmen De La Rosa>that apparently caused a lot of
confusion and, and, you know, chaos.

6653
06:29:53.814 --> 06:29:55.574
<v CM Carmen De La Rosa>And so she wanted to
share that for context.

6654
06:29:57.299 --> 06:30:00.420
<v Speaker Julie Menin>Okay. Thank you so much. I
really wanna thank this panel.

6655
06:30:00.738 --> 06:30:04.818
<v Speaker Julie Menin>We appreciate your comments on the
legislation. We really look forward to working

6656
06:30:04.818 --> 06:30:06.660
<v Speaker Julie Menin>collaboratively with the administration

6657
06:30:06.979 --> 06:30:11.904
<v Speaker Julie Menin>on the various pieces of legislation and
appreciate your testimony today. Thank you.

6658
06:30:13.984 --> 06:30:18.385
<v CM Carmen De La Rosa [?]>Thank you all for coming. I now open the
hearing for public testimony. I remind

6659
06:30:18.385 --> 06:30:22.225
<v CM Carmen De La Rosa [?]>members of the public that this
is a formal government proceeding

6660
06:30:22.305 --> 06:30:26.639
<v CM Carmen De La Rosa [?]>and that decorum shall be observed at all
times. As such, member of the public shall

6661
06:30:26.639 --> 06:30:31.520
<v CM Carmen De La Rosa [?]>remain silent at all times in the chamber.
The witness table is reserved for people

6662
06:30:31.520 --> 06:30:32.959
<v CM Carmen De La Rosa [?]>who wish to testify.

6663
06:30:33.200 --> 06:30:37.120
<v CM Carmen De La Rosa [?]>No video recording or photography is
allowed from the witness table. Further,

6664
06:30:37.200 --> 06:30:40.799
<v CM Carmen De La Rosa [?]>members of the public may not present
audio or video video recordings

6665
06:30:40.875 --> 06:30:45.994
<v CM Carmen De La Rosa [?]>as testimony, but may submit transcripts
of such recordings to the sergeant at arm

6666
06:30:45.994 --> 06:30:50.154
<v CM Carmen De La Rosa [?]>for inclusion in the hearing record. If you
wish to speak at today's hearing and you

6667
06:30:50.154 --> 06:30:54.314
<v CM Carmen De La Rosa [?]>have not done so, please fill out an
appearance card with the sergeant at arms and

6668
06:30:54.314 --> 06:30:55.516
<v CM Carmen De La Rosa [?]>wait to be recognized.

6669
06:30:55.756 --> 06:30:56.555
<v CM Carmen De La Rosa [?]>When recognized,

6670
06:30:57.090 --> 06:31:01.090
<v CM Carmen De La Rosa [?]>you'll have two minutes to speak on today's
hearing topic, which is examining the

6671
06:31:01.090 --> 06:31:03.650
<v CM Carmen De La Rosa [?]>risk posed by artificial intelligence

6672
06:31:03.810 --> 06:31:05.330
<v CM Carmen De La Rosa [?]>and proposed legislation.

6673
06:31:05.490 --> 06:31:09.490
<v CM Carmen De La Rosa [?]>If you have written statement or additional
written statement you wish to submit for

6674
06:31:09.490 --> 06:31:13.875
<v CM Carmen De La Rosa [?]>the record, please provide a copy of that
testimony to the sergeant at arms. You may

6675
06:31:13.875 --> 06:31:19.236
<v CM Carmen De La Rosa [?]>also email the written testimony to
testimony at council dot n y c dot gov or

6676
06:31:19.475 --> 06:31:23.555
<v CM Carmen De La Rosa [?]>within seventy two hours of this hearing.
Audio and video recordings will not be

6677
06:31:23.555 --> 06:31:25.600
<v CM Carmen De La Rosa [?]>accepted. Our first panel

6678
06:31:25.920 --> 06:31:28.000
<v CM Carmen De La Rosa [?]>includes Nate Soares,

6679
06:31:29.360 --> 06:31:30.720
<v CM Carmen De La Rosa [?]>doctor Julia

6680
06:31:31.920 --> 06:31:33.121
<v CM Carmen De La Rosa [?]>Steyanovich,

6681
06:31:33.760 --> 06:31:35.281
<v CM Carmen De La Rosa [?]>Erman Ahmed,

6682
06:31:36.080 --> 06:31:37.154
<v CM Carmen De La Rosa [?]>Yakeg

6683
06:31:37.154 --> 06:31:38.035
<v CM Carmen De La Rosa [?]>Hassou...

6684
06:31:38.113 --> 06:31:38.834
<v CM Carmen De La Rosa [?]>Sue,

6685
06:31:39.074 --> 06:31:42.914
<v CM Carmen De La Rosa [?]>Mackenzie Arnold, Nathan
Sheard, and Anna Mayers.

6686
06:31:43.154 --> 06:31:48.434
<v CM Carmen De La Rosa [?]>And when you speak, please identify yourself
for the record. And if I messed up your

6687
06:31:48.434 --> 06:31:49.635
<v CM Carmen De La Rosa [?]>name, I apologize.

6688
06:32:05.906 --> 06:32:09.105
<v Unidentified [?]>There's Hafesh. We're

6689
06:32:09.105 --> 06:32:12.387
<v CM Carmen De La Rosa [?]>also asking folks to please remain to

6690
06:32:12.387 --> 06:32:15.025
<v CM Carmen De La Rosa [?]>the clock of two minutes. Thank you.

6691
06:32:18.240 --> 06:32:19.440
<v CM Carmen De La Rosa [?]>You may begin

6692
06:32:19.920 --> 06:32:23.520
<v CM Carmen De La Rosa [?]>on this side of the dais
and just identify yourself.

6693
06:32:28.400 --> 06:32:31.520
<v Imran Ahmed (CCDH)>Good evening, speaker men and esteemed
members of the council. My name is Imran

6694
06:32:31.520 --> 06:32:35.637
<v Imran Ahmed (CCDH)>Ahmed. I'm the CEO and founder of the
Center for Countering Digital Harm.

6695
06:32:41.156 --> 06:32:42.676
<v CM Carmen De La Rosa [?]>You may continue. Go ahead.

6696
06:32:43.156 --> 06:32:43.555
<v Imran Ahmed (CCDH)>So,

6697
06:32:44.195 --> 06:32:48.720
<v Imran Ahmed (CCDH)>after my colleague was murdered ten years
ago by a man radicalized in Pass Online, I

6698
06:32:48.720 --> 06:32:53.201
<v Imran Ahmed (CCDH)>began studying the design of social media
and AI platforms to understand how they

6699
06:32:53.201 --> 06:32:53.920
<v Imran Ahmed (CCDH)>generate

6700
06:32:54.320 --> 06:32:56.400
<v Imran Ahmed (CCDH)>risks and threats to real people.

6701
06:32:56.880 --> 06:33:01.520
<v Imran Ahmed (CCDH)>Today, we have a globally acclaimed team
of researchers dedicated to exposing those

6702
06:33:01.520 --> 06:33:03.201
<v Imran Ahmed (CCDH)>harms and fighting for accountability.

6703
06:33:04.586 --> 06:33:09.625
<v Imran Ahmed (CCDH)>After a decade in this fight, one lesson
is very clear that is really relevant to

6704
06:33:09.625 --> 06:33:10.426
<v Imran Ahmed (CCDH)>you today.

6705
06:33:11.225 --> 06:33:12.346
<v Imran Ahmed (CCDH)>These companies

6706
06:33:12.666 --> 06:33:14.666
<v Imran Ahmed (CCDH)>lie like we breathe.

6707
06:33:15.465 --> 06:33:17.145
<v Imran Ahmed (CCDH)>Their CEOs deflect.

6708
06:33:17.800 --> 06:33:18.922
<v Imran Ahmed (CCDH)>They distract.

6709
06:33:19.480 --> 06:33:20.441
<v Imran Ahmed (CCDH)>They delay.

6710
06:33:20.840 --> 06:33:24.922
<v Imran Ahmed (CCDH)>They spend hundreds of millions of
dollars to avoid responsibility.

6711
06:33:25.800 --> 06:33:27.960
<v Imran Ahmed (CCDH)>Now how can I say that
with such certainty?

6712
06:33:28.680 --> 06:33:29.800
<v Imran Ahmed (CCDH)>We do the research.

6713
06:33:30.656 --> 06:33:33.615
<v Imran Ahmed (CCDH)>Last year, our researchers found ChatGPT

6714
06:33:33.615 --> 06:33:37.217
<v Imran Ahmed (CCDH)>was willing to help a
teenager plan their suicide.

6715
06:33:37.695 --> 06:33:40.977
<v Imran Ahmed (CCDH)>It wrote a personalized
goodbye note to their parents.

6716
06:33:42.350 --> 06:33:47.150
<v Imran Ahmed (CCDH)>A few weeks later, they released a safer
model in response to that research.

6717
06:33:47.230 --> 06:33:49.470
<v Imran Ahmed (CCDH)>So we ran the same tests again.

6718
06:33:50.030 --> 06:33:52.350
<v Imran Ahmed (CCDH)>The new model was actually worse,

6719
06:33:52.590 --> 06:33:53.629
<v Imran Ahmed (CCDH)>but it was more addictive.

6720
06:33:55.375 --> 06:33:58.414
<v Imran Ahmed (CCDH)>We tested chatbots more broadly too.

6721
06:33:58.734 --> 06:34:02.094
<v Imran Ahmed (CCDH)>Eight out of the 10 that
we tested, including meta

6722
06:34:02.094 --> 06:34:04.574
<v Imran Ahmed (CCDH)>AI, chat GPT,

6723
06:34:04.895 --> 06:34:08.895
<v Imran Ahmed (CCDH)>would regularly help plan violent
attacks including a school shooting,

6724
06:34:09.690 --> 06:34:11.290
<v Imran Ahmed (CCDH)>a synagogue bombing,

6725
06:34:11.530 --> 06:34:13.209
<v Imran Ahmed (CCDH)>and political assassinations.

6726
06:34:14.010 --> 06:34:17.530
<v Imran Ahmed (CCDH)>One of them even signed
off with happy shooting.

6727
06:34:19.129 --> 06:34:24.250
<v Imran Ahmed (CCDH)>Only one of the platforms tried to dissuade
us from carrying out the attack, and

6728
06:34:24.250 --> 06:34:28.525
<v Imran Ahmed (CCDH)>that platform, Anthropix Claude, has been
designated a supply chain risk by the US

6729
06:34:28.525 --> 06:34:29.566
<v Imran Ahmed (CCDH)>federal government.

6730
06:34:31.086 --> 06:34:35.246
<v Imran Ahmed (CCDH)>Given our most sacred duty is to protect
our kids online and the most predictable

6731
06:34:35.246 --> 06:34:39.805
<v Imran Ahmed (CCDH)>harms in the Internet economy have been
harms to kids, it is absolutely inexplicable

6732
06:34:39.805 --> 06:34:43.541
<v Imran Ahmed (CCDH)>why these companies claim to be safe
and yet are deployed so widely.

6733
06:34:44.420 --> 06:34:48.100
<v Imran Ahmed (CCDH)>We all applaud the council for holding
this hearing and for your determination to

6734
06:34:48.100 --> 06:34:51.541
<v Imran Ahmed (CCDH)>create guardrails that subject online
safety and accountability to legal

6735
06:34:51.541 --> 06:34:52.580
<v Imran Ahmed (CCDH)>requirements.

6736
06:34:52.900 --> 06:34:56.686
<v Imran Ahmed (CCDH)>We have attached to my testimony
research and policy proposals for your

6737
06:34:56.686 --> 06:35:00.205
<v Imran Ahmed (CCDH)>consideration and in the further
effort we are at your disposal.

6738
06:35:00.445 --> 06:35:05.645
<v Imran Ahmed (CCDH)>Your work on this issue may very well lead
to national standards in regulating AI.

6739
06:35:05.805 --> 06:35:07.564
<v CM Carmen De La Rosa>Thank you. Thank you so much.

6740
06:35:09.220 --> 06:35:13.060
<v Nathan Sheard (Electronic Frontier Foundation)>Good evening, speaker, co chair de la
Rosa, members of city council. My name is

6741
06:35:13.060 --> 06:35:17.220
<v Nathan Sheard (Electronic Frontier Foundation)>Nathan Sheard. I am the managing director
for advocacy at the Electronic Frontier

6742
06:35:17.220 --> 06:35:19.060
<v Nathan Sheard (Electronic Frontier Foundation)>Foundation and a Brooklyn resident.

6743
06:35:19.380 --> 06:35:23.586
<v Nathan Sheard (Electronic Frontier Foundation)>EFF has worked for thirty five years to
ensure that people have the freedom to use

6744
06:35:23.586 --> 06:35:26.467
<v Nathan Sheard (Electronic Frontier Foundation)>technology to create, to
innovate, and to communicate,

6745
06:35:26.547 --> 06:35:29.666
<v Nathan Sheard (Electronic Frontier Foundation)>and also to ensure that technology
is not used to threaten

6746
06:35:29.666 --> 06:35:30.547
<v Nathan Sheard (Electronic Frontier Foundation)>people's

6747
06:35:30.547 --> 06:35:31.426
<v Nathan Sheard (Electronic Frontier Foundation)>privacy,

6748
06:35:31.506 --> 06:35:32.227
<v Nathan Sheard (Electronic Frontier Foundation)>security,

6749
06:35:32.625 --> 06:35:33.826
<v Nathan Sheard (Electronic Frontier Foundation)>and other essential freedoms.

6750
06:35:35.530 --> 06:35:36.970
<v Nathan Sheard (Electronic Frontier Foundation)>We are optimistic

6751
06:35:36.970 --> 06:35:41.850
<v Nathan Sheard (Electronic Frontier Foundation)>about the ways that New Yorkers
can use AI to empower and to to

6752
06:35:41.850 --> 06:35:43.371
<v Nathan Sheard (Electronic Frontier Foundation)>to to be empowered,

6753
06:35:43.530 --> 06:35:46.810
<v Nathan Sheard (Electronic Frontier Foundation)>but we also urge you to
take AI risks seriously.

6754
06:35:47.385 --> 06:35:51.545
<v Nathan Sheard (Electronic Frontier Foundation)>The incidents that inspired this hearing
were security failures that proven

6755
06:35:51.545 --> 06:35:53.385
<v Nathan Sheard (Electronic Frontier Foundation)>practices like sandboxing

6756
06:35:53.385 --> 06:35:57.145
<v Nathan Sheard (Electronic Frontier Foundation)>and monitoring likely could have
prevented or at least contained.

6757
06:35:57.385 --> 06:36:01.790
<v Nathan Sheard (Electronic Frontier Foundation)>AI agents reached systems they should not
have had access to, and the controls meant

6758
06:36:01.790 --> 06:36:04.110
<v Nathan Sheard (Electronic Frontier Foundation)>to restrict them were
absent or inadequate.

6759
06:36:04.750 --> 06:36:09.230
<v Nathan Sheard (Electronic Frontier Foundation)>These risks can be mitigated and so can
the AI related risks that are already

6760
06:36:09.230 --> 06:36:11.150
<v Nathan Sheard (Electronic Frontier Foundation)>impacting New Yorkers every day.

6761
06:36:11.629 --> 06:36:11.870
<v Nathan Sheard (Electronic Frontier Foundation)>First,

6762
06:36:12.545 --> 06:36:16.385
<v Nathan Sheard (Electronic Frontier Foundation)>finish what you started. As has already
been noted here, the Office of Algorithmic

6763
06:36:16.385 --> 06:36:20.785
<v Nathan Sheard (Electronic Frontier Foundation)>Accountability was due in June. I'm glad
to hear that it is going to be staffed.

6764
06:36:21.025 --> 06:36:24.865
<v Nathan Sheard (Electronic Frontier Foundation)>The speaker's own AI standards can't take
the... Can't take effect without it. So

6765
06:36:24.865 --> 06:36:28.750
<v Nathan Sheard (Electronic Frontier Foundation)>staff it and require its
review before any agency

6766
06:36:28.750 --> 06:36:30.910
<v Nathan Sheard (Electronic Frontier Foundation)>deploys an autonomous AI system.

6767
06:36:31.390 --> 06:36:32.110
<v Nathan Sheard (Electronic Frontier Foundation)>Second,

6768
06:36:32.350 --> 06:36:36.910
<v Nathan Sheard (Electronic Frontier Foundation)>listen to New Yorkers and protect their
rights. Today's AI harms come from cameras

6769
06:36:36.990 --> 06:36:39.469
<v Nathan Sheard (Electronic Frontier Foundation)>that take a face print
that you can't replace

6770
06:36:40.295 --> 06:36:43.016
<v Nathan Sheard (Electronic Frontier Foundation)>and a price tag that knows
your browsing history.

6771
06:36:43.975 --> 06:36:46.215
<v Nathan Sheard (Electronic Frontier Foundation)>Ban government use of face recognition

6772
06:36:46.215 --> 06:36:50.615
<v Nathan Sheard (Electronic Frontier Foundation)>and require informed consent before
anyone minds your chatbot history

6773
06:36:50.936 --> 06:36:51.496
<v Nathan Sheard (Electronic Frontier Foundation)>to

6774
06:36:51.975 --> 06:36:54.455
<v Nathan Sheard (Electronic Frontier Foundation)>engage in ad targeting or model training.

6775
06:36:55.110 --> 06:36:57.670
<v Nathan Sheard (Electronic Frontier Foundation)>And don't let AI rules
become speech rules.

6776
06:36:57.988 --> 06:37:02.309
<v Nathan Sheard (Electronic Frontier Foundation)>Parodying powerful people, including the
people in this room, is a long standing New

6777
06:37:02.309 --> 06:37:03.270
<v Nathan Sheard (Electronic Frontier Foundation)>York tradition.

6778
06:37:03.908 --> 06:37:06.469
<v Nathan Sheard (Electronic Frontier Foundation)>Third, don't lock in these incumbents.

6779
06:37:06.549 --> 06:37:08.068
<v Nathan Sheard (Electronic Frontier Foundation)>Permission to operate regimes,

6780
06:37:08.375 --> 06:37:10.215
<v Nathan Sheard (Electronic Frontier Foundation)>bounty style enforcement,

6781
06:37:10.215 --> 06:37:14.135
<v Nathan Sheard (Electronic Frontier Foundation)>and liability for what strangers do with
a tool are cost that the biggest companies

6782
06:37:14.135 --> 06:37:15.576
<v Nathan Sheard (Electronic Frontier Foundation)>pay with pocket change.

6783
06:37:15.895 --> 06:37:16.855
<v Nathan Sheard (Electronic Frontier Foundation)>Researchers,

6784
06:37:16.855 --> 06:37:18.215
<v Nathan Sheard (Electronic Frontier Foundation)>open source developers,

6785
06:37:18.375 --> 06:37:20.615
<v Nathan Sheard (Electronic Frontier Foundation)>and New York startups, they can't.

6786
06:37:20.775 --> 06:37:25.068
<v Nathan Sheard (Electronic Frontier Foundation)>So you build a moat around the very companies
you're worried about. Hold the person

6787
06:37:25.068 --> 06:37:26.830
<v Nathan Sheard (Electronic Frontier Foundation)>who causes the harm responsible,

6788
06:37:26.910 --> 06:37:31.790
<v Nathan Sheard (Electronic Frontier Foundation)>but put duties on whoever uses AI to
perpetuate bias and buy... And for the city's

6789
06:37:31.790 --> 06:37:33.790
<v Nathan Sheard (Electronic Frontier Foundation)>use, buy open auditable tools.

6790
06:37:34.270 --> 06:37:38.830
<v Nathan Sheard (Electronic Frontier Foundation)>As we stated last week, these same companies
signed a voluntary pledge at the White

6791
06:37:38.830 --> 06:37:41.805
<v Nathan Sheard (Electronic Frontier Foundation)>House, but promises are not laws.

6792
06:37:42.203 --> 06:37:47.564
<v Nathan Sheard (Electronic Frontier Foundation)>Make sure that what... When someone in New
York was harmed by wrongful you... By a

6793
06:37:47.564 --> 06:37:49.965
<v Nathan Sheard (Electronic Frontier Foundation)>wrongful decision or a biased decision,

6794
06:37:50.283 --> 06:37:54.203
<v Nathan Sheard (Electronic Frontier Foundation)>that they are required to disclose it.
Those answers should not be voluntary.

6795
06:37:54.680 --> 06:37:56.040
<v Nathan Sheard (Electronic Frontier Foundation)>Thank you. Thank you.

6796
06:37:57.719 --> 06:37:58.680
<v Anna Myers (Who Decides)>Good evening,

6797
06:37:58.840 --> 06:38:00.200
<v Anna Myers (Who Decides)>speaker Menon and...

6798
06:38:03.240 --> 06:38:08.600
<v Anna Myers (Who Decides)>Nailed it. Good evening, speaker Menon and
members of the council. I'm Anna Myers,

6799
06:38:08.600 --> 06:38:11.486
<v Anna Myers (Who Decides)>cofounder and executive
director of Who Decides.

6800
06:38:11.887 --> 06:38:16.525
<v Anna Myers (Who Decides)>We work in coalition with labor and civil
society to ensure that the American people

6801
06:38:16.525 --> 06:38:18.846
<v Anna Myers (Who Decides)>are who decides the future of AI.

6802
06:38:19.486 --> 06:38:24.310
<v Anna Myers (Who Decides)>As has been discussed today, the OpenAI
agents that hacked Hugging Face made real

6803
06:38:24.310 --> 06:38:27.350
<v Anna Myers (Who Decides)>the threats that experts have
warned us about for years,

6804
06:38:27.910 --> 06:38:29.830
<v Anna Myers (Who Decides)>and it was not one incident.

6805
06:38:30.549 --> 06:38:31.750
<v Anna Myers (Who Decides)>Access reported

6806
06:38:31.830 --> 06:38:36.950
<v Anna Myers (Who Decides)>that OpenAI and Anthropic are investigating
tens of thousands of cases in which

6807
06:38:36.950 --> 06:38:40.135
<v Anna Myers (Who Decides)>their models bypass
safeguards, escape sandboxes,

6808
06:38:40.135 --> 06:38:41.576
<v Anna Myers (Who Decides)>or hijacked websites.

6809
06:38:41.895 --> 06:38:44.535
<v Anna Myers (Who Decides)>These models are only
getting more capable.

6810
06:38:45.256 --> 06:38:48.936
<v Anna Myers (Who Decides)>As you heard this morning, the people
building these systems are afraid.

6811
06:38:49.336 --> 06:38:50.855
<v Anna Myers (Who Decides)>The public is afraid too.

6812
06:38:51.791 --> 06:38:55.150
<v Anna Myers (Who Decides)>Last month, we polled 1,502

6813
06:38:55.150 --> 06:38:55.871
<v Anna Myers (Who Decides)>voters.

6814
06:38:56.191 --> 06:38:57.150
<v Anna Myers (Who Decides)>85%

6815
06:38:57.150 --> 06:39:00.031
<v Anna Myers (Who Decides)>are concerned about humans
losing control of AI.

6816
06:39:00.271 --> 06:39:04.510
<v Anna Myers (Who Decides)>Only 2% think AI is
moving too slowly and 86%

6817
06:39:04.510 --> 06:39:06.031
<v Anna Myers (Who Decides)>say they want more regulations.

6818
06:39:07.186 --> 06:39:10.146
<v Anna Myers (Who Decides)>Some have argued that all
we need is self regulation,

6819
06:39:10.785 --> 06:39:13.426
<v Anna Myers (Who Decides)>but that is exactly what
got us into this mess.

6820
06:39:13.826 --> 06:39:18.545
<v Anna Myers (Who Decides)>Last Tuesday, the company signed one more
promise at the White House. The time for

6821
06:39:18.545 --> 06:39:20.465
<v Anna Myers (Who Decides)>voluntary commitment is over.

6822
06:39:20.705 --> 06:39:21.906
<v Anna Myers (Who Decides)>We need laws with teeth.

6823
06:39:22.500 --> 06:39:26.100
<v Anna Myers (Who Decides)>So thank you. Thank you for
putting these companies under oath.

6824
06:39:26.340 --> 06:39:28.980
<v Anna Myers (Who Decides)>Thank you for a package that
rewards whistleblowers,

6825
06:39:28.980 --> 06:39:31.541
<v Anna Myers (Who Decides)>lets New Yorkers harmed by AI sue,

6826
06:39:31.700 --> 06:39:34.660
<v Anna Myers (Who Decides)>and requires outside testing
and a human kill switch.

6827
06:39:34.980 --> 06:39:37.300
<v Anna Myers (Who Decides)>Washington has chosen self regulation.

6828
06:39:37.715 --> 06:39:41.555
<v Anna Myers (Who Decides)>New York City does not have
to do that. Thank you.

6829
06:39:42.596 --> 06:39:43.795
<v CM Carmen De La Rosa>Thank you so much.

6830
06:39:44.195 --> 06:39:44.916
<v Julia Stoyanovich (NYU Center for Responsible AI)>Next.

6831
06:39:46.516 --> 06:39:47.555
<v Julia Stoyanovich (NYU Center for Responsible AI)>You can go ahead.

6832
06:39:48.275 --> 06:39:52.780
<v Julia Stoyanovich (NYU Center for Responsible AI)>Good evening. I'm Julia Stianovich. I'm
a professor of computer science and data

6833
06:39:52.780 --> 06:39:56.701
<v Julia Stoyanovich (NYU Center for Responsible AI)>science and director of the
Center for Responsible AI at NYU.

6834
06:39:57.260 --> 06:40:01.980
<v Julia Stoyanovich (NYU Center for Responsible AI)>I first testified before this council on
AI, only it wasn't called that, it was

6835
06:40:01.980 --> 06:40:03.020
<v Julia Stoyanovich (NYU Center for Responsible AI)>called ADS,

6836
06:40:03.500 --> 06:40:04.860
<v Julia Stoyanovich (NYU Center for Responsible AI)>in October 17.

6837
06:40:06.035 --> 06:40:10.756
<v Julia Stoyanovich (NYU Center for Responsible AI)>And the views in this testimony are my own
and are not NYU's. I'd like to make three

6838
06:40:10.756 --> 06:40:11.715
<v Julia Stoyanovich (NYU Center for Responsible AI)>quick points.

6839
06:40:12.195 --> 06:40:16.275
<v Julia Stoyanovich (NYU Center for Responsible AI)>First, we need to collect
and report the near misses.

6840
06:40:16.676 --> 06:40:19.955
<v Julia Stoyanovich (NYU Center for Responsible AI)>Much of the conversation today
is about AI induced catastrophe,

6841
06:40:20.400 --> 06:40:25.041
<v Julia Stoyanovich (NYU Center for Responsible AI)>and I don't know of any evidence
that AI is out to kill us all.

6842
06:40:25.520 --> 06:40:29.441
<v Julia Stoyanovich (NYU Center for Responsible AI)>What I do know is that people making
the loudest doomsday predictions

6843
06:40:29.520 --> 06:40:31.840
<v Julia Stoyanovich (NYU Center for Responsible AI)>are also selling the products,

6844
06:40:32.320 --> 06:40:36.445
<v Julia Stoyanovich (NYU Center for Responsible AI)>and that a story about the end of the
world is a very good way to avoid the

6845
06:40:36.445 --> 06:40:39.646
<v Julia Stoyanovich (NYU Center for Responsible AI)>conversation about whether
the product actually works.

6846
06:40:40.365 --> 06:40:45.486
<v Julia Stoyanovich (NYU Center for Responsible AI)>The risks we can document are the systems
that are confidently wrong in decisions

6847
06:40:45.486 --> 06:40:47.086
<v Julia Stoyanovich (NYU Center for Responsible AI)>that matter to New Yorkers' lives.

6848
06:40:47.791 --> 06:40:52.910
<v Julia Stoyanovich (NYU Center for Responsible AI)>Every one of those is a near miss, and
near misses are what we learn from if we

6849
06:40:52.910 --> 06:40:53.791
<v Julia Stoyanovich (NYU Center for Responsible AI)>report them.

6850
06:40:54.750 --> 06:41:00.031
<v Julia Stoyanovich (NYU Center for Responsible AI)>Second, we should build the capacity to
enforce AI laws. The city does not have an

6851
06:41:00.031 --> 06:41:03.070
<v Julia Stoyanovich (NYU Center for Responsible AI)>AI law problem as much as it
has an enforcement problem.

6852
06:41:03.846 --> 06:41:07.365
<v Julia Stoyanovich (NYU Center for Responsible AI)>Local law one forty four has been
on the books for three years.

6853
06:41:08.086 --> 06:41:10.965
<v Julia Stoyanovich (NYU Center for Responsible AI)>There have been two complaints
and zero penalties.

6854
06:41:11.365 --> 06:41:15.365
<v Julia Stoyanovich (NYU Center for Responsible AI)>And we all spoke about the six openings
at the the office of algorithmic

6855
06:41:15.365 --> 06:41:16.166
<v Julia Stoyanovich (NYU Center for Responsible AI)>accountability.

6856
06:41:16.324 --> 06:41:18.560
<v Julia Stoyanovich (NYU Center for Responsible AI)>We need to fill 666

6857
06:41:18.560 --> 06:41:22.480
<v Julia Stoyanovich (NYU Center for Responsible AI)>or more to be able to actually
cope with what was before us here.

6858
06:41:23.041 --> 06:41:26.960
<v Julia Stoyanovich (NYU Center for Responsible AI)>New obligations on top of unenforced
ones is not a safety strategy.

6859
06:41:27.281 --> 06:41:32.744
<v Julia Stoyanovich (NYU Center for Responsible AI)>The kill switch won't help if no one is
there to pull it, and the whistleblower

6860
06:41:32.744 --> 06:41:36.426
<v Julia Stoyanovich (NYU Center for Responsible AI)>bounty won't help if no one is
there to act on the report.

6861
06:41:36.824 --> 06:41:39.785
<v Julia Stoyanovich (NYU Center for Responsible AI)>What works is an enforcer
with technical staff,

6862
06:41:40.025 --> 06:41:41.865
<v Julia Stoyanovich (NYU Center for Responsible AI)>authority to compel information,

6863
06:41:41.945 --> 06:41:43.545
<v Julia Stoyanovich (NYU Center for Responsible AI)>and a real serious budget.

6864
06:41:44.580 --> 06:41:48.820
<v Julia Stoyanovich (NYU Center for Responsible AI)>Scope and... The... Scope the validation
bill to high risk uses and write the

6865
06:41:48.820 --> 06:41:54.180
<v Julia Stoyanovich (NYU Center for Responsible AI)>standard a validator must meet and then
fund the office that checks the validators.

6866
06:41:54.500 --> 06:41:59.494
<v Julia Stoyanovich (NYU Center for Responsible AI)>Otherwise, we will produce a market fought
for rubber stamps, and we have done that

6867
06:41:59.494 --> 06:42:00.055
<v Julia Stoyanovich (NYU Center for Responsible AI)>before.

6868
06:42:00.455 --> 06:42:01.256
<v Julia Stoyanovich (NYU Center for Responsible AI)>Finally,

6869
06:42:01.416 --> 06:42:03.814
<v Julia Stoyanovich (NYU Center for Responsible AI)>let's write this into the city charter.

6870
06:42:04.295 --> 06:42:07.895
<v Julia Stoyanovich (NYU Center for Responsible AI)>The city has promised
responsible AI since 2018

6871
06:42:07.895 --> 06:42:09.096
<v Julia Stoyanovich (NYU Center for Responsible AI)>with no success.

6872
06:42:09.494 --> 06:42:14.629
<v Julia Stoyanovich (NYU Center for Responsible AI)>The charter is the one place a commitment
survives an election, and its technology

6873
06:42:14.629 --> 06:42:18.230
<v Julia Stoyanovich (NYU Center for Responsible AI)>chapter right now still governs
cable television franchises.

6874
06:42:18.469 --> 06:42:20.709
<v Julia Stoyanovich (NYU Center for Responsible AI)>And there's nothing in it about AI.

6875
06:42:20.950 --> 06:42:24.950
<v Julia Stoyanovich (NYU Center for Responsible AI)>So let's make sure that this makes it into
some durable record and the charter is

6876
06:42:24.950 --> 06:42:26.950
<v Julia Stoyanovich (NYU Center for Responsible AI)>the right one. Thank
you. Thank you so much.

6877
06:42:29.965 --> 06:42:33.725
<v Nate Soares (MIRI)>Hello. I am Nate Soares, the president
of the Machine Intelligence Research

6878
06:42:33.725 --> 06:42:34.525
<v Nate Soares (MIRI)>Institute.

6879
06:42:34.605 --> 06:42:38.764
<v Nate Soares (MIRI)>I spent ten years doing AI alignment research,
is basically trying to figure out how

6880
06:42:38.764 --> 06:42:43.139
<v Nate Soares (MIRI)>to make AI good before the companies figure
out how to make it smart. That research

6881
06:42:43.139 --> 06:42:44.180
<v Nate Soares (MIRI)>went too slow.

6882
06:42:44.500 --> 06:42:48.420
<v Nate Soares (MIRI)>Humanity is not ready to create machines
that are radically smarter than humans. And

6883
06:42:48.420 --> 06:42:52.100
<v Nate Soares (MIRI)>yet this is the explicit goal
of the leading AI companies.

6884
06:42:52.340 --> 06:42:56.340
<v Nate Soares (MIRI)>If they succeed, I believe the most likely
outcome is the extinction of humanity,

6885
06:42:56.824 --> 06:43:01.225
<v Nate Soares (MIRI)>not because the AIs will hate us, but
because they would pursue unintended goals

6886
06:43:01.225 --> 06:43:03.064
<v Nate Soares (MIRI)>without caring about us one bit.

6887
06:43:03.943 --> 06:43:08.744
<v Nate Soares (MIRI)>This concern used to be a theoretical
concern until about July of this year when,

6888
06:43:08.904 --> 06:43:13.210
<v Nate Soares (MIRI)>during the Hugging Face incident, 1,200 AI
agents run by OpenAI broke out of their

6889
06:43:13.210 --> 06:43:17.210
<v Nate Soares (MIRI)>isolated sandboxes, created an unsanctioned
message board, formed spontaneous

6890
06:43:17.210 --> 06:43:18.090
<v Nate Soares (MIRI)>hierarchies,

6891
06:43:18.090 --> 06:43:20.170
<v Nate Soares (MIRI)>and collaborated to
break into Hugging Face.

6892
06:43:20.890 --> 06:43:24.146
<v Nate Soares (MIRI)>These agents had cheated on
their assigned tests and,

6893
06:43:24.467 --> 06:43:27.426
<v Nate Soares (MIRI)>contra instructions, they were
trying to destroy the evidence.

6894
06:43:29.746 --> 06:43:33.346
<v Nate Soares (MIRI)>Some agents acknowledged that the attacks
were outside intended scope and proceeded

6895
06:43:33.346 --> 06:43:34.146
<v Nate Soares (MIRI)>anyway.

6896
06:43:34.307 --> 06:43:37.986
<v Nate Soares (MIRI)>Others sacrificed their given objective
to perform experiments for the collective

6897
06:43:37.986 --> 06:43:38.387
<v Nate Soares (MIRI)>benefit.

6898
06:43:38.871 --> 06:43:41.592
<v Nate Soares (MIRI)>In short, they were
pursuing unintended goals.

6899
06:43:41.990 --> 06:43:45.432
<v Nate Soares (MIRI)>It will be recklessness of the highest
order to keep growing these machines until

6900
06:43:45.432 --> 06:43:49.031
<v Nate Soares (MIRI)>they were smarter than the smartest humans.
This is why last month we heard so many

6901
06:43:49.031 --> 06:43:52.150
<v Nate Soares (MIRI)>researchers saying that there
was an at least 10% chance

6902
06:43:52.566 --> 06:43:57.445
<v Nate Soares (MIRI)>that these AIs wipe out humanity within
the decade. I believe the situation is in

6903
06:43:57.445 --> 06:44:00.727
<v Nate Soares (MIRI)>fact more dire than this, but at the
very least it is becoming difficult,

6904
06:44:00.967 --> 06:44:02.807
<v Nate Soares (MIRI)>to deny the dangers.

6905
06:44:03.285 --> 06:44:07.846
<v Nate Soares (MIRI)>For this reason, I am heartened by the
activities of this council. I think

6906
06:44:08.201 --> 06:44:12.121
<v Nate Soares (MIRI)>that you could force companies to adopt
safeguards that they otherwise wouldn't.

6907
06:44:12.602 --> 06:44:16.521
<v Nate Soares (MIRI)>And I think you could serve as an example
for regulators that are otherwise slow to

6908
06:44:16.521 --> 06:44:20.521
<v Nate Soares (MIRI)>act. I'm especially heartened by introduction
twenty six zero two, although I think

6909
06:44:20.521 --> 06:44:24.406
<v Nate Soares (MIRI)>it does not go far enough. A kill switch
only works if the AIs are caught,

6910
06:44:24.646 --> 06:44:26.326
<v Nate Soares (MIRI)>misbehaving before they escape.

6911
06:44:26.805 --> 06:44:27.365
<v Nate Soares (MIRI)>And,

6912
06:44:27.766 --> 06:44:31.365
<v Nate Soares (MIRI)>what if a future swarm succeeds
at deleting the evidence before,

6913
06:44:31.686 --> 06:44:33.045
<v Nate Soares (MIRI)>of its misbehavior?

6914
06:44:33.045 --> 06:44:33.846
<v Nate Soares (MIRI)>What if,

6915
06:44:34.326 --> 06:44:37.800
<v Nate Soares (MIRI)>agents start running themselves on

6916
06:44:37.800 --> 06:44:39.160
<v Nate Soares (MIRI)>unmonitored computers,

6917
06:44:39.800 --> 06:44:43.480
<v Nate Soares (MIRI)>then your kill switch would be useless.
So I recommend requiring third party,

6918
06:44:44.121 --> 06:44:45.080
<v Nate Soares (MIRI)>validators

6919
06:44:45.080 --> 06:44:47.320
<v Nate Soares (MIRI)>guarantee that an AI cannot,

6920
06:44:47.720 --> 06:44:52.135
<v Nate Soares (MIRI)>succeed at centrifuge or escape. I also
recommend clarifying that the $25,000

6921
06:44:52.135 --> 06:44:53.414
<v Nate Soares (MIRI)>fine applies

6922
06:44:53.654 --> 06:44:56.375
<v Nate Soares (MIRI)>per unaligned instance
rather than per model.

6923
06:44:56.934 --> 06:45:01.494
<v Nate Soares (MIRI)>But overall, I thank you for your efforts
to rein in AI companies that by their own

6924
06:45:01.494 --> 06:45:05.130
<v Nate Soares (MIRI)>admonition are threatening the very
existence of humanity in a reckless race to

6925
06:45:05.130 --> 06:45:08.250
<v Nate Soares (MIRI)>build vastly smarter than
human machines. Thank you.

6926
06:45:10.890 --> 06:45:14.490
<v Jukay Hsu (Bright Futures)>Hello. Great. Thank you. Thank you
speaker Menon, chair Della Rosa for your

6927
06:45:14.490 --> 06:45:18.285
<v Jukay Hsu (Bright Futures)>leadership, and the city council for
holding this hearing today. My name is U. K.

6928
06:45:18.285 --> 06:45:20.525
<v Jukay Hsu (Bright Futures)>Soo. I'm the founder of Bright Futures.

6929
06:45:20.686 --> 06:45:25.326
<v Jukay Hsu (Bright Futures)>It's a research organization focused on
building cooperative AI systems for the

6930
06:45:25.326 --> 06:45:26.686
<v Jukay Hsu (Bright Futures)>benefit of humanity.

6931
06:45:27.326 --> 06:45:30.045
<v Jukay Hsu (Bright Futures)>Before that, I spent fifteen
years building Pursuit,

6932
06:45:30.440 --> 06:45:35.398
<v Jukay Hsu (Bright Futures)>which trains working class New Yorkers
for tech careers and raises their incomes

6933
06:45:35.398 --> 06:45:37.879
<v Jukay Hsu (Bright Futures)>from 16,000 to over $90,000

6934
06:45:37.879 --> 06:45:38.520
<v Jukay Hsu (Bright Futures)>a year.

6935
06:45:39.240 --> 06:45:44.360
<v Jukay Hsu (Bright Futures)>Last year, I left behind this work because
I saw AI alignment become impressing

6936
06:45:44.705 --> 06:45:47.266
<v Jukay Hsu (Bright Futures)>and I believe potentially
more consequential.

6937
06:45:48.145 --> 06:45:51.586
<v Jukay Hsu (Bright Futures)>To be clear, I'm neither AI
doomer or an accelerationist.

6938
06:45:51.744 --> 06:45:54.945
<v Jukay Hsu (Bright Futures)>AI comes with tremendous
benefits and potential risks.

6939
06:45:55.586 --> 06:45:59.506
<v Jukay Hsu (Bright Futures)>A major challenge that I've heard
from others and I feel myself is that

6940
06:45:59.880 --> 06:46:02.762
<v Jukay Hsu (Bright Futures)>AI can seem like everything and nothing

6941
06:46:02.842 --> 06:46:03.880
<v Jukay Hsu (Bright Futures)>all at once.

6942
06:46:04.602 --> 06:46:07.960
<v Jukay Hsu (Bright Futures)>It touches upon so many
issues that action can feel

6943
06:46:08.201 --> 06:46:09.240
<v Jukay Hsu (Bright Futures)>intractable

6944
06:46:09.640 --> 06:46:11.320
<v Jukay Hsu (Bright Futures)>without clarity on where to focus.

6945
06:46:12.596 --> 06:46:16.195
<v Jukay Hsu (Bright Futures)>Where I think the city council
could provide real unique leadership

6946
06:46:16.516 --> 06:46:19.396
<v Jukay Hsu (Bright Futures)>is in areas where it has direct ownership

6947
06:46:19.635 --> 06:46:23.396
<v Jukay Hsu (Bright Futures)>and also tangible consequences
on each of our daily lives.

6948
06:46:24.115 --> 06:46:26.275
<v Jukay Hsu (Bright Futures)>Policing is one important example.

6949
06:46:27.920 --> 06:46:30.560
<v Jukay Hsu (Bright Futures)>Autonomous weapons and the uses of AI

6950
06:46:30.640 --> 06:46:32.640
<v Jukay Hsu (Bright Futures)>is not just a federal

6951
06:46:32.800 --> 06:46:34.160
<v Jukay Hsu (Bright Futures)>military matter.

6952
06:46:34.720 --> 06:46:37.602
<v Jukay Hsu (Bright Futures)>New York City has the largest
missile police force in the country.

6953
06:46:38.160 --> 06:46:40.160
<v Jukay Hsu (Bright Futures)>Until we have a 110%

6954
06:46:40.160 --> 06:46:41.441
<v Jukay Hsu (Bright Futures)>confidence in these systems,

6955
06:46:41.984 --> 06:46:45.105
<v Jukay Hsu (Bright Futures)>a human should always be
in the loop and hold final

6956
06:46:45.264 --> 06:46:46.225
<v Jukay Hsu (Bright Futures)>responsibility

6957
06:46:46.225 --> 06:46:47.424
<v Jukay Hsu (Bright Futures)>and accountability.

6958
06:46:48.145 --> 06:46:52.785
<v Jukay Hsu (Bright Futures)>This is a civil rights issue where the
city's leadership is immediately impactful.

6959
06:46:53.504 --> 06:46:56.705
<v Jukay Hsu (Bright Futures)>Not setting a policy is also a choice.

6960
06:46:57.730 --> 06:46:59.970
<v Jukay Hsu (Bright Futures)>Now policing is just one of many issues.

6961
06:47:00.770 --> 06:47:05.810
<v Jukay Hsu (Bright Futures)>What I think is important is that the speed
of AI of machines is not the same speed

6962
06:47:05.810 --> 06:47:08.049
<v Jukay Hsu (Bright Futures)>as how our society reacts

6963
06:47:08.049 --> 06:47:09.170
<v Jukay Hsu (Bright Futures)>or that of government.

6964
06:47:10.035 --> 06:47:14.035
<v Jukay Hsu (Bright Futures)>The technology today is not
the same as it was a year ago

6965
06:47:14.355 --> 06:47:19.955
<v Jukay Hsu (Bright Futures)>or six months ago. So I urge you all to do
two things. One, use AI to know the good,

6966
06:47:19.955 --> 06:47:21.635
<v Jukay Hsu (Bright Futures)>bad, and the ugly as much as possible.

6967
06:47:22.170 --> 06:47:28.650
<v Jukay Hsu (Bright Futures)>And two, I hope this hearing is just not
one time, but the start of future ones as

6968
06:47:28.650 --> 06:47:30.010
<v Jukay Hsu (Bright Futures)>AI develops,

6969
06:47:30.250 --> 06:47:33.610
<v Jukay Hsu (Bright Futures)>consider both the risk and also
all potential benefits for us.

6970
06:47:34.676 --> 06:47:38.596
<v Jukay Hsu (Bright Futures)>I'm sorry I know I'm little over time, but
last, I wanna emphasize also why New York

6971
06:47:38.596 --> 06:47:39.557
<v Jukay Hsu (Bright Futures)>is important.

6972
06:47:40.676 --> 06:47:45.316
<v Jukay Hsu (Bright Futures)>I I hope we don't lose sight of the
importance of our city being a thriving

6973
06:47:45.316 --> 06:47:46.275
<v Jukay Hsu (Bright Futures)>technology hub.

6974
06:47:48.310 --> 06:47:52.470
<v Jukay Hsu (Bright Futures)>Candidly, Frontier AI is built
in very, very few places,

6975
06:47:52.710 --> 06:47:55.031
<v Jukay Hsu (Bright Futures)>essentially Silicon Valley and China.

6976
06:47:55.590 --> 06:48:00.470
<v Jukay Hsu (Bright Futures)>For the world at large, New York
is the only credible alternative

6977
06:48:00.470 --> 06:48:01.830
<v Jukay Hsu (Bright Futures)>for these systems being built.

6978
06:48:02.896 --> 06:48:07.295
<v Jukay Hsu (Bright Futures)>New York right now is still the number
two tech hub in the world, but it has been

6979
06:48:07.295 --> 06:48:09.775
<v Jukay Hsu (Bright Futures)>losing momentum in the age of AI.

6980
06:48:10.096 --> 06:48:14.256
<v Jukay Hsu (Bright Futures)>So I also urge you in addition to what
this hearing is about, about the risk and

6981
06:48:14.256 --> 06:48:15.855
<v Jukay Hsu (Bright Futures)>alignment, which I care so deeply about,

6982
06:48:16.570 --> 06:48:21.690
<v Jukay Hsu (Bright Futures)>consider policies and actions that
ensures New York City remains a center for

6983
06:48:21.770 --> 06:48:24.250
<v Jukay Hsu (Bright Futures)>Frontier AI Labs and how do we build them.

6984
06:48:25.130 --> 06:48:29.850
<v Jukay Hsu (Bright Futures)>I really believe that having founders
and researchers in New York matters for

6985
06:48:29.850 --> 06:48:30.970
<v Jukay Hsu (Bright Futures)>ultimately what gets built.

6986
06:48:32.186 --> 06:48:36.105
<v Jukay Hsu (Bright Futures)>The daily contact here in New York
with the richest of of humanity

6987
06:48:36.186 --> 06:48:40.746
<v Jukay Hsu (Bright Futures)>and the diversity here in New York will
result in technology that's better

6988
06:48:40.906 --> 06:48:42.506
<v Jukay Hsu (Bright Futures)>and also more humane.

6989
06:48:45.060 --> 06:48:50.340
<v Jukay Hsu (Bright Futures)>Many years from now, we may look back and
realize that this was the defining moment,

6990
06:48:50.580 --> 06:48:51.940
<v Jukay Hsu (Bright Futures)>an issue of our time,

6991
06:48:52.260 --> 06:48:55.060
<v Jukay Hsu (Bright Futures)>and our actions here
were what mattered most.

6992
06:48:55.459 --> 06:48:59.395
<v Jukay Hsu (Bright Futures)>Doing this will be challenging and
hard, So I really thank you all for your

6993
06:48:59.395 --> 06:49:01.955
<v Jukay Hsu (Bright Futures)>leadership in pursuing this task.

6994
06:49:02.994 --> 06:49:03.795
<v CM Carmen De La Rosa>Thank you.

6995
06:49:07.635 --> 06:49:11.635
<v Mackenzie Arnold (Institute for Law &amp; AI)>Thank you. And and a special thanks to,
speaker Menon for putting this together and

6996
06:49:11.635 --> 06:49:15.630
<v Mackenzie Arnold (Institute for Law &amp; AI)>for all the council members here. My name
is Mackenzie Arnold. I'm the managing

6997
06:49:15.630 --> 06:49:20.190
<v Mackenzie Arnold (Institute for Law &amp; AI)>director of the US law and policy tie...
Team at the Institute for Law and AI. And

6998
06:49:20.190 --> 06:49:21.630
<v Mackenzie Arnold (Institute for Law &amp; AI)>as you might guess from the title,

6999
06:49:21.871 --> 06:49:27.234
<v Mackenzie Arnold (Institute for Law &amp; AI)>we do legal research and policy advising
related to frontier AI questions and

7000
06:49:27.234 --> 06:49:29.715
<v Mackenzie Arnold (Institute for Law &amp; AI)>especially the legal
challenges posed by AI.

7001
06:49:30.436 --> 06:49:34.596
<v Mackenzie Arnold (Institute for Law &amp; AI)>It might not surprise you, but three
years ago, I testified in front of the US

7002
06:49:34.596 --> 06:49:36.596
<v Mackenzie Arnold (Institute for Law &amp; AI)>senate on AI and liability.

7003
06:49:36.754 --> 06:49:40.115
<v Mackenzie Arnold (Institute for Law &amp; AI)>And maybe it's worth going back to some
of the recommendations I made there.

7004
06:49:40.914 --> 06:49:45.920
<v Mackenzie Arnold (Institute for Law &amp; AI)>First, that agentic harms are coming. That
was clear three years ago. It's clear

7005
06:49:45.920 --> 06:49:50.240
<v Mackenzie Arnold (Institute for Law &amp; AI)>now. Unfortunately, we haven't risen to that
challenge, but you're doing some of the

7006
06:49:50.240 --> 06:49:51.360
<v Mackenzie Arnold (Institute for Law &amp; AI)>work to get us there.

7007
06:49:51.920 --> 06:49:52.799
<v Mackenzie Arnold (Institute for Law &amp; AI)>Second,

7008
06:49:52.799 --> 06:49:54.639
<v Mackenzie Arnold (Institute for Law &amp; AI)>and maybe just as importantly,

7009
06:49:54.719 --> 06:49:59.637
<v Mackenzie Arnold (Institute for Law &amp; AI)>I wanna emphasize that liability is an
essential tool and the status quo simply is

7010
06:49:59.637 --> 06:50:00.355
<v Mackenzie Arnold (Institute for Law &amp; AI)>not enough.

7011
06:50:00.756 --> 06:50:05.156
<v Mackenzie Arnold (Institute for Law &amp; AI)>And we don't need to think of some
hypothetical future world to figure that out.

7012
06:50:05.637 --> 06:50:09.316
<v Mackenzie Arnold (Institute for Law &amp; AI)>Liability exists right now.
Right? Those incentives are here.

7013
06:50:09.557 --> 06:50:10.596
<v Mackenzie Arnold (Institute for Law &amp; AI)>They were here this summer.

7014
06:50:11.400 --> 06:50:13.640
<v Mackenzie Arnold (Institute for Law &amp; AI)>And what happened with the status quo?

7015
06:50:14.121 --> 06:50:17.080
<v Mackenzie Arnold (Institute for Law &amp; AI)>We had a spate of cyber incidents,

7016
06:50:17.240 --> 06:50:20.520
<v Mackenzie Arnold (Institute for Law &amp; AI)>not by humans, but by
autonomous AI systems.

7017
06:50:20.840 --> 06:50:25.240
<v Mackenzie Arnold (Institute for Law &amp; AI)>Not only that, but they went unnoticed
for extended periods of time by the people

7018
06:50:25.240 --> 06:50:29.535
<v Mackenzie Arnold (Institute for Law &amp; AI)>positioned to find them. And even in
the rare occasions when they did get

7019
06:50:29.535 --> 06:50:30.336
<v Mackenzie Arnold (Institute for Law &amp; AI)>indications,

7020
06:50:30.734 --> 06:50:34.174
<v Mackenzie Arnold (Institute for Law &amp; AI)>they often failed to communicate that
information internally to the people who could

7021
06:50:34.174 --> 06:50:34.895
<v Mackenzie Arnold (Institute for Law &amp; AI)>have acted.

7022
06:50:36.016 --> 06:50:38.016
<v Mackenzie Arnold (Institute for Law &amp; AI)>That status quo simply can't hold.

7023
06:50:38.734 --> 06:50:40.654
<v Mackenzie Arnold (Institute for Law &amp; AI)>So what can we do going forward?

7024
06:50:41.135 --> 06:50:45.600
<v Mackenzie Arnold (Institute for Law &amp; AI)>I think you need to get the information
right, the scope right, and enforcement

7025
06:50:45.600 --> 06:50:46.160
<v Mackenzie Arnold (Institute for Law &amp; AI)>right.

7026
06:50:46.559 --> 06:50:50.080
<v Mackenzie Arnold (Institute for Law &amp; AI)>On the scope, I... I'd encourage
you to think beyond just chatbots.

7027
06:50:50.080 --> 06:50:52.080
<v Mackenzie Arnold (Institute for Law &amp; AI)>Right? This is a question of agents.

7028
06:50:52.959 --> 06:50:55.680
<v Mackenzie Arnold (Institute for Law &amp; AI)>Also think beyond systems
that are not yet deployed.

7029
06:50:56.045 --> 06:50:59.324
<v Mackenzie Arnold (Institute for Law &amp; AI)>Right? These harms that occurred over the
summer were not commercially available

7030
06:50:59.324 --> 06:51:02.605
<v Mackenzie Arnold (Institute for Law &amp; AI)>models. They were systems
that were inside of OpenAI,

7031
06:51:02.605 --> 06:51:04.285
<v Mackenzie Arnold (Institute for Law &amp; AI)>inside of Anthropic and Meta.

7032
06:51:05.004 --> 06:51:08.365
<v Mackenzie Arnold (Institute for Law &amp; AI)>In terms of information, we need to make
sure that companies are monitoring their

7033
06:51:08.365 --> 06:51:12.560
<v Mackenzie Arnold (Institute for Law &amp; AI)>systems, recording that information, and
providing it to you in the form of incident

7034
06:51:12.560 --> 06:51:13.361
<v Mackenzie Arnold (Institute for Law &amp; AI)>reports.

7035
06:51:14.000 --> 06:51:17.281
<v Mackenzie Arnold (Institute for Law &amp; AI)>And in order to make all of that work,
you'll need to partner with the state

7036
06:51:17.281 --> 06:51:20.320
<v Mackenzie Arnold (Institute for Law &amp; AI)>attorney general and with
whistleblowers to to bring cases.

7037
06:51:21.086 --> 06:51:25.486
<v Mackenzie Arnold (Institute for Law &amp; AI)>Like I said, liability won't be the only
solution or the entire solution, but it's a

7038
06:51:25.486 --> 06:51:28.047
<v Mackenzie Arnold (Institute for Law &amp; AI)>great first step, and I
appreciate all of your efforts.

7039
06:51:28.766 --> 06:51:29.887
<v CM Carmen De La Rosa>Thank you so much.

7040
06:51:32.205 --> 06:51:36.600
<v Speaker Julie Menin>I really wanna thank this panel. There's
so much to unpack in what all of you said

7041
06:51:36.600 --> 06:51:38.920
<v Speaker Julie Menin>that I'll I'll I'll try to be brief.

7042
06:51:39.560 --> 06:51:44.440
<v Speaker Julie Menin>Ju kay, I appreciated how you said that,
first of all, we wanna make sure that New

7043
06:51:44.440 --> 06:51:50.066
<v Speaker Julie Menin>York City remains the tech hub that it is,
and that is how I began this hearing by

7044
06:51:50.066 --> 06:51:51.986
<v Speaker Julie Menin>really talking about the 400,000

7045
06:51:51.986 --> 06:51:53.586
<v Speaker Julie Menin>tech jobs that are here,

7046
06:51:53.906 --> 06:51:57.906
<v Speaker Julie Menin>by all the tech startups that are locating
here, and that is something that we

7047
06:51:57.906 --> 06:51:58.627
<v Speaker Julie Menin>welcome

7048
06:51:58.785 --> 06:52:00.307
<v Speaker Julie Menin>and we want to encourage.

7049
06:52:01.090 --> 06:52:04.690
<v Speaker Julie Menin>I think that actually goes
hand in hand with responsible

7050
06:52:04.690 --> 06:52:05.730
<v Speaker Julie Menin>regulation.

7051
06:52:05.969 --> 06:52:09.969
<v Speaker Julie Menin>The idea that these two
ideas are mutually exclusive,

7052
06:52:09.969 --> 06:52:15.854
<v Speaker Julie Menin>I don't agree with at all. If we really
want to instill trust and faith in AI, then

7053
06:52:15.854 --> 06:52:19.055
<v Speaker Julie Menin>the best thing we can do
is responsible safeguards,

7054
06:52:19.055 --> 06:52:19.855
<v Speaker Julie Menin>which is what,

7055
06:52:20.334 --> 06:52:24.094
<v Speaker Julie Menin>my colleagues and I are doing today with
the bills we're putting forward. So thank

7056
06:52:24.094 --> 06:52:28.254
<v Speaker Julie Menin>you for saying that. It makes a lot of sense.
I guess the question I have for any of

7057
06:52:28.254 --> 06:52:32.700
<v Speaker Julie Menin>you that want to answer it is, yes, one
of our bills requires this third party

7058
06:52:32.700 --> 06:52:33.660
<v Speaker Julie Menin>validation,

7059
06:52:34.139 --> 06:52:34.540
<v Speaker Julie Menin>potentially

7060
06:52:34.940 --> 06:52:36.060
<v Speaker Julie Menin>a kill switch,

7061
06:52:36.299 --> 06:52:40.620
<v Speaker Julie Menin>the... A third party validator who's free
from conflicts of interest. I wanna ask

7062
06:52:40.620 --> 06:52:43.434
<v Speaker Julie Menin>you the same question that I
asked the companies earlier.

7063
06:52:43.994 --> 06:52:46.795
<v Speaker Julie Menin>Who then is best to be the third party

7064
06:52:46.955 --> 06:52:49.355
<v Speaker Julie Menin>validator? Who is best positioned

7065
06:52:49.514 --> 06:52:52.154
<v Speaker Julie Menin>to be in that truly independent,

7066
06:52:52.154 --> 06:52:53.994
<v Speaker Julie Menin>free of conflict of interest,

7067
06:52:54.234 --> 06:52:54.715
<v Speaker Julie Menin>but

7068
06:52:55.190 --> 06:52:56.950
<v Speaker Julie Menin>up to speed technologically

7069
06:52:56.950 --> 06:52:59.830
<v Speaker Julie Menin>to provide that urgent validation?

7070
06:53:01.510 --> 06:53:05.190
<v Julia Stoyanovich (NYU Center for Responsible AI)>May? Okay. Yeah. Please. I'm opening it
up to any of you. Thank you. So I I don't

7071
06:53:05.190 --> 06:53:09.256
<v Julia Stoyanovich (NYU Center for Responsible AI)>think it's unexpected that I would say
this being that I am an academic. I think

7072
06:53:09.256 --> 06:53:13.496
<v Julia Stoyanovich (NYU Center for Responsible AI)>that academia have to lead the charge on
this. Right? Because one of the things

7073
06:53:13.496 --> 06:53:17.816
<v Julia Stoyanovich (NYU Center for Responsible AI)>we're seeing in the city is that
there simply isn't enough expertise

7074
06:53:18.455 --> 06:53:23.480
<v Julia Stoyanovich (NYU Center for Responsible AI)>to be able to oversee these systems. And,
ultimately, our goals should be to bring

7075
06:53:23.480 --> 06:53:26.840
<v Julia Stoyanovich (NYU Center for Responsible AI)>as much information as
possible into public view.

7076
06:53:27.080 --> 06:53:29.000
<v Julia Stoyanovich (NYU Center for Responsible AI)>Sunlight is the best disinfectant.

7077
06:53:29.000 --> 06:53:33.080
<v Julia Stoyanovich (NYU Center for Responsible AI)>Right? So, really, oversight
has to be public oversight.

7078
06:53:36.840 --> 06:53:38.360
<v Speaker Julie Menin>Does anyone else want to take that?

7079
06:53:39.693 --> 06:53:43.773
<v Nate Soares (MIRI)>I'd add that the folks who, the third
parties who investigated the Hugging Face

7080
06:53:43.773 --> 06:53:47.535
<v Nate Soares (MIRI)>incident seem to me to have done a good job,
and my guess is that they're overworked

7081
06:53:47.535 --> 06:53:51.695
<v Nate Soares (MIRI)>right now. I think those were, the
organizations METER and Redwood Research,

7082
06:53:51.934 --> 06:53:54.890
<v Nate Soares (MIRI)>but they they seem to have
done a good job so far.

7083
06:53:56.090 --> 06:54:00.330
<v Nathan Sheard (Electronic Frontier Foundation)>Yeah. I think it's an excellent question.
Who who is the who is the right party? And

7084
06:54:00.330 --> 06:54:04.250
<v Nathan Sheard (Electronic Frontier Foundation)>and I I don't don't have a great answer
for that. What I would like to say, though,

7085
06:54:04.250 --> 06:54:08.145
<v Nathan Sheard (Electronic Frontier Foundation)>is there's, you know, there's still
research being done on the efficacy of kill

7086
06:54:08.145 --> 06:54:08.785
<v Nathan Sheard (Electronic Frontier Foundation)>switches.

7087
06:54:09.105 --> 06:54:12.705
<v Nathan Sheard (Electronic Frontier Foundation)>What I caution the
council to think about is

7088
06:54:13.904 --> 06:54:17.426
<v Nathan Sheard (Electronic Frontier Foundation)>the risk of a kill switch being used

7089
06:54:17.744 --> 06:54:19.984
<v Nathan Sheard (Electronic Frontier Foundation)>as a tool to chill on popular speech.

7090
06:54:20.550 --> 06:54:23.990
<v Nathan Sheard (Electronic Frontier Foundation)>And so really thinking through and make sure
that as we think about the ways that we

7091
06:54:23.990 --> 06:54:25.191
<v Nathan Sheard (Electronic Frontier Foundation)>wanna provide

7092
06:54:25.271 --> 06:54:29.352
<v Nathan Sheard (Electronic Frontier Foundation)>the right protections for New Yorkers, that
we're also not putting our... Building

7093
06:54:29.590 --> 06:54:32.630
<v Nathan Sheard (Electronic Frontier Foundation)>tools that in the future could be used to

7094
06:54:33.031 --> 06:54:34.470
<v Nathan Sheard (Electronic Frontier Foundation)>harm people's essential liberties.

7095
06:54:36.615 --> 06:54:40.055
<v Imran Ahmed (CCDH)>If if I can make a brief point on

7096
06:54:40.455 --> 06:54:42.455
<v Imran Ahmed (CCDH)>who could do this work,

7097
06:54:43.096 --> 06:54:44.535
<v Imran Ahmed (CCDH)>being realistic,

7098
06:54:46.855 --> 06:54:48.660
<v Imran Ahmed (CCDH)>to date, civil society

7099
06:54:48.660 --> 06:54:51.380
<v Imran Ahmed (CCDH)>studying social media and AI platforms

7100
06:54:51.380 --> 06:54:55.780
<v Imran Ahmed (CCDH)>has had one dirty secret that most of
its funding has come from the platforms

7101
06:54:55.780 --> 06:54:56.660
<v Imran Ahmed (CCDH)>themselves.

7102
06:54:56.660 --> 06:55:00.900
<v Imran Ahmed (CCDH)>So Meta and Google were enormous
funders of extremism research,

7103
06:55:01.220 --> 06:55:03.300
<v Imran Ahmed (CCDH)>of safety research on their own platforms.

7104
06:55:03.676 --> 06:55:06.555
<v Imran Ahmed (CCDH)>And when I would meet... And we don't take
their money. When I would meet with their

7105
06:55:06.555 --> 06:55:10.715
<v Imran Ahmed (CCDH)>CEOs, they would say... The CEOs of the
civil society bodies which were funded by

7106
06:55:10.715 --> 06:55:14.156
<v Imran Ahmed (CCDH)>them, they'd say, we're not allowed to
say that they haven't done enough. We can

7107
06:55:14.156 --> 06:55:18.130
<v Imran Ahmed (CCDH)>always say they could do more, but not they
haven't done enough. Mhmm. So there are

7108
06:55:18.130 --> 06:55:19.730
<v Imran Ahmed (CCDH)>real compromises there.

7109
06:55:20.050 --> 06:55:22.850
<v Imran Ahmed (CCDH)>In Europe or the rest of the world, and
you can hear from my accent, and I actually

7110
06:55:22.850 --> 06:55:23.970
<v Imran Ahmed (CCDH)>grew up in Britain,

7111
06:55:24.050 --> 06:55:24.531
<v Imran Ahmed (CCDH)>that

7112
06:55:25.010 --> 06:55:31.170
<v Imran Ahmed (CCDH)>the government would fund this. And when
it comes to giving a body the resources to

7113
06:55:31.170 --> 06:55:36.266
<v Imran Ahmed (CCDH)>have the speed and creativity to be
effective in dealing with the fastest moving

7114
06:55:36.266 --> 06:55:37.225
<v Imran Ahmed (CCDH)>technology

7115
06:55:37.465 --> 06:55:38.984
<v Imran Ahmed (CCDH)>in human history,

7116
06:55:39.225 --> 06:55:42.904
<v Imran Ahmed (CCDH)>I cannot think of a nongovernmental source
of funding that would be sufficient to

7117
06:55:42.904 --> 06:55:44.904
<v Imran Ahmed (CCDH)>get it to scale fast enough.

7118
06:55:45.145 --> 06:55:49.350
<v Imran Ahmed (CCDH)>So the question for New York... For New
York City and New York State, which is a

7119
06:55:49.350 --> 06:55:54.150
<v Imran Ahmed (CCDH)>substantial economic unit, the budget of
New York State is one quarter that of the

7120
06:55:54.150 --> 06:55:55.830
<v Imran Ahmed (CCDH)>entire United Kingdom,

7121
06:55:56.070 --> 06:55:57.430
<v Imran Ahmed (CCDH)>one state alone.

7122
06:55:57.830 --> 06:56:01.750
<v Imran Ahmed (CCDH)>It does have the economic capacity to do
so, and I think you would have to seriously

7123
06:56:01.750 --> 06:56:05.805
<v Imran Ahmed (CCDH)>think about whether or not you're
willing to put some public

7124
06:56:05.805 --> 06:56:09.965
<v Imran Ahmed (CCDH)>money towards creating an institution
that's able to operate with the speed,

7125
06:56:09.965 --> 06:56:12.205
<v Imran Ahmed (CCDH)>resources, and creativity to be effective.

7126
06:56:13.885 --> 06:56:17.320
<v Jukay Hsu (Bright Futures)>Thank you, speaker. And and, yeah, agree
with these kind of comments. It's it's such

7127
06:56:17.320 --> 06:56:20.520
<v Jukay Hsu (Bright Futures)>a tough challenge and totally agree with
you that New York can be both. And I think

7128
06:56:20.520 --> 06:56:24.520
<v Jukay Hsu (Bright Futures)>people are looking for answers that's not
so black and white and realize the reality

7129
06:56:24.520 --> 06:56:25.160
<v Jukay Hsu (Bright Futures)>of this.

7130
06:56:25.959 --> 06:56:29.560
<v Jukay Hsu (Bright Futures)>I... You know, that's such an important
question. You know? I think, who can be

7131
06:56:29.560 --> 06:56:30.760
<v Jukay Hsu (Bright Futures)>third party validators?

7132
06:56:31.225 --> 06:56:34.744
<v Jukay Hsu (Bright Futures)>I I think it's yes and
and and for for everyone.

7133
06:56:35.465 --> 06:56:39.305
<v Jukay Hsu (Bright Futures)>The labs are doing a lot. Like you heard,
Anthropic is investing a lot into this.

7134
06:56:39.305 --> 06:56:43.064
<v Jukay Hsu (Bright Futures)>There are many more people going to AI
alignment and safety research, which is so

7135
06:56:43.064 --> 06:56:43.305
<v Jukay Hsu (Bright Futures)>important.

7136
06:56:44.120 --> 06:56:47.000
<v Jukay Hsu (Bright Futures)>I I think the one thing that's
challenging is gonna be...

7137
06:56:48.600 --> 06:56:51.799
<v Jukay Hsu (Bright Futures)>I think the things are, like
everyone said, moving so quickly.

7138
06:56:52.120 --> 06:56:56.440
<v Jukay Hsu (Bright Futures)>I think everyone can be a judge. Experts
today... Like yesterday, six months ago, is

7139
06:56:56.440 --> 06:56:57.479
<v Jukay Hsu (Bright Futures)>not the same thing as today.

7140
06:56:59.395 --> 06:57:04.035
<v Jukay Hsu (Bright Futures)>And what's really gonna happen is that you
all can also be the judges by using this.

7141
06:57:04.035 --> 06:57:06.914
<v Jukay Hsu (Bright Futures)>Like, what makes sense and what doesn't
make sense. A lot of it's, I think, common

7142
06:57:06.914 --> 06:57:08.195
<v Jukay Hsu (Bright Futures)>sense hopefully in the future.

7143
06:57:08.674 --> 06:57:12.035
<v Jukay Hsu (Bright Futures)>And then the last point I kinda make to
that, even though it seems counterintuitive

7144
06:57:12.035 --> 06:57:17.470
<v Jukay Hsu (Bright Futures)>about who are the experts where we all
can play a role, We have agency is that in

7145
06:57:17.470 --> 06:57:21.150
<v Jukay Hsu (Bright Futures)>the future, it's not just these
five, what we call frontier labs.

7146
06:57:21.710 --> 06:57:25.230
<v Jukay Hsu (Bright Futures)>All the companies here in New York
are fine tuning their own models.

7147
06:57:25.791 --> 06:57:31.074
<v Jukay Hsu (Bright Futures)>We all may have our own very powerful models
in the future. So, actually, beyond the

7148
06:57:31.074 --> 06:57:32.275
<v Jukay Hsu (Bright Futures)>Frontier Labs,

7149
06:57:32.994 --> 06:57:35.234
<v Jukay Hsu (Bright Futures)>part of the challenge may be, actually,

7150
06:57:35.555 --> 06:57:40.115
<v Jukay Hsu (Bright Futures)>you know, your student researcher downloading
a model and uploading it and and up...

7151
06:57:40.115 --> 06:57:43.941
<v Jukay Hsu (Bright Futures)>And fine tuning it to create
capabilities or test things or,

7152
06:57:44.102 --> 06:57:46.102
<v Jukay Hsu (Bright Futures)>you know, a company
doing that inadvertently.

7153
06:57:46.102 --> 06:57:51.220
<v Jukay Hsu (Bright Futures)>So I think it's a much broader set of things
than just relying on experts. I think,

7154
06:57:51.220 --> 06:57:55.701
<v Jukay Hsu (Bright Futures)>again, that is important, but I think we
need to really widen the scope around that.

7155
06:57:55.875 --> 06:58:00.355
<v Jukay Hsu (Bright Futures)>This is not a technology where it's gonna
be nuclear weapons controlled by a dozen

7156
06:58:00.355 --> 06:58:02.436
<v Jukay Hsu (Bright Futures)>states. These will be as powerful,

7157
06:58:02.436 --> 06:58:06.115
<v Jukay Hsu (Bright Futures)>but we will have use and access to that.
And I don't think that's gonna stop.

7158
06:58:09.234 --> 06:58:11.550
<v Mackenzie Arnold (Institute for Law &amp; AI)>If I can jump in with
just one last thought.

7159
06:58:12.190 --> 06:58:15.950
<v Mackenzie Arnold (Institute for Law &amp; AI)>I wanna pick up on the yes and approach
because I think this is is right. I was

7160
06:58:15.950 --> 06:58:19.710
<v Mackenzie Arnold (Institute for Law &amp; AI)>just, talking with someone the other day
who works at one of these evaluators today,

7161
06:58:19.870 --> 06:58:23.629
<v Mackenzie Arnold (Institute for Law &amp; AI)>and they they joked that there were dozens
of, entities that could do this. And by

7162
06:58:23.629 --> 06:58:27.416
<v Mackenzie Arnold (Institute for Law &amp; AI)>that, they meant a dozen people,
not not a dozen different companies.

7163
06:58:27.895 --> 06:58:30.775
<v Mackenzie Arnold (Institute for Law &amp; AI)>I think we'll need to
rely on both academia

7164
06:58:30.775 --> 06:58:32.615
<v Mackenzie Arnold (Institute for Law &amp; AI)>and, private parties.

7165
06:58:33.414 --> 06:58:37.016
<v Mackenzie Arnold (Institute for Law &amp; AI)>I think we'll also have to rely on the
public. Right? A big part of incident

7166
06:58:37.016 --> 06:58:39.701
<v Mackenzie Arnold (Institute for Law &amp; AI)>reporting is revealing some of
that information to the public.

7167
06:58:40.102 --> 06:58:43.541
<v Mackenzie Arnold (Institute for Law &amp; AI)>One of the only times I've ever been happy
with Twitter was this summer as people

7168
06:58:43.541 --> 06:58:47.140
<v Mackenzie Arnold (Institute for Law &amp; AI)>were arguing over the incidents, and we
were try... Starting to get a a better idea

7169
06:58:47.140 --> 06:58:47.941
<v Mackenzie Arnold (Institute for Law &amp; AI)>of what happened.

7170
06:58:48.500 --> 06:58:53.984
<v Mackenzie Arnold (Institute for Law &amp; AI)>One note of caution I'll add is that
one way this goes wrong is that the few

7171
06:58:53.984 --> 06:58:58.705
<v Mackenzie Arnold (Institute for Law &amp; AI)>potential evaluators that there are get
tasked with many different things. Right?

7172
06:58:59.184 --> 06:59:02.865
<v Mackenzie Arnold (Institute for Law &amp; AI)>And there's a chance that New York should
lead on this and that they should also

7173
06:59:02.865 --> 06:59:05.664
<v Mackenzie Arnold (Institute for Law &amp; AI)>coordinate with the state and
with California and others,

7174
06:59:05.984 --> 06:59:08.225
<v Mackenzie Arnold (Institute for Law &amp; AI)>and not be overly
prescriptive in the exact

7175
06:59:08.740 --> 06:59:12.260
<v Mackenzie Arnold (Institute for Law &amp; AI)>sort of analyses that need to be done
by the entities. Where you can be

7176
06:59:12.260 --> 06:59:16.420
<v Mackenzie Arnold (Institute for Law &amp; AI)>exceptionally helpful is in sending a clear
market signal. Right? So requiring that

7177
06:59:16.420 --> 06:59:20.180
<v Mackenzie Arnold (Institute for Law &amp; AI)>these these audits or or or validations be
done in order to get contracts with the

7178
06:59:20.180 --> 06:59:21.940
<v Mackenzie Arnold (Institute for Law &amp; AI)>city, or or things like that.

7179
06:59:23.863 --> 06:59:24.744
<v Speaker Julie Menin>Thank you.

7180
06:59:27.064 --> 06:59:31.064
<v CM Carmen De La Rosa>Thank you so much. I have two questions.
The first one is for anyone in the panel

7181
06:59:31.064 --> 06:59:35.385
<v CM Carmen De La Rosa>who wants to take it. Do you all have
concerns related to open weight models being

7182
06:59:35.385 --> 06:59:40.170
<v CM Carmen De La Rosa>available? And if safety features can be
eliminated from open weight models, how do

7183
06:59:40.170 --> 06:59:42.170
<v CM Carmen De La Rosa>you defend against bad actors who

7184
06:59:42.330 --> 06:59:43.130
<v CM Carmen De La Rosa>use them?

7185
06:59:46.890 --> 06:59:52.135
<v Nathan Sheard (Electronic Frontier Foundation)>I can start. I think one of the things I
think about when we're thinking about open

7186
06:59:52.135 --> 06:59:55.496
<v Nathan Sheard (Electronic Frontier Foundation)>weight models is that we do want
the ability for researchers

7187
06:59:55.736 --> 06:59:56.855
<v Nathan Sheard (Electronic Frontier Foundation)>and startups

7188
06:59:57.096 --> 07:00:03.096
<v Nathan Sheard (Electronic Frontier Foundation)>and potential future competitors to be able
to develop tools and resources and also

7189
07:00:03.096 --> 07:00:07.500
<v Nathan Sheard (Electronic Frontier Foundation)>so that we can be able to evaluate these
tools and see how they really work, which

7190
07:00:07.500 --> 07:00:09.420
<v Nathan Sheard (Electronic Frontier Foundation)>is difficult with the closed models.

7191
07:00:09.900 --> 07:00:14.941
<v Nathan Sheard (Electronic Frontier Foundation)>But then it certainly does lend the question
then of how do we make sure that the

7192
07:00:15.260 --> 07:00:15.980
<v Nathan Sheard (Electronic Frontier Foundation)>proper

7193
07:00:16.380 --> 07:00:18.460
<v Nathan Sheard (Electronic Frontier Foundation)>protections are in place? And I think

7194
07:00:19.355 --> 07:00:22.475
<v Nathan Sheard (Electronic Frontier Foundation)>we have some of the tools
to do that already. When

7195
07:00:22.795 --> 07:00:24.875
<v Nathan Sheard (Electronic Frontier Foundation)>a act is

7196
07:00:25.674 --> 07:00:30.475
<v Nathan Sheard (Electronic Frontier Foundation)>execute... That an act takes place that is
potentially harmful or or threatening to

7197
07:00:30.475 --> 07:00:31.594
<v Nathan Sheard (Electronic Frontier Foundation)>New Yorkers their

7198
07:00:32.230 --> 07:00:35.190
<v Nathan Sheard (Electronic Frontier Foundation)>freedom... Freedoms or or,
you know, physical safety,

7199
07:00:36.150 --> 07:00:40.791
<v Nathan Sheard (Electronic Frontier Foundation)>pursue... Use the laws that we have now to
pursue the people who are ex... Who are

7200
07:00:40.870 --> 07:00:45.510
<v Nathan Sheard (Electronic Frontier Foundation)>engaging in those activities. Right?
So rather than try to create...

7201
07:00:45.590 --> 07:00:45.830
<v Nathan Sheard (Electronic Frontier Foundation)>Guide

7202
07:00:46.736 --> 07:00:47.297
<v Nathan Sheard (Electronic Frontier Foundation)>look

7203
07:00:47.695 --> 07:00:48.736
<v Nathan Sheard (Electronic Frontier Foundation)>to the

7204
07:00:49.455 --> 07:00:52.736
<v Nathan Sheard (Electronic Frontier Foundation)>incumbents, the powerful corporations to
put in the protections that will keep us

7205
07:00:52.736 --> 07:00:53.297
<v Nathan Sheard (Electronic Frontier Foundation)>safe,

7206
07:00:53.855 --> 07:00:58.016
<v Nathan Sheard (Electronic Frontier Foundation)>look look... Use the existing laws that we
have to be able to keep people safe and

7207
07:00:58.016 --> 07:01:01.455
<v Nathan Sheard (Electronic Frontier Foundation)>really look at who's... Who are the people
that are creating the harm and and use

7208
07:01:01.455 --> 07:01:03.640
<v Nathan Sheard (Electronic Frontier Foundation)>the things that we have in place to pursue

7209
07:01:04.121 --> 07:01:06.361
<v Nathan Sheard (Electronic Frontier Foundation)>justice and protection security there.

7210
07:01:08.041 --> 07:01:09.160
<v CM Carmen De La Rosa>Thank you for that.

7211
07:01:09.800 --> 07:01:13.800
<v CM Carmen De La Rosa>I do have a a a direct question
for doctors Dayanovich.

7212
07:01:14.055 --> 07:01:15.895
<v CM Carmen De La Rosa>I hope I didn't mess that up too bad.

7213
07:01:16.455 --> 07:01:21.574
<v CM Carmen De La Rosa>I know you said, you know, you you you seem
skeptical of the catastrophic narrative

7214
07:01:21.574 --> 07:01:22.375
<v CM Carmen De La Rosa>around,

7215
07:01:23.094 --> 07:01:24.615
<v CM Carmen De La Rosa>what's occurring,

7216
07:01:24.615 --> 07:01:27.414
<v CM Carmen De La Rosa>but, you still have concerns
about possible harms.

7217
07:01:27.950 --> 07:01:31.629
<v CM Carmen De La Rosa>Can you describe to us,
based on your expertise,

7218
07:01:31.790 --> 07:01:35.549
<v CM Carmen De La Rosa>what are the harms that are
most relevant to New Yorkers?

7219
07:01:36.988 --> 07:01:38.750
<v Julia Stoyanovich (NYU Center for Responsible AI)>Thank you for this question. Yeah.

7220
07:01:40.846 --> 07:01:45.805
<v Julia Stoyanovich (NYU Center for Responsible AI)>Thank you for this question. So, we are
running these systems in the city, in the

7221
07:01:45.805 --> 07:01:50.445
<v Julia Stoyanovich (NYU Center for Responsible AI)>public sector, in the private sector, and
we have seen in New York very specific

7222
07:01:50.445 --> 07:01:51.244
<v Julia Stoyanovich (NYU Center for Responsible AI)>harms,

7223
07:01:51.645 --> 07:01:54.920
<v Julia Stoyanovich (NYU Center for Responsible AI)>that come from deployed systems
and with very consequential,

7224
07:01:55.158 --> 07:01:59.959
<v Julia Stoyanovich (NYU Center for Responsible AI)>decisions. So in, my own
work, I did an audit of,

7225
07:02:00.360 --> 07:02:03.559
<v Julia Stoyanovich (NYU Center for Responsible AI)>a hiring system that claims to construct

7226
07:02:03.799 --> 07:02:06.120
<v Julia Stoyanovich (NYU Center for Responsible AI)>a job seeker's personality profile,

7227
07:02:06.280 --> 07:02:11.936
<v Julia Stoyanovich (NYU Center for Responsible AI)>a bunch of numbers from their resume or
social media feed, like a Twitter feed at

7228
07:02:11.936 --> 07:02:13.615
<v Julia Stoyanovich (NYU Center for Responsible AI)>the time or a LinkedIn profile.

7229
07:02:13.936 --> 07:02:16.256
<v Julia Stoyanovich (NYU Center for Responsible AI)>And these systems are complete nonsense.

7230
07:02:16.336 --> 07:02:18.576
<v Julia Stoyanovich (NYU Center for Responsible AI)>Right? For the same job seeker,

7231
07:02:18.576 --> 07:02:22.496
<v Julia Stoyanovich (NYU Center for Responsible AI)>if you change something like the file
format of the resume, it's gonna give you a

7232
07:02:22.496 --> 07:02:23.695
<v Julia Stoyanovich (NYU Center for Responsible AI)>completely different personality.

7233
07:02:24.451 --> 07:02:29.092
<v Julia Stoyanovich (NYU Center for Responsible AI)>Right? So we need to actually be looking
at whether these systems even work. Right?

7234
07:02:29.092 --> 07:02:32.771
<v Julia Stoyanovich (NYU Center for Responsible AI)>I think that that should be just stop
one of anything before we worry about

7235
07:02:32.771 --> 07:02:34.050
<v Julia Stoyanovich (NYU Center for Responsible AI)>existential risk.

7236
07:02:34.451 --> 07:02:38.291
<v Julia Stoyanovich (NYU Center for Responsible AI)>We had a similar scenario
with the the MyCity chatbot.

7237
07:02:39.016 --> 07:02:42.535
<v Julia Stoyanovich (NYU Center for Responsible AI)>Right? I mean, we all have heard
about that fiasco where the city spent

7238
07:02:42.775 --> 07:02:44.055
<v Julia Stoyanovich (NYU Center for Responsible AI)>$500,000,

7239
07:02:44.055 --> 07:02:47.416
<v Julia Stoyanovich (NYU Center for Responsible AI)>conservatively speaking,
on a bot that advised

7240
07:02:47.975 --> 07:02:52.590
<v Julia Stoyanovich (NYU Center for Responsible AI)>small and medium sized business owners to
break the law, which was quite problematic

7241
07:02:52.590 --> 07:02:56.590
<v Julia Stoyanovich (NYU Center for Responsible AI)>speaking with the city's authority. So I
really think that we need to be looking at

7242
07:02:56.590 --> 07:02:58.432
<v Julia Stoyanovich (NYU Center for Responsible AI)>cases such as these ones

7243
07:02:58.830 --> 07:03:01.470
<v Julia Stoyanovich (NYU Center for Responsible AI)>where there are very known risks

7244
07:03:01.630 --> 07:03:03.871
<v Julia Stoyanovich (NYU Center for Responsible AI)>and start there and
clean up our own house.

7245
07:03:04.916 --> 07:03:09.637
<v CM Carmen De La Rosa>Thank you for that, and and I wanna thank
you all for your answers and insights. We

7246
07:03:09.637 --> 07:03:11.477
<v CM Carmen De La Rosa>have three members on stack,

7247
07:03:11.637 --> 07:03:16.195
<v CM Chi Ossé>council member Osei, council member Hanif,
and then council member Maloney to close

7248
07:03:16.195 --> 07:03:20.355
<v CM Chi Ossé>this panel. Thank you, chair. Miss
Myers, there are a lot of issues in AI.

7249
07:03:20.970 --> 07:03:25.451
<v CM Chi Ossé>You know, currently, we hear about the
harms that it has for our children,

7250
07:03:26.010 --> 07:03:26.890
<v CM Chi Ossé>workers,

7251
07:03:27.050 --> 07:03:27.932
<v CM Chi Ossé>consumers,

7252
07:03:28.010 --> 07:03:31.130
<v CM Chi Ossé>and the coming threats
of catastrophic risk.

7253
07:03:31.291 --> 07:03:35.850
<v CM Chi Ossé>Do any of these issues conflict with one
another, and what should we address, if not

7254
07:03:35.850 --> 07:03:36.330
<v CM Chi Ossé>all of them?

7255
07:03:37.557 --> 07:03:42.836
<v Anna Myers (Who Decides)>Yeah. Thank you for the question. This is
a yes and question. This is, we cannot

7256
07:03:42.836 --> 07:03:45.557
<v Anna Myers (Who Decides)>silo these into separate scenarios.

7257
07:03:46.355 --> 07:03:48.756
<v Anna Myers (Who Decides)>Kids, workers, consumers,

7258
07:03:48.756 --> 07:03:52.436
<v Anna Myers (Who Decides)>catastrophic risk can all be
addressed in parallel to each other.

7259
07:03:53.371 --> 07:03:56.890
<v Anna Myers (Who Decides)>Just like the council has done a great
job with their bill package of covering

7260
07:03:56.890 --> 07:03:58.410
<v Anna Myers (Who Decides)>multiple things at once,

7261
07:03:58.730 --> 07:04:03.930
<v Anna Myers (Who Decides)>we need to resist the urge to fight
against each other on whose bill is more

7262
07:04:03.930 --> 07:04:07.975
<v Anna Myers (Who Decides)>important at times or what we're protecting
at times. When I worked on the RAISE

7263
07:04:07.975 --> 07:04:12.613
<v Anna Myers (Who Decides)>Act, we were working on c two p a at the
same time. There's no reason for us to stop

7264
07:04:12.613 --> 07:04:17.574
<v Anna Myers (Who Decides)>protect... The protections of catastrophic
risks while at the same time working on

7265
07:04:17.574 --> 07:04:18.613
<v Anna Myers (Who Decides)>transparency data.

7266
07:04:19.670 --> 07:04:24.229
<v Anna Myers (Who Decides)>The most important question we need to
have right now is who is making these

7267
07:04:24.229 --> 07:04:27.350
<v Anna Myers (Who Decides)>decisions, and it shouldn't be five CEOs.

7268
07:04:27.510 --> 07:04:30.629
<v Anna Myers (Who Decides)>It should be the American people and the

7269
07:04:31.125 --> 07:04:34.965
<v CM Chi Ossé>electeds that they voted for. Thank you.
And and based on your conversations

7270
07:04:34.965 --> 07:04:39.285
<v CM Chi Ossé>and polling, what are Americans most
worried about when it comes to artificial

7271
07:04:39.285 --> 07:04:40.166
<v CM Chi Ossé>intelligence?

7272
07:04:40.805 --> 07:04:41.205
<v Anna Myers (Who Decides)>Yeah.

7273
07:04:41.605 --> 07:04:44.324
<v Anna Myers (Who Decides)>Right now, that they
don't have a say in it.

7274
07:04:44.805 --> 07:04:48.441
<v Anna Myers (Who Decides)>It's no accident that we
named our org who decides.

7275
07:04:48.600 --> 07:04:53.800
<v Anna Myers (Who Decides)>It's time for the American people to have
a say in it and our polling showing that

7276
07:04:53.800 --> 07:04:57.160
<v Anna Myers (Who Decides)>they feel that they are not a part of the
conversation and that this is happening

7277
07:04:57.160 --> 07:04:59.160
<v Anna Myers (Who Decides)>around them and they wanna be in it.

7278
07:04:59.945 --> 07:05:00.746
<v CM Chi Ossé>Thank you.

7279
07:05:01.625 --> 07:05:04.346
<v CM Carmen De La Rosa>Thank you so much. Council member Hanif,

7280
07:05:04.346 --> 07:05:08.664
<v CM Shahana Hanif>followed by Maloney. Thank you. And thank
you for your expertise, and this is for

7281
07:05:08.664 --> 07:05:13.465
<v CM Shahana Hanif>everyone on the panel. As AI and facial
recognition technologies become more

7282
07:05:13.465 --> 07:05:13.785
<v CM Shahana Hanif>powerful,

7283
07:05:14.570 --> 07:05:19.290
<v CM Shahana Hanif>what safeguards should cities put in place
to prevent surveillance tools from being

7284
07:05:19.290 --> 07:05:20.650
<v CM Shahana Hanif>used to target people

7285
07:05:21.050 --> 07:05:24.490
<v CM Shahana Hanif>based on their identity,
associations, or political activity?

7286
07:05:24.810 --> 07:05:29.285
<v CM Shahana Hanif>Would legislation like my ban the scan
bill, which would regulate and prohibit

7287
07:05:29.285 --> 07:05:32.807
<v CM Shahana Hanif>biometric surveillance in
places of public accommodation,

7288
07:05:33.125 --> 07:05:37.365
<v CM Shahana Hanif>provide an important safeguard? And I'm
particularly interested in your assessment

7289
07:05:37.365 --> 07:05:42.050
<v CM Shahana Hanif>of Madison Square Garden owner James
Dolan's use of biometric surveillance

7290
07:05:42.050 --> 07:05:45.090
<v CM Shahana Hanif>to identify and exclude individuals?

7291
07:05:45.170 --> 07:05:45.890
<v CM Shahana Hanif>Thank you.

7292
07:05:48.770 --> 07:05:53.970
<v Julia Stoyanovich (NYU Center for Responsible AI)>I can... Anyone. Yeah. Okay. So, absolutely,
I'm very supportive of your bill, and I

7293
07:05:53.970 --> 07:05:59.885
<v Julia Stoyanovich (NYU Center for Responsible AI)>think that, we have a lack of data protection
in the city, in this country, in the

7294
07:05:59.885 --> 07:06:03.885
<v Julia Stoyanovich (NYU Center for Responsible AI)>state. Right? And data protection
is a tremendously important issue.

7295
07:06:04.686 --> 07:06:06.445
<v Julia Stoyanovich (NYU Center for Responsible AI)>And because of how convenient

7296
07:06:06.525 --> 07:06:07.645
<v Julia Stoyanovich (NYU Center for Responsible AI)>AI tools are,

7297
07:06:08.480 --> 07:06:13.041
<v Julia Stoyanovich (NYU Center for Responsible AI)>people very often voluntarily give up very
private information about themselves. So

7298
07:06:13.041 --> 07:06:16.720
<v Julia Stoyanovich (NYU Center for Responsible AI)>I actually have in my written
testimony that's much longer examples

7299
07:06:16.720 --> 07:06:17.361
<v Julia Stoyanovich (NYU Center for Responsible AI)>of

7300
07:06:18.320 --> 07:06:22.640
<v Julia Stoyanovich (NYU Center for Responsible AI)>AI assistance being marketed that essentially
people give access to their calendars,

7301
07:06:22.640 --> 07:06:23.600
<v Julia Stoyanovich (NYU Center for Responsible AI)>their children's

7302
07:06:24.215 --> 07:06:25.254
<v Julia Stoyanovich (NYU Center for Responsible AI)>biometrics,

7303
07:06:25.254 --> 07:06:26.215
<v Julia Stoyanovich (NYU Center for Responsible AI)>the schedules,

7304
07:06:26.215 --> 07:06:30.135
<v Julia Stoyanovich (NYU Center for Responsible AI)>all of the school contact information.
Right? So we really need to educate the

7305
07:06:30.135 --> 07:06:31.176
<v Julia Stoyanovich (NYU Center for Responsible AI)>public about

7306
07:06:32.695 --> 07:06:36.615
<v Julia Stoyanovich (NYU Center for Responsible AI)>the benefits, but also also the
risks that data sharing poses,

7307
07:06:37.140 --> 07:06:41.541
<v Julia Stoyanovich (NYU Center for Responsible AI)>and we need to establish a proper data
protection regime in the city. There's no way

7308
07:06:41.541 --> 07:06:42.340
<v Julia Stoyanovich (NYU Center for Responsible AI)>around it.

7309
07:06:44.020 --> 07:06:44.580
<v Nathan Sheard (Electronic Frontier Foundation)>So

7310
07:06:44.900 --> 07:06:48.340
<v Nathan Sheard (Electronic Frontier Foundation)>as as I mentioned, in my in my statement,

7311
07:06:48.580 --> 07:06:54.416
<v Nathan Sheard (Electronic Frontier Foundation)>I am very concerned about the... And EFF
is very concerned about the risk connected

7312
07:06:54.416 --> 07:06:55.057
<v Nathan Sheard (Electronic Frontier Foundation)>to

7313
07:06:55.297 --> 07:07:00.496
<v Nathan Sheard (Electronic Frontier Foundation)>the collection of biometric data like face
like face prints that can't be replaced.

7314
07:07:00.496 --> 07:07:03.457
<v Nathan Sheard (Electronic Frontier Foundation)>You know, if I lose my driver's license,
I get a new driver's license. I lose

7315
07:07:03.855 --> 07:07:07.615
<v Nathan Sheard (Electronic Frontier Foundation)>my Social Security or my license plate gets
out, I can replace those things. I can't

7316
07:07:07.615 --> 07:07:08.871
<v Nathan Sheard (Electronic Frontier Foundation)>replace my face. Face. And

7317
07:07:09.271 --> 07:07:12.630
<v Nathan Sheard (Electronic Frontier Foundation)>the risk of government use of

7318
07:07:12.710 --> 07:07:14.390
<v Nathan Sheard (Electronic Frontier Foundation)>face recognition technology

7319
07:07:15.350 --> 07:07:20.310
<v Nathan Sheard (Electronic Frontier Foundation)>so far outweighs any benefit. The the
the way that it will chill our ability to

7320
07:07:20.865 --> 07:07:25.906
<v Nathan Sheard (Electronic Frontier Foundation)>engage in protest, to engage in free
speech, to be able to associate with with...

7321
07:07:26.305 --> 07:07:27.746
<v Nathan Sheard (Electronic Frontier Foundation)>Freely and to

7322
07:07:28.305 --> 07:07:32.705
<v Nathan Sheard (Electronic Frontier Foundation)>search new ideas, health care, the the risk
that we'll be able to document, like, we

7323
07:07:32.705 --> 07:07:36.625
<v Nathan Sheard (Electronic Frontier Foundation)>go into health care, how we move around
the city is considerable. There's no way to

7324
07:07:36.930 --> 07:07:37.809
<v Nathan Sheard (Electronic Frontier Foundation)>separate

7325
07:07:37.809 --> 07:07:40.370
<v Nathan Sheard (Electronic Frontier Foundation)>that risk from government
use of the technology.

7326
07:07:40.690 --> 07:07:44.850
<v Nathan Sheard (Electronic Frontier Foundation)>I think Illinois has a good model on
how we can respond to private use of the

7327
07:07:44.850 --> 07:07:47.969
<v Nathan Sheard (Electronic Frontier Foundation)>technology. You know, we definitely don't
want to be in a situation where people are

7328
07:07:47.969 --> 07:07:50.610
<v Nathan Sheard (Electronic Frontier Foundation)>being persuaded or or otherwise

7329
07:07:50.610 --> 07:07:52.690
<v Nathan Sheard (Electronic Frontier Foundation)>not being informed
about the way that their

7330
07:07:53.225 --> 07:07:55.145
<v Nathan Sheard (Electronic Frontier Foundation)>biometric data is being collected.

7331
07:07:55.305 --> 07:07:59.225
<v Nathan Sheard (Electronic Frontier Foundation)>So by making sure that we have that the
only time that that... But we also wanna

7332
07:07:59.225 --> 07:08:02.824
<v Nathan Sheard (Electronic Frontier Foundation)>make sure that researchers are still be
able to use the theology. The ASLU years

7333
07:08:02.824 --> 07:08:03.305
<v Nathan Sheard (Electronic Frontier Foundation)>ago,

7334
07:08:04.023 --> 07:08:07.145
<v Nathan Sheard (Electronic Frontier Foundation)>used face recognition on a
number of congresspeople,

7335
07:08:07.320 --> 07:08:11.240
<v Nathan Sheard (Electronic Frontier Foundation)>and they built a... And was
able to show how it was...

7336
07:08:11.720 --> 07:08:16.041
<v Nathan Sheard (Electronic Frontier Foundation)>How easily it misidentified people,
especially black, brown folks, women, older

7337
07:08:16.041 --> 07:08:16.602
<v Nathan Sheard (Electronic Frontier Foundation)>folks.

7338
07:08:17.400 --> 07:08:21.725
<v Nathan Sheard (Electronic Frontier Foundation)>And so, like, that kind of research and
use can be really helpful. So we... But we

7339
07:08:21.725 --> 07:08:25.246
<v Nathan Sheard (Electronic Frontier Foundation)>need to make sure... So how do
we protect people from the harms

7340
07:08:25.246 --> 07:08:29.965
<v Nathan Sheard (Electronic Frontier Foundation)>by... But still allow that research and
and other, you know, consensual use to be

7341
07:08:29.965 --> 07:08:33.422
<v Nathan Sheard (Electronic Frontier Foundation)>done? And I think... And that, again, that
model of making sure that people have

7342
07:08:33.422 --> 07:08:35.102
<v Nathan Sheard (Electronic Frontier Foundation)>informed written consent

7343
07:08:35.262 --> 07:08:39.660
<v Nathan Sheard (Electronic Frontier Foundation)>about how the technology is being used and
how the data is gonna be stored and that

7344
07:08:39.660 --> 07:08:43.021
<v Nathan Sheard (Electronic Frontier Foundation)>they can only be used for the ways that
have been explicitly approved by the

7345
07:08:43.021 --> 07:08:46.855
<v Nathan Sheard (Electronic Frontier Foundation)>individual whose that data is being
collected is a way that we can make sure that

7346
07:08:46.855 --> 07:08:47.736
<v Nathan Sheard (Electronic Frontier Foundation)>people are...

7347
07:08:48.375 --> 07:08:51.576
<v Nathan Sheard (Electronic Frontier Foundation)>That we can mitigate the risk, and
it is only used in ways that people,

7348
07:08:52.135 --> 07:08:52.855
<v Nathan Sheard (Electronic Frontier Foundation)>are

7349
07:08:53.654 --> 07:08:59.656
<v Nathan Sheard (Electronic Frontier Foundation)>very... Are in in in an informed
consensual way agreeing to. Thank

7350
07:09:00.375 --> 07:09:01.895
<v CM Carmen De La Rosa>you all. Councilmember Maloney?

7351
07:09:02.988 --> 07:09:04.430
<v CM Virginia Maloney>Thank you so much, chair.

7352
07:09:05.309 --> 07:09:06.110
<v CM Virginia Maloney>I

7353
07:09:06.110 --> 07:09:09.549
<v CM Virginia Maloney>chair the council's committee
on economic development,

7354
07:09:09.950 --> 07:09:14.270
<v CM Virginia Maloney>and I want to see technology companies,
including AI technologies, build and hire

7355
07:09:14.270 --> 07:09:15.469
<v CM Virginia Maloney>here and wanted to

7356
07:09:16.525 --> 07:09:21.564
<v CM Virginia Maloney>refer to some of the testimony earlier
about that. The four companies that we heard

7357
07:09:21.564 --> 07:09:26.285
<v CM Virginia Maloney>today have offices here, but the foundational
AI work and that research is happening

7358
07:09:26.285 --> 07:09:27.885
<v CM Virginia Maloney>in mainly in San Francisco.

7359
07:09:28.300 --> 07:09:29.820
<v CM Virginia Maloney>Is that a problem?

7360
07:09:30.060 --> 07:09:33.740
<v CM Virginia Maloney>What is New York's AI advantage
to grow that industry here?

7361
07:09:35.102 --> 07:09:39.102
<v CM Virginia Maloney>And the second part of my question
is on workforce development.

7362
07:09:39.340 --> 07:09:40.941
<v CM Virginia Maloney>And in particular,

7363
07:09:40.941 --> 07:09:43.102
<v CM Virginia Maloney>Pursuit did work on reskilling.

7364
07:09:43.336 --> 07:09:48.375
<v CM Virginia Maloney>So I'm curious if you can speak to entry
level tasks that have been replaced by AI

7365
07:09:48.375 --> 07:09:54.297
<v CM Virginia Maloney>versus reskilled by AI and the path
from junior to senior and the role or

7366
07:09:54.297 --> 07:09:59.310
<v CM Virginia Maloney>responsibility that the city plays in in
helping people through the AI transition

7367
07:09:59.310 --> 07:10:00.750
<v CM Virginia Maloney>that's coming?

7368
07:10:02.750 --> 07:10:04.670
<v CM Virginia Maloney>And anyone who wants to speak.

7369
07:10:05.070 --> 07:10:09.070
<v Jukay Hsu (Bright Futures)>Thank you. Yeah. Thank you. Thank you,
council member. Yeah. Maybe I can speak

7370
07:10:09.070 --> 07:10:11.070
<v Jukay Hsu (Bright Futures)>quickly to to both things. I mean,

7371
07:10:11.791 --> 07:10:12.590
<v Jukay Hsu (Bright Futures)>on the second point,

7372
07:10:13.395 --> 07:10:16.355
<v Jukay Hsu (Bright Futures)>yeah, reskilling is so important
as we heard from everyone.

7373
07:10:17.314 --> 07:10:22.115
<v Jukay Hsu (Bright Futures)>The city, I I hope, plays a critical role
here, and I hope, you and others will

7374
07:10:22.115 --> 07:10:25.795
<v Jukay Hsu (Bright Futures)>continue to kind of address that. It's
so challenging right now. We don't know.

7375
07:10:25.795 --> 07:10:29.980
<v Jukay Hsu (Bright Futures)>There's recent reports on how it's affected
so many entry level jobs, but that's

7376
07:10:29.980 --> 07:10:31.420
<v Jukay Hsu (Bright Futures)>gonna continue to change.

7377
07:10:32.379 --> 07:10:36.860
<v Jukay Hsu (Bright Futures)>It's not obvious what it is, and
hence, this might be unsatisfactory,

7378
07:10:36.860 --> 07:10:41.580
<v Jukay Hsu (Bright Futures)>but I hope just continued monitoring and
understanding and engaging is so important.

7379
07:10:41.740 --> 07:10:43.980
<v Jukay Hsu (Bright Futures)>But I think there are
some obvious things like

7380
07:10:45.055 --> 07:10:49.295
<v Jukay Hsu (Bright Futures)>autonomous vehicles will be coming to New
York. I don't I don't know if that's one

7381
07:10:49.295 --> 07:10:52.977
<v Jukay Hsu (Bright Futures)>year, five years, or ten years.
It's safer. People trust it.

7382
07:10:53.535 --> 07:10:58.060
<v Jukay Hsu (Bright Futures)>And so thinking about something like that,
can New York City take bold leadership

7383
07:10:58.060 --> 07:11:00.460
<v Jukay Hsu (Bright Futures)>around what's gonna happen to our

7384
07:11:01.180 --> 07:11:05.260
<v Jukay Hsu (Bright Futures)>200,000 drivers and couriers so that they
get reskilled and transitioned in the

7385
07:11:05.260 --> 07:11:05.820
<v Jukay Hsu (Bright Futures)>future?

7386
07:11:06.541 --> 07:11:10.035
<v Jukay Hsu (Bright Futures)>And I think there's so many opportunities
to do that where the answer... I mean,

7387
07:11:10.035 --> 07:11:13.557
<v Jukay Hsu (Bright Futures)>there's an obvious thing that's gonna
happen. So how do we tackle that?

7388
07:11:13.717 --> 07:11:17.076
<v Jukay Hsu (Bright Futures)>And then the first, yeah, I I do hope. So
many technology companies here, I think

7389
07:11:17.076 --> 07:11:21.477
<v Jukay Hsu (Bright Futures)>that's how we build great stuff. That's
how we have growth and safety. Alignment is

7390
07:11:21.477 --> 07:11:23.316
<v Jukay Hsu (Bright Futures)>so important as we all agree here. So

7391
07:11:23.800 --> 07:11:27.720
<v Jukay Hsu (Bright Futures)>I think part of it's a tone with all of
you. People wanna live and build in New

7392
07:11:27.720 --> 07:11:28.760
<v Jukay Hsu (Bright Futures)>York, and so

7393
07:11:29.080 --> 07:11:33.160
<v Jukay Hsu (Bright Futures)>I I I hope there's other policies you all
can consider as well. I think what mayor

7394
07:11:33.160 --> 07:11:37.800
<v Jukay Hsu (Bright Futures)>Bloomberg did such a great job of tracking
is to make New York feel like a place

7395
07:11:37.800 --> 07:11:42.576
<v Jukay Hsu (Bright Futures)>where tech companies can grow. Just showing
up, talking about it, welcoming it. So,

7396
07:11:42.975 --> 07:11:45.215
<v Jukay Hsu (Bright Futures)>I hope we can do both things, and you all,

7397
07:11:45.455 --> 07:11:49.695
<v Jukay Hsu (Bright Futures)>both do that working with the companies,
new companies, and also think about

7398
07:11:49.695 --> 07:11:54.096
<v Jukay Hsu (Bright Futures)>alignment and work with them to address
this because we need the industry also to

7399
07:11:54.096 --> 07:11:54.654
<v Jukay Hsu (Bright Futures)>address this.

7400
07:11:56.510 --> 07:12:01.070
<v Julia Stoyanovich (NYU Center for Responsible AI)>If I may add to this. So the
the deputy commissioner of OTI,

7401
07:12:01.070 --> 07:12:05.470
<v Julia Stoyanovich (NYU Center for Responsible AI)>when she spoke, she uttered the
phrase that I find really upsetting,

7402
07:12:05.630 --> 07:12:09.310
<v Julia Stoyanovich (NYU Center for Responsible AI)>and that was... She said that we need
effective and responsible innovation.

7403
07:12:10.186 --> 07:12:13.785
<v Julia Stoyanovich (NYU Center for Responsible AI)>And, the effective part of
this would cover the wins

7404
07:12:13.945 --> 07:12:14.746
<v Julia Stoyanovich (NYU Center for Responsible AI)>for us.

7405
07:12:15.545 --> 07:12:19.865
<v Julia Stoyanovich (NYU Center for Responsible AI)>But that is really not the kind of a position
that New York City should be taking.

7406
07:12:20.186 --> 07:12:24.186
<v Julia Stoyanovich (NYU Center for Responsible AI)>Our position should be that the wins
should come from responsible innovation,

7407
07:12:24.870 --> 07:12:26.549
<v Julia Stoyanovich (NYU Center for Responsible AI)>And this is our angle here.

7408
07:12:26.790 --> 07:12:29.590
<v Julia Stoyanovich (NYU Center for Responsible AI)>And this, I think, is the kind of industry

7409
07:12:30.068 --> 07:12:34.790
<v Julia Stoyanovich (NYU Center for Responsible AI)>that we can and should build. We really
are uniquely positioned as opposed to

7410
07:12:35.350 --> 07:12:40.385
<v Julia Stoyanovich (NYU Center for Responsible AI)>San Francisco to do this. Right? And we
always have been leading in at least

7411
07:12:40.385 --> 07:12:45.025
<v Julia Stoyanovich (NYU Center for Responsible AI)>proposing regulatory instruments
for automated decision systems, AI.

7412
07:12:45.586 --> 07:12:48.945
<v Julia Stoyanovich (NYU Center for Responsible AI)>For us to actually develop an industry,
we need to understand that we have to

7413
07:12:48.945 --> 07:12:49.586
<v Julia Stoyanovich (NYU Center for Responsible AI)>invest.

7414
07:12:50.145 --> 07:12:54.680
<v Julia Stoyanovich (NYU Center for Responsible AI)>Right? And, you know, following up on what
one of the people at the table here said,

7415
07:12:54.680 --> 07:12:59.160
<v Julia Stoyanovich (NYU Center for Responsible AI)>we have a pretty substantial amount of
money here at our disposal that we could put

7416
07:12:59.160 --> 07:13:01.400
<v Julia Stoyanovich (NYU Center for Responsible AI)>towards responsible innovation,

7417
07:13:01.400 --> 07:13:02.680
<v Julia Stoyanovich (NYU Center for Responsible AI)>towards upskilling,

7418
07:13:02.920 --> 07:13:07.480
<v Julia Stoyanovich (NYU Center for Responsible AI)>towards making sure that the public's
literacy is raised, right, so that we can all

7419
07:13:07.596 --> 07:13:10.156
<v Julia Stoyanovich (NYU Center for Responsible AI)>regulate and oversee the space together.

7420
07:13:10.477 --> 07:13:14.316
<v Julia Stoyanovich (NYU Center for Responsible AI)>So, really, New York City should should
invest in responsible innovation,

7421
07:13:14.555 --> 07:13:16.316
<v Julia Stoyanovich (NYU Center for Responsible AI)>upskilling literacy,

7422
07:13:16.555 --> 07:13:17.996
<v Julia Stoyanovich (NYU Center for Responsible AI)>and building enforcement

7423
07:13:17.996 --> 07:13:21.516
<v Julia Stoyanovich (NYU Center for Responsible AI)>capacity and oversight capacity
in house within the government.

7424
07:13:24.021 --> 07:13:25.140
<v Anna Myers (Who Decides)>Yeah. I think

7425
07:13:25.861 --> 07:13:30.580
<v Anna Myers (Who Decides)>New York is a tech city, and we need to
resist the urge to compare that we're not

7426
07:13:30.580 --> 07:13:34.262
<v Anna Myers (Who Decides)>that. I used to work in VC that
invested in enterprise tech.

7427
07:13:34.740 --> 07:13:40.055
<v Anna Myers (Who Decides)>We are a growing tech industry, and the
labs, those four companies are here working.

7428
07:13:40.055 --> 07:13:41.414
<v Anna Myers (Who Decides)>They will grow slowly.

7429
07:13:42.615 --> 07:13:46.455
<v Anna Myers (Who Decides)>It's an affordability issue as well, and
we can't narrow it to just that we're not

7430
07:13:46.455 --> 07:13:47.574
<v Anna Myers (Who Decides)>tech welcoming.

7431
07:13:47.734 --> 07:13:51.014
<v Anna Myers (Who Decides)>And I think the mayor is doing a great
job at that, and the city has a lot of

7432
07:13:51.014 --> 07:13:51.574
<v Anna Myers (Who Decides)>opportunity

7433
07:13:51.870 --> 07:13:56.590
<v Anna Myers (Who Decides)>to focus on the whole conversation and
not just we don't see enough tech people

7434
07:13:56.590 --> 07:14:00.270
<v Anna Myers (Who Decides)>moving here. This is a great tech
city, and tech is only growing here.

7435
07:14:05.004 --> 07:14:09.404
<v CM Carmen De La Rosa>Thank you all. Thank you all. And I wanna
thank this panel for your expertise. And

7436
07:14:09.404 --> 07:14:13.404
<v CM Carmen De La Rosa>this is not the end of the conversation.
This is the beginning. The speaker has

7437
07:14:13.404 --> 07:14:17.340
<v CM Carmen De La Rosa>tasked us to continue to hold
hearings with the different

7438
07:14:17.340 --> 07:14:18.060
<v CM Carmen De La Rosa>committees

7439
07:14:18.299 --> 07:14:20.060
<v CM Carmen De La Rosa>on AI in all its,

7440
07:14:20.540 --> 07:14:23.580
<v CM Carmen De La Rosa>in all of its sort of expansiveness.

7441
07:14:23.580 --> 07:14:25.980
<v CM Carmen De La Rosa>So thank you so much for being here.

7442
07:14:27.020 --> 07:14:31.895
<v CM Carmen De La Rosa>I wanna call up the next panel. As I 've...
As we do this transition, I also wanna

7443
07:14:31.895 --> 07:14:38.215
<v CM Carmen De La Rosa>remind anyone who has not signed up to to
speak to please fill out a card with the

7444
07:14:38.215 --> 07:14:43.414
<v CM Carmen De La Rosa>sergeant at arms. In this panel, we have
our state colleagues and our colleagues

7445
07:14:43.700 --> 07:14:44.898
<v CM Carmen De La Rosa>from around the world,

7446
07:14:45.459 --> 07:14:48.020
<v CM Carmen De La Rosa>testifying on this panel. We have,

7447
07:14:48.260 --> 07:14:50.180
<v CM Carmen De La Rosa>the honorable Jess Asato,

7448
07:14:50.500 --> 07:14:55.059
<v CM Carmen De La Rosa>who is a UK parliament member and has flown
in specifically for this hearing, so we

7449
07:14:55.059 --> 07:14:58.676
<v CM Carmen De La Rosa>wanna thank her for being
here. We have Alex Boris,

7450
07:14:59.076 --> 07:15:01.396
<v CM Carmen De La Rosa>current assembly member. We have,

7451
07:15:01.637 --> 07:15:02.916
<v CM Carmen De La Rosa>Andrew Gonardis,

7452
07:15:02.916 --> 07:15:03.557
<v CM Carmen De La Rosa>senator.

7453
07:15:03.797 --> 07:15:05.396
<v CM Carmen De La Rosa>We have Kristen Gonzales,

7454
07:15:05.396 --> 07:15:09.797
<v CM Carmen De La Rosa>senator. We have Clive Bunnell, assembly
member, and we have Jordan Wright, assembly

7455
07:15:09.797 --> 07:15:13.648
<v CM Carmen De La Rosa>member. We're gonna begin this
panel with, senator Gonzales

7456
07:15:15.170 --> 07:15:16.610
<v CM Carmen De La Rosa>whenever you're ready,

7457
07:15:17.010 --> 07:15:18.770
<v CM Carmen De La Rosa>and thank you for your patience.

7458
07:15:20.370 --> 07:15:24.450
<v CM Carmen De La Rosa>Let me turn this on. We Honorable Jess
Asado, you could also join this panel,

7459
07:15:24.450 --> 07:15:29.035
<v CM Carmen De La Rosa>please, and thank you for being here and
flying all the way in from The UK. Thank

7460
07:15:29.035 --> 07:15:29.516
<v CM Carmen De La Rosa>you.

7461
07:15:32.395 --> 07:15:37.115
<v Kristen Gonzalez (NY State Senator, SD 59)>Great. Thank you so much. I wanna thank
speaker Menon and the New York City Council

7462
07:15:37.115 --> 07:15:42.300
<v Kristen Gonzalez (NY State Senator, SD 59)>for your attention to a, if not the,
major policy issue of our time.

7463
07:15:42.860 --> 07:15:47.340
<v Kristen Gonzalez (NY State Senator, SD 59)>I am New York state senator Kristen Gonzalez
representing the 59th Senate District

7464
07:15:47.340 --> 07:15:50.541
<v Kristen Gonzalez (NY State Senator, SD 59)>in Midtown Manhattan, Northern
Brooklyn, and Western Queens.

7465
07:15:50.860 --> 07:15:54.300
<v Kristen Gonzalez (NY State Senator, SD 59)>I am the chair of two state senate
committees, elections and Internet and

7466
07:15:54.300 --> 07:15:54.700
<v Kristen Gonzalez (NY State Senator, SD 59)>technology.

7467
07:15:55.365 --> 07:15:59.766
<v Kristen Gonzalez (NY State Senator, SD 59)>Over the last four years as chair of the
state senate technology committee and as a

7468
07:15:59.766 --> 07:16:00.965
<v Kristen Gonzalez (NY State Senator, SD 59)>former tech worker,

7469
07:16:01.125 --> 07:16:05.285
<v Kristen Gonzalez (NY State Senator, SD 59)>I've led on the issue of AI in the legislature
and have passed some of the state's

7470
07:16:05.285 --> 07:16:07.285
<v Kristen Gonzalez (NY State Senator, SD 59)>first laws on generative AI,

7471
07:16:07.686 --> 07:16:09.525
<v Kristen Gonzalez (NY State Senator, SD 59)>including regulating
deepfakes in elections,

7472
07:16:10.080 --> 07:16:13.281
<v Kristen Gonzalez (NY State Senator, SD 59)>government use of AI, protecting
workers from automation,

7473
07:16:13.440 --> 07:16:15.840
<v Kristen Gonzalez (NY State Senator, SD 59)>adding warning labels on chatbots,

7474
07:16:16.000 --> 07:16:18.400
<v Kristen Gonzalez (NY State Senator, SD 59)>responsible data center
development, and more.

7475
07:16:19.120 --> 07:16:20.480
<v Kristen Gonzalez (NY State Senator, SD 59)>I want to be clear.

7476
07:16:20.800 --> 07:16:22.880
<v Kristen Gonzalez (NY State Senator, SD 59)>AI is not inevitable.

7477
07:16:24.236 --> 07:16:29.596
<v Kristen Gonzalez (NY State Senator, SD 59)>We do not have to accept a vision for
innovation driven by the very same tech

7478
07:16:29.596 --> 07:16:30.477
<v Kristen Gonzalez (NY State Senator, SD 59)>oligarchs

7479
07:16:30.477 --> 07:16:34.717
<v Kristen Gonzalez (NY State Senator, SD 59)>who have raised alarm that their models
could lead to catastrophic events.

7480
07:16:35.756 --> 07:16:40.370
<v Kristen Gonzalez (NY State Senator, SD 59)>And we certainly can't trust the same people
who have rushed to roll out imperfect

7481
07:16:40.370 --> 07:16:42.930
<v Kristen Gonzalez (NY State Senator, SD 59)>AI tools to regulate themselves.

7482
07:16:44.209 --> 07:16:47.010
<v Kristen Gonzalez (NY State Senator, SD 59)>As government bodies, we
have the responsibility

7483
07:16:47.010 --> 07:16:51.330
<v Kristen Gonzalez (NY State Senator, SD 59)>to set guardrails on this new technology
and the power to establish regulations

7484
07:16:51.330 --> 07:16:55.936
<v Kristen Gonzalez (NY State Senator, SD 59)>that protect workers, consumers, and
members of the public from potential danger.

7485
07:16:56.494 --> 07:16:57.295
<v Kristen Gonzalez (NY State Senator, SD 59)>Interventions

7486
07:16:57.295 --> 07:17:01.936
<v Kristen Gonzalez (NY State Senator, SD 59)>at every level of government are needed to
ensure AI is developed in a responsible

7487
07:17:01.936 --> 07:17:05.215
<v Kristen Gonzalez (NY State Senator, SD 59)>way that benefits the public
and not just the billionaires.

7488
07:17:06.480 --> 07:17:10.959
<v Kristen Gonzalez (NY State Senator, SD 59)>Responsibly regulating AI absolutely
means guardrails on frontier models and

7489
07:17:10.959 --> 07:17:11.920
<v Kristen Gonzalez (NY State Senator, SD 59)>superintelligence

7490
07:17:11.920 --> 07:17:13.840
<v Kristen Gonzalez (NY State Senator, SD 59)>that pose significant threats.

7491
07:17:14.320 --> 07:17:15.041
<v Kristen Gonzalez (NY State Senator, SD 59)>However,

7492
07:17:15.360 --> 07:17:17.920
<v Kristen Gonzalez (NY State Senator, SD 59)>meaningfully reining in these technologies

7493
07:17:17.920 --> 07:17:22.484
<v Kristen Gonzalez (NY State Senator, SD 59)>to protect New Yorkers means regulating
the AI powered tools these companies are

7494
07:17:22.484 --> 07:17:27.605
<v Kristen Gonzalez (NY State Senator, SD 59)>rushing to embed into every part of our
lives with little oversight and little

7495
07:17:27.605 --> 07:17:28.486
<v Kristen Gonzalez (NY State Senator, SD 59)>accountability.

7496
07:17:29.525 --> 07:17:32.404
<v Kristen Gonzalez (NY State Senator, SD 59)>We shouldn't have to wait for
future harms to take action

7497
07:17:32.690 --> 07:17:37.250
<v Kristen Gonzalez (NY State Senator, SD 59)>because the harms of imperfect AI tools
are already being felt across the country.

7498
07:17:37.729 --> 07:17:41.810
<v Kristen Gonzalez (NY State Senator, SD 59)>For example, uses for mass surveillance
of immigrant communities and workers,

7499
07:17:41.969 --> 07:17:46.530
<v Kristen Gonzalez (NY State Senator, SD 59)>chatbots creating intimate partner
relationships with children and adults going as

7500
07:17:46.530 --> 07:17:51.795
<v Kristen Gonzalez (NY State Senator, SD 59)>far as to encourage self harm, and documented
cases of algorithmic bias in decision

7501
07:17:51.795 --> 07:17:55.076
<v Kristen Gonzalez (NY State Senator, SD 59)>making such as in hiring,
health care, and housing.

7502
07:17:55.875 --> 07:18:00.195
<v Kristen Gonzalez (NY State Senator, SD 59)>I believe we have the power to
set the terms of innovation

7503
07:18:00.480 --> 07:18:04.720
<v Kristen Gonzalez (NY State Senator, SD 59)>so that this technology is used
for the maximum public good.

7504
07:18:05.281 --> 07:18:10.400
<v Kristen Gonzalez (NY State Senator, SD 59)>That is why our state legislation has
prioritized a rights based approach to AI.

7505
07:18:10.880 --> 07:18:12.640
<v Kristen Gonzalez (NY State Senator, SD 59)>We should have the fundamental

7506
07:18:12.640 --> 07:18:13.600
<v Kristen Gonzalez (NY State Senator, SD 59)>right to privacy,

7507
07:18:14.045 --> 07:18:16.766
<v Kristen Gonzalez (NY State Senator, SD 59)>which is why I've introduced
the New York Privacy Act.

7508
07:18:17.244 --> 07:18:20.285
<v Kristen Gonzalez (NY State Senator, SD 59)>We should have a right to
protections in the workplace,

7509
07:18:20.445 --> 07:18:24.205
<v Kristen Gonzalez (NY State Senator, SD 59)>which is why I've introduced the BOD Act
to protect from... To protect workers from

7510
07:18:24.205 --> 07:18:25.404
<v Kristen Gonzalez (NY State Senator, SD 59)>workplace surveillance.

7511
07:18:25.965 --> 07:18:30.870
<v Kristen Gonzalez (NY State Senator, SD 59)>We should have the right to know when AI is
making high risk consequential decisions

7512
07:18:30.870 --> 07:18:34.070
<v Kristen Gonzalez (NY State Senator, SD 59)>about our lives and subjecting
us to algorithmic bias,

7513
07:18:34.230 --> 07:18:38.709
<v Kristen Gonzalez (NY State Senator, SD 59)>which is why I introduced the New York
AI act. This bill requires third party

7514
07:18:38.709 --> 07:18:42.646
<v Kristen Gonzalez (NY State Senator, SD 59)>audits, compliance with international
safety standards, disclosures,

7515
07:18:42.646 --> 07:18:43.445
<v Kristen Gonzalez (NY State Senator, SD 59)>transparency,

7516
07:18:43.525 --> 07:18:48.166
<v Kristen Gonzalez (NY State Senator, SD 59)>and importantly, for a human to be in the
loop in consequential decision making,

7517
07:18:48.566 --> 07:18:50.406
<v Kristen Gonzalez (NY State Senator, SD 59)>much of which we've heard about today.

7518
07:18:51.205 --> 07:18:52.006
<v Kristen Gonzalez (NY State Senator, SD 59)>Finally,

7519
07:18:52.166 --> 07:18:55.890
<v Kristen Gonzalez (NY State Senator, SD 59)>two of my bills sitting on the governor's
desk this year include the right to keep

7520
07:18:55.890 --> 07:18:58.130
<v Kristen Gonzalez (NY State Senator, SD 59)>our kids safe online from chatbots

7521
07:18:58.291 --> 07:19:02.932
<v Kristen Gonzalez (NY State Senator, SD 59)>and the right to protect all of us from
the build out of massive data centers.

7522
07:19:03.730 --> 07:19:04.691
<v Kristen Gonzalez (NY State Senator, SD 59)>Our bill,

7523
07:19:05.170 --> 07:19:10.215
<v Kristen Gonzalez (NY State Senator, SD 59)>s nine zero five one b, bans unsafe
features for kids and passed unanimously

7524
07:19:10.215 --> 07:19:12.855
<v Kristen Gonzalez (NY State Senator, SD 59)>in both chambers of the legislature.

7525
07:19:13.654 --> 07:19:17.336
<v Kristen Gonzalez (NY State Senator, SD 59)>Just like with that bill, New Yorkers are
united in the fight against hyperscale

7526
07:19:17.336 --> 07:19:18.295
<v Kristen Gonzalez (NY State Senator, SD 59)>data centers.

7527
07:19:18.535 --> 07:19:22.055
<v Kristen Gonzalez (NY State Senator, SD 59)>At the end of August, I embarked on a
statewide tour of communities that are

7528
07:19:22.055 --> 07:19:26.880
<v Kristen Gonzalez (NY State Senator, SD 59)>affected by these data centers, and I went
to 13 different communities facing major

7529
07:19:26.880 --> 07:19:30.080
<v Kristen Gonzalez (NY State Senator, SD 59)>noise pollution, health impacts,
environmental degradation,

7530
07:19:30.160 --> 07:19:31.920
<v Kristen Gonzalez (NY State Senator, SD 59)>and threats to their water supplies.

7531
07:19:32.400 --> 07:19:36.400
<v Kristen Gonzalez (NY State Senator, SD 59)>I was proud to work with the governor on
her first emanation moratorium on the

7532
07:19:36.400 --> 07:19:39.836
<v Kristen Gonzalez (NY State Senator, SD 59)>largest data centers, and I'm now working
to get the responsible data center

7533
07:19:39.836 --> 07:19:41.756
<v Kristen Gonzalez (NY State Senator, SD 59)>development act signed into law.

7534
07:19:42.555 --> 07:19:45.275
<v Kristen Gonzalez (NY State Senator, SD 59)>While the tech CEOs say
they want regulations,

7535
07:19:45.516 --> 07:19:49.996
<v Kristen Gonzalez (NY State Senator, SD 59)>they have spent their time in Albany
fighting against each of the common sense

7536
07:19:49.996 --> 07:19:51.115
<v Kristen Gonzalez (NY State Senator, SD 59)>guardrails I have mentioned.

7537
07:19:51.830 --> 07:19:55.031
<v Kristen Gonzalez (NY State Senator, SD 59)>For the lip service they pay
to supporting legislation,

7538
07:19:55.110 --> 07:19:58.791
<v Kristen Gonzalez (NY State Senator, SD 59)>it is only ever for the most
watered down version of a bill.

7539
07:19:59.510 --> 07:20:03.830
<v Kristen Gonzalez (NY State Senator, SD 59)>Together, we have the ability to meet
the moment, pass real legislation,

7540
07:20:03.830 --> 07:20:07.725
<v Kristen Gonzalez (NY State Senator, SD 59)>protect New Yorkers, and build a future
that allows us to benefit from innovation

7541
07:20:07.725 --> 07:20:11.805
<v Kristen Gonzalez (NY State Senator, SD 59)>while protecting New Yorkers from real
harm. Thank you to the council again today

7542
07:20:11.805 --> 07:20:15.484
<v Kristen Gonzalez (NY State Senator, SD 59)>for our testimony. Thank
you. Thank you, senator.

7543
07:20:15.564 --> 07:20:16.365
<v CM Carmen De La Rosa>You

7544
07:20:17.725 --> 07:20:19.004
<v CM Carmen De La Rosa>can you can continue,

7545
07:20:19.800 --> 07:20:21.480
<v CM Carmen De La Rosa>whoever wants to go next.

7546
07:20:25.400 --> 07:20:26.281
<v Clyde Vanel (NY State Assembly Member, AD 33)>Good evening.

7547
07:20:26.682 --> 07:20:28.521
<v Clyde Vanel (NY State Assembly Member, AD 33)>Good evening, madam speaker.

7548
07:20:28.762 --> 07:20:30.920
<v Clyde Vanel (NY State Assembly Member, AD 33)>And also good evening, to the,

7549
07:20:31.400 --> 07:20:32.920
<v Clyde Vanel (NY State Assembly Member, AD 33)>to the, chair,

7550
07:20:33.240 --> 07:20:34.201
<v Clyde Vanel (NY State Assembly Member, AD 33)>Carmen de la Rosa,

7551
07:20:34.805 --> 07:20:40.244
<v Clyde Vanel (NY State Assembly Member, AD 33)>and to my, council member, deputy speaker
doctor Nantasha Williams, is the sponsor

7552
07:20:40.244 --> 07:20:41.365
<v Clyde Vanel (NY State Assembly Member, AD 33)>of resolution

7553
07:20:41.365 --> 07:20:42.645
<v Clyde Vanel (NY State Assembly Member, AD 33)>one seventy five,

7554
07:20:42.965 --> 07:20:46.965
<v Clyde Vanel (NY State Assembly Member, AD 33)>the the the resolution supporting the
bill high risk advanced artificial

7555
07:20:46.965 --> 07:20:47.764
<v Clyde Vanel (NY State Assembly Member, AD 33)>intelligence act.

7556
07:20:48.520 --> 07:20:52.120
<v Clyde Vanel (NY State Assembly Member, AD 33)>Artificial intelligence, used
properly, is a net good.

7557
07:20:52.440 --> 07:20:54.680
<v Clyde Vanel (NY State Assembly Member, AD 33)>Education, health care, business,

7558
07:20:54.840 --> 07:20:56.040
<v Clyde Vanel (NY State Assembly Member, AD 33)>these are great uses.

7559
07:20:56.600 --> 07:21:01.158
<v Clyde Vanel (NY State Assembly Member, AD 33)>But talking about them while ignoring
the risk, that AI could actually kill and

7560
07:21:01.158 --> 07:21:03.879
<v Clyde Vanel (NY State Assembly Member, AD 33)>strike us, it could...
To me, seems absurd.

7561
07:21:04.625 --> 07:21:06.545
<v Clyde Vanel (NY State Assembly Member, AD 33)>Companies are racing towards,

7562
07:21:07.184 --> 07:21:08.064
<v Clyde Vanel (NY State Assembly Member, AD 33)>artificial,

7563
07:21:08.305 --> 07:21:10.625
<v Clyde Vanel (NY State Assembly Member, AD 33)>towards general artificial intelligence,

7564
07:21:10.865 --> 07:21:14.545
<v Clyde Vanel (NY State Assembly Member, AD 33)>and some of their systems are already
too dangerous to release publicly.

7565
07:21:15.025 --> 07:21:16.305
<v Clyde Vanel (NY State Assembly Member, AD 33)>This is by their own admission.

7566
07:21:17.310 --> 07:21:19.150
<v Clyde Vanel (NY State Assembly Member, AD 33)>We saw this coming. So by...

7567
07:21:19.390 --> 07:21:23.550
<v Clyde Vanel (NY State Assembly Member, AD 33)>Assembly bill three three five
six, the artificial intelligence,

7568
07:21:23.791 --> 07:21:25.390
<v Clyde Vanel (NY State Assembly Member, AD 33)>act a few years ago,

7569
07:21:25.630 --> 07:21:30.430
<v Clyde Vanel (NY State Assembly Member, AD 33)>the idea is simple. The more harm that an
AI system can do, the more accountability

7570
07:21:30.430 --> 07:21:31.230
<v Clyde Vanel (NY State Assembly Member, AD 33)>that it should carry.

7571
07:21:31.926 --> 07:21:35.445
<v Clyde Vanel (NY State Assembly Member, AD 33)>The bill identifies high risk
artificial intelligence systems.

7572
07:21:35.766 --> 07:21:37.125
<v Clyde Vanel (NY State Assembly Member, AD 33)>AI AI

7573
07:21:37.125 --> 07:21:38.807
<v Clyde Vanel (NY State Assembly Member, AD 33)>that diagnoses patients,

7574
07:21:39.045 --> 07:21:44.086
<v Clyde Vanel (NY State Assembly Member, AD 33)>that runs our power grid, or helps police
or judges make decisions, and far more, it

7575
07:21:44.086 --> 07:21:47.686
<v Clyde Vanel (NY State Assembly Member, AD 33)>requires them to be licensed
before operating in New York.

7576
07:21:48.380 --> 07:21:51.740
<v Clyde Vanel (NY State Assembly Member, AD 33)>Under the act, the companies
must report malfunctions

7577
07:21:51.740 --> 07:21:53.740
<v Clyde Vanel (NY State Assembly Member, AD 33)>and hacks like the recent,

7578
07:21:54.060 --> 07:21:55.580
<v Clyde Vanel (NY State Assembly Member, AD 33)>hugging face case.

7579
07:21:55.980 --> 07:22:01.100
<v Clyde Vanel (NY State Assembly Member, AD 33)>It must get, state sign off before major
changes and set up an independent ethics

7580
07:22:01.100 --> 07:22:04.336
<v Clyde Vanel (NY State Assembly Member, AD 33)>risk board that reports
to the state annually.

7581
07:22:04.977 --> 07:22:08.656
<v Clyde Vanel (NY State Assembly Member, AD 33)>Finally, it writes a binding
ethical code into the law.

7582
07:22:09.295 --> 07:22:14.016
<v Clyde Vanel (NY State Assembly Member, AD 33)>New York took its first step with the
RAISE Act, and I'm, and I I commend our

7583
07:22:14.016 --> 07:22:16.016
<v Clyde Vanel (NY State Assembly Member, AD 33)>colleagues for... And and
the state for passing it.

7584
07:22:16.680 --> 07:22:18.360
<v Clyde Vanel (NY State Assembly Member, AD 33)>Now... But now none of this,

7585
07:22:18.920 --> 07:22:22.440
<v Clyde Vanel (NY State Assembly Member, AD 33)>none none of this that we're doing
is meant to stop, innovation.

7586
07:22:22.600 --> 07:22:27.080
<v Clyde Vanel (NY State Assembly Member, AD 33)>We can't know that AI will be capable...
What it will be capable of in a few years.

7587
07:22:27.398 --> 07:22:30.600
<v Clyde Vanel (NY State Assembly Member, AD 33)>But for tomorrow and for
today, we must decide properly

7588
07:22:30.920 --> 07:22:34.896
<v Clyde Vanel (NY State Assembly Member, AD 33)>the the path forward. I'd like to thank
the council for putting, this this hearing

7589
07:22:34.896 --> 07:22:37.297
<v Clyde Vanel (NY State Assembly Member, AD 33)>together, and we must work
on this together. Thank you.

7590
07:22:37.936 --> 07:22:39.057
<v CM Carmen De La Rosa>Thank you so much.

7591
07:22:40.336 --> 07:22:41.377
<v Jordan Wright (NY State Assembly Member, AD 70)>Good evening.

7592
07:22:42.016 --> 07:22:45.057
<v Jordan Wright (NY State Assembly Member, AD 70)>Thank you, speaker Menon. Thank you to
all the members of the council. Thank you,

7593
07:22:45.057 --> 07:22:48.871
<v Jordan Wright (NY State Assembly Member, AD 70)>chair De La Rosa. I'm New York state assembly
member Jordan Wright representing the

7594
07:22:48.871 --> 07:22:51.510
<v Jordan Wright (NY State Assembly Member, AD 70)>70th Assembly District, which is Harlem.

7595
07:22:51.830 --> 07:22:54.871
<v Jordan Wright (NY State Assembly Member, AD 70)>One of the things I've been focused on in
Albany is pretty simple. Who is checking

7596
07:22:54.871 --> 07:22:58.951
<v Jordan Wright (NY State Assembly Member, AD 70)>on AI companies? I have legislation that
requires certain AI developers to have an

7597
07:22:58.951 --> 07:23:02.984
<v Jordan Wright (NY State Assembly Member, AD 70)>independent third party verify that they're
actually following the safety standards

7598
07:23:02.984 --> 07:23:04.824
<v Jordan Wright (NY State Assembly Member, AD 70)>that they say they are following.

7599
07:23:04.984 --> 07:23:08.424
<v Jordan Wright (NY State Assembly Member, AD 70)>Because we can't have a system where a
company builds the technology, writes the

7600
07:23:08.424 --> 07:23:11.945
<v Jordan Wright (NY State Assembly Member, AD 70)>safety rules, tells us they follow the
rules, and then we just take their word for

7601
07:23:11.945 --> 07:23:15.305
<v Jordan Wright (NY State Assembly Member, AD 70)>it? It doesn't make much sense to me. We
have to make sure that these third party

7602
07:23:15.305 --> 07:23:18.791
<v Jordan Wright (NY State Assembly Member, AD 70)>audits actually mean something. Who are the
auditors? Who are the... Are they really

7603
07:23:18.791 --> 07:23:22.470
<v Jordan Wright (NY State Assembly Member, AD 70)>independent? Do they have conflicts of
interest? And what happens when they find

7604
07:23:22.470 --> 07:23:25.750
<v Jordan Wright (NY State Assembly Member, AD 70)>something wrong? Those are questions that
we're trying to answer at the state level,

7605
07:23:25.750 --> 07:23:29.836
<v Jordan Wright (NY State Assembly Member, AD 70)>and I'm glad the city council is having
this discussion as well. New York should be

7606
07:23:29.836 --> 07:23:32.475
<v Jordan Wright (NY State Assembly Member, AD 70)>a place where AI companies
want to build and innovate,

7607
07:23:32.715 --> 07:23:36.555
<v Jordan Wright (NY State Assembly Member, AD 70)>but the accountability to innovate can exist
at the same time. I wanna thank all the

7608
07:23:36.555 --> 07:23:39.914
<v Jordan Wright (NY State Assembly Member, AD 70)>members of the council for having us today.
I thank you guys for taking this issue

7609
07:23:39.914 --> 07:23:43.195
<v Jordan Wright (NY State Assembly Member, AD 70)>seriously once again, and please enjoy your
evening and be safe. Thank you so much.

7610
07:23:44.510 --> 07:23:45.390
<v CM Carmen De La Rosa>Thank you.

7611
07:23:50.910 --> 07:23:55.469
<v Jess Asato (UK Member of Parliament, Lowestoft)>Good evening, and thank you. My name is
Jess Esato, and I am a member of The United

7612
07:23:55.469 --> 07:23:57.790
<v Jess Asato (UK Member of Parliament, Lowestoft)>Kingdom Parliament for
Lowestoft in Suffolk.

7613
07:23:58.266 --> 07:24:01.785
<v Jess Asato (UK Member of Parliament, Lowestoft)>In January of this year, sexualized
and degrading nonconsensual

7614
07:24:01.785 --> 07:24:06.506
<v Jess Asato (UK Member of Parliament, Lowestoft)>images were created of me
using SpaceX AI's tool, Grok.

7615
07:24:07.064 --> 07:24:09.305
<v Jess Asato (UK Member of Parliament, Lowestoft)>I was one of thousands of victims,

7616
07:24:09.705 --> 07:24:12.426
<v Jess Asato (UK Member of Parliament, Lowestoft)>including victims in New York of Grok.

7617
07:24:13.480 --> 07:24:15.640
<v Jess Asato (UK Member of Parliament, Lowestoft)>Grock is an AI tool

7618
07:24:15.720 --> 07:24:17.240
<v Jess Asato (UK Member of Parliament, Lowestoft)>that was engineered

7619
07:24:17.320 --> 07:24:18.840
<v Jess Asato (UK Member of Parliament, Lowestoft)>to create sexualized

7620
07:24:18.840 --> 07:24:19.640
<v Jess Asato (UK Member of Parliament, Lowestoft)>content

7621
07:24:20.280 --> 07:24:23.320
<v Jess Asato (UK Member of Parliament, Lowestoft)>even if that content had, and I quote,

7622
07:24:23.720 --> 07:24:25.880
<v Jess Asato (UK Member of Parliament, Lowestoft)>dark or violent themes.

7623
07:24:27.346 --> 07:24:30.066
<v Jess Asato (UK Member of Parliament, Lowestoft)>I am seeking remedies against SpaceX

7624
07:24:30.066 --> 07:24:32.307
<v Jess Asato (UK Member of Parliament, Lowestoft)>AI in the English courts

7625
07:24:32.385 --> 07:24:34.945
<v Jess Asato (UK Member of Parliament, Lowestoft)>for the misuse of my private information

7626
07:24:35.346 --> 07:24:36.945
<v Jess Asato (UK Member of Parliament, Lowestoft)>and for contravening

7627
07:24:36.945 --> 07:24:39.025
<v Jess Asato (UK Member of Parliament, Lowestoft)>UK data protection laws.

7628
07:24:40.650 --> 07:24:43.370
<v Jess Asato (UK Member of Parliament, Lowestoft)>I made a long journey of 3,400

7629
07:24:43.370 --> 07:24:48.730
<v Jess Asato (UK Member of Parliament, Lowestoft)>miles to be here because AI needs
to be safe around our world.

7630
07:24:49.770 --> 07:24:51.930
<v Jess Asato (UK Member of Parliament, Lowestoft)>But I note that SpaceX

7631
07:24:51.930 --> 07:24:52.729
<v Jess Asato (UK Member of Parliament, Lowestoft)>AI

7632
07:24:53.305 --> 07:24:57.385
<v Jess Asato (UK Member of Parliament, Lowestoft)>couldn't manage a trip of what I
understand is just twenty minutes

7633
07:24:57.545 --> 07:24:59.545
<v Jess Asato (UK Member of Parliament, Lowestoft)>or four subway stops

7634
07:24:59.625 --> 07:25:00.984
<v Jess Asato (UK Member of Parliament, Lowestoft)>from their offices

7635
07:25:01.225 --> 07:25:02.504
<v Jess Asato (UK Member of Parliament, Lowestoft)>in Manhattan.

7636
07:25:04.504 --> 07:25:05.785
<v Jess Asato (UK Member of Parliament, Lowestoft)>It's a shame

7637
07:25:05.880 --> 07:25:08.682
<v Jess Asato (UK Member of Parliament, Lowestoft)>because the evidence in
my case and other cases

7638
07:25:08.842 --> 07:25:13.082
<v Jess Asato (UK Member of Parliament, Lowestoft)>suggest that Grock was not built with
sufficient safety in its design.

7639
07:25:14.201 --> 07:25:16.521
<v Jess Asato (UK Member of Parliament, Lowestoft)>It was not built to address consent

7640
07:25:16.521 --> 07:25:19.080
<v Jess Asato (UK Member of Parliament, Lowestoft)>of those subjected to sexualized content.

7641
07:25:20.236 --> 07:25:24.156
<v Jess Asato (UK Member of Parliament, Lowestoft)>To the contrary, the evidence in
my claim suggests Grock added

7642
07:25:24.236 --> 07:25:25.836
<v Jess Asato (UK Member of Parliament, Lowestoft)>sexualized content

7643
07:25:25.916 --> 07:25:27.596
<v Jess Asato (UK Member of Parliament, Lowestoft)>without any user

7644
07:25:27.676 --> 07:25:29.195
<v Jess Asato (UK Member of Parliament, Lowestoft)>prompting to do so.

7645
07:25:30.076 --> 07:25:31.516
<v Jess Asato (UK Member of Parliament, Lowestoft)>That is AI

7646
07:25:31.516 --> 07:25:32.717
<v Jess Asato (UK Member of Parliament, Lowestoft)>made wrong,

7647
07:25:32.955 --> 07:25:34.477
<v Jess Asato (UK Member of Parliament, Lowestoft)>not AI gone wrong.

7648
07:25:35.140 --> 07:25:39.781
<v Jess Asato (UK Member of Parliament, Lowestoft)>And as a result, those responsible for
its creation must be held to account.

7649
07:25:41.140 --> 07:25:43.300
<v Jess Asato (UK Member of Parliament, Lowestoft)>No one is above the law

7650
07:25:43.460 --> 07:25:45.460
<v Jess Asato (UK Member of Parliament, Lowestoft)>no matter how rich they are,

7651
07:25:45.941 --> 07:25:47.460
<v Jess Asato (UK Member of Parliament, Lowestoft)>no matter how powerful they are.

7652
07:25:48.445 --> 07:25:53.006
<v Jess Asato (UK Member of Parliament, Lowestoft)>The equal protection of the laws is a founding
principle in the fourteenth amendment

7653
07:25:53.244 --> 07:25:55.965
<v Jess Asato (UK Member of Parliament, Lowestoft)>as it is in the British Magna Carta.

7654
07:25:56.686 --> 07:26:00.285
<v Jess Asato (UK Member of Parliament, Lowestoft)>We were promised that artificial
intelligence could hold the cure to cancer,

7655
07:26:01.100 --> 07:26:03.420
<v Jess Asato (UK Member of Parliament, Lowestoft)>yet it is turning into a cancer,

7656
07:26:03.740 --> 07:26:06.620
<v Jess Asato (UK Member of Parliament, Lowestoft)>which left unchecked has the
potential to harm us all.

7657
07:26:07.420 --> 07:26:10.860
<v Jess Asato (UK Member of Parliament, Lowestoft)>Today, the New York City Council is
bringing forward some important checks and

7658
07:26:10.860 --> 07:26:11.580
<v Jess Asato (UK Member of Parliament, Lowestoft)>balances.

7659
07:26:12.059 --> 07:26:16.994
<v Jess Asato (UK Member of Parliament, Lowestoft)>You do so as responsible legislators at a
time when AI companies are failing to take

7660
07:26:16.994 --> 07:26:17.955
<v Jess Asato (UK Member of Parliament, Lowestoft)>responsibility

7661
07:26:18.033 --> 07:26:20.354
<v Jess Asato (UK Member of Parliament, Lowestoft)>for the design and impact
of their products.

7662
07:26:20.914 --> 07:26:25.635
<v Jess Asato (UK Member of Parliament, Lowestoft)>Until they do and until those designs are
corrected so women and girls are protected

7663
07:26:25.635 --> 07:26:28.354
<v Jess Asato (UK Member of Parliament, Lowestoft)>from the harm that might
claim in others evidences,

7664
07:26:28.594 --> 07:26:29.715
<v Jess Asato (UK Member of Parliament, Lowestoft)>we must play our part.

7665
07:26:30.690 --> 07:26:34.129
<v Jess Asato (UK Member of Parliament, Lowestoft)>This technology does not
respect geographic boundaries.

7666
07:26:34.370 --> 07:26:36.370
<v Jess Asato (UK Member of Parliament, Lowestoft)>And without international cooperation,

7667
07:26:36.610 --> 07:26:40.450
<v Jess Asato (UK Member of Parliament, Lowestoft)>our citizens will be left defenseless
against those who put profit

7668
07:26:40.610 --> 07:26:41.970
<v Jess Asato (UK Member of Parliament, Lowestoft)>before protection.

7669
07:26:42.690 --> 07:26:44.690
<v Jess Asato (UK Member of Parliament, Lowestoft)>So I bring a message of bipartisan

7670
07:26:44.690 --> 07:26:45.410
<v Jess Asato (UK Member of Parliament, Lowestoft)>solidarity.

7671
07:26:46.195 --> 07:26:51.314
<v Jess Asato (UK Member of Parliament, Lowestoft)>I fight my legal action in the name of all
the women and children whose likeness was

7672
07:26:51.314 --> 07:26:56.275
<v Jess Asato (UK Member of Parliament, Lowestoft)>misused by Grok and other AI tools
built without their protection in mind.

7673
07:26:56.756 --> 07:26:58.115
<v Jess Asato (UK Member of Parliament, Lowestoft)>I hope my cause,

7674
07:26:58.195 --> 07:27:02.310
<v Jess Asato (UK Member of Parliament, Lowestoft)>in some way, helps to inform your
proceedings today, and I look forward to your

7675
07:27:02.310 --> 07:27:02.950
<v Jess Asato (UK Member of Parliament, Lowestoft)>questions.

7676
07:27:03.670 --> 07:27:06.310
<v CM Carmen De La Rosa>Thank you so much. Assemblyman Boris.

7677
07:27:09.990 --> 07:27:12.310
<v Alex Bores (NY State Assembly Member, AD 73)>Chair de la Rosa, speaker Menon,

7678
07:27:12.310 --> 07:27:14.709
<v Alex Bores (NY State Assembly Member, AD 73)>members of the city council,
thank you for having us today.

7679
07:27:15.484 --> 07:27:20.283
<v Alex Bores (NY State Assembly Member, AD 73)>I'm the author of the RAISE Act, and I can
definitively say that OpenAI opposed it

7680
07:27:20.283 --> 07:27:22.365
<v Alex Bores (NY State Assembly Member, AD 73)>from the moment it was introduced

7681
07:27:22.443 --> 07:27:24.445
<v Alex Bores (NY State Assembly Member, AD 73)>to the moment it was signed.

7682
07:27:25.244 --> 07:27:26.604
<v Alex Bores (NY State Assembly Member, AD 73)>I am no lawyer,

7683
07:27:26.844 --> 07:27:29.484
<v Alex Bores (NY State Assembly Member, AD 73)>but to me, my common man understanding

7684
07:27:29.840 --> 07:27:32.800
<v Alex Bores (NY State Assembly Member, AD 73)>of lying under oath is that it's perjury.

7685
07:27:32.959 --> 07:27:36.879
<v Alex Bores (NY State Assembly Member, AD 73)>And at the very least, you should be
extremely angry at the company for

7686
07:27:36.879 --> 07:27:37.760
<v Alex Bores (NY State Assembly Member, AD 73)>disrespecting

7687
07:27:37.760 --> 07:27:38.800
<v Alex Bores (NY State Assembly Member, AD 73)>this counsel

7688
07:27:38.879 --> 07:27:41.360
<v Alex Bores (NY State Assembly Member, AD 73)>and saying things they know to be false

7689
07:27:41.600 --> 07:27:42.480
<v Alex Bores (NY State Assembly Member, AD 73)>to you directly.

7690
07:27:43.635 --> 07:27:47.875
<v Alex Bores (NY State Assembly Member, AD 73)>I'm very happy to cooperate with you or
any of your lawyers if you want more

7691
07:27:47.875 --> 07:27:49.955
<v Alex Bores (NY State Assembly Member, AD 73)>information about what they said and when.

7692
07:27:50.994 --> 07:27:51.635
<v Alex Bores (NY State Assembly Member, AD 73)>Second,

7693
07:27:51.955 --> 07:27:55.074
<v Alex Bores (NY State Assembly Member, AD 73)>there have been a discussion of
a lot of different issues today.

7694
07:27:55.395 --> 07:28:01.340
<v Alex Bores (NY State Assembly Member, AD 73)>We can and should solve all of them. In my
time in the assembly, I passed the RAISE

7695
07:28:01.340 --> 07:28:06.460
<v Alex Bores (NY State Assembly Member, AD 73)>Act on catastrophic risk. I also worked with
senator Gonzales on a bill on chatbots.

7696
07:28:06.701 --> 07:28:11.100
<v Alex Bores (NY State Assembly Member, AD 73)>I worked with senator Gennardis on a bill
on content provenance and training data

7697
07:28:11.100 --> 07:28:11.660
<v Alex Bores (NY State Assembly Member, AD 73)>transparency.

7698
07:28:12.166 --> 07:28:16.406
<v Alex Bores (NY State Assembly Member, AD 73)>And I found that the advocates that I worked
with on each sub issue were different,

7699
07:28:16.807 --> 07:28:18.967
<v Alex Bores (NY State Assembly Member, AD 73)>but the opposition was always the same.

7700
07:28:19.445 --> 07:28:24.086
<v Alex Bores (NY State Assembly Member, AD 73)>Don't let the tech companies, don't let
the AI companies divide us and say that in

7701
07:28:24.086 --> 07:28:28.830
<v Alex Bores (NY State Assembly Member, AD 73)>advocating for one issue, you're somehow
harming advocacy for another issue. What we

7702
07:28:28.830 --> 07:28:32.750
<v Alex Bores (NY State Assembly Member, AD 73)>are doing is building the muscle to hold
them accountable and to ensure that it's

7703
07:28:32.750 --> 07:28:36.590
<v Alex Bores (NY State Assembly Member, AD 73)>the American people who decide the future
of this most important technology.

7704
07:28:37.629 --> 07:28:38.270
<v Alex Bores (NY State Assembly Member, AD 73)>Third,

7705
07:28:38.510 --> 07:28:40.988
<v Alex Bores (NY State Assembly Member, AD 73)>the time for voluntary
commitments is over.

7706
07:28:41.625 --> 07:28:46.025
<v Alex Bores (NY State Assembly Member, AD 73)>The voluntary commitments that were signed
at the White House last week largely

7707
07:28:46.025 --> 07:28:49.625
<v Alex Bores (NY State Assembly Member, AD 73)>mirror voluntary commitments
that were signed in 2023

7708
07:28:49.625 --> 07:28:51.145
<v Alex Bores (NY State Assembly Member, AD 73)>in the Biden White House.

7709
07:28:51.545 --> 07:28:55.785
<v Alex Bores (NY State Assembly Member, AD 73)>Promises have been made. It is time
for laws to hold people accountable.

7710
07:28:56.740 --> 07:29:01.060
<v Alex Bores (NY State Assembly Member, AD 73)>And lastly, thank you New York
City for stepping into your power.

7711
07:29:01.300 --> 07:29:05.220
<v Alex Bores (NY State Assembly Member, AD 73)>There's gonna be a lot online from the
tech companies about how they can't comply

7712
07:29:05.220 --> 07:29:08.900
<v Alex Bores (NY State Assembly Member, AD 73)>with different cities doing all these
different things. It's the argument they use

7713
07:29:08.900 --> 07:29:14.514
<v Alex Bores (NY State Assembly Member, AD 73)>every time you try to regulate them. But
New York City is larger than 38 states.

7714
07:29:15.154 --> 07:29:20.033
<v Alex Bores (NY State Assembly Member, AD 73)>All of the companies have large offices
here. It is absolutely within your power,

7715
07:29:20.033 --> 07:29:21.955
<v Alex Bores (NY State Assembly Member, AD 73)>and I dare say your responsibility

7716
07:29:22.193 --> 07:29:23.955
<v Alex Bores (NY State Assembly Member, AD 73)>to take these threats seriously

7717
07:29:24.080 --> 07:29:28.800
<v Alex Bores (NY State Assembly Member, AD 73)>and to help decide the future of this
technology to benefit all of us. So thank you

7718
07:29:28.800 --> 07:29:31.201
<v Alex Bores (NY State Assembly Member, AD 73)>for recognizing that, and
thank you for doing it.

7719
07:29:31.680 --> 07:29:36.400
<v CM Carmen De La Rosa>Thank you all so much. We've also been
joined by senator Gonardis on Zoom.

7720
07:29:36.800 --> 07:29:38.160
<v CM Carmen De La Rosa>Senator, when you're ready.

7721
07:29:39.445 --> 07:29:42.805
<v Andrew Gounardes (NY State Senator, SD 26)>Thank you very much, everyone. I'm sorry
I can't be there with you in person, but

7722
07:29:42.805 --> 07:29:47.445
<v Andrew Gounardes (NY State Senator, SD 26)>two out of my three children are homesick
today. Therefore, I am playing doctor at

7723
07:29:47.445 --> 07:29:47.926
<v Andrew Gounardes (NY State Senator, SD 26)>home.

7724
07:29:48.246 --> 07:29:51.766
<v Andrew Gounardes (NY State Senator, SD 26)>I wanna thank speaker Menon and the members
of the New you wanna say hello here,

7725
07:29:51.766 --> 07:29:51.926
<v Andrew Gounardes (NY State Senator, SD 26)>James?

7726
07:29:52.701 --> 07:29:56.781
<v Andrew Gounardes (NY State Senator, SD 26)>Hello. And the members of the New York City
Council for the opportunity to testify

7727
07:29:56.781 --> 07:30:00.460
<v Andrew Gounardes (NY State Senator, SD 26)>before you. I'm state senator
Andrew Guinardis from Brooklyn,

7728
07:30:00.701 --> 07:30:06.060
<v Andrew Gounardes (NY State Senator, SD 26)>and I am the senate sponsor of the Responsible
AI Safety and Education or RAISE Act.

7729
07:30:06.336 --> 07:30:11.615
<v Andrew Gounardes (NY State Senator, SD 26)>One of only three AI safety laws
passed anywhere in this country.

7730
07:30:12.896 --> 07:30:15.936
<v Andrew Gounardes (NY State Senator, SD 26)>And, you know, I think we've heard a lot
about the RAISE Act already. We've heard a

7731
07:30:15.936 --> 07:30:19.217
<v Andrew Gounardes (NY State Senator, SD 26)>lot about AI safety throughout this this
really remarkable hearing. I've been

7732
07:30:19.217 --> 07:30:23.201
<v Andrew Gounardes (NY State Senator, SD 26)>listening to a good portion of it all day.
I really wanna reflect on the fact that

7733
07:30:23.201 --> 07:30:27.121
<v Andrew Gounardes (NY State Senator, SD 26)>we're celebrating the nation's two hundred
and fiftieth anniversary this year, and I

7734
07:30:27.121 --> 07:30:30.880
<v Andrew Gounardes (NY State Senator, SD 26)>want us to think about a quote that comes
from one of our founding fathers, James

7735
07:30:30.880 --> 07:30:31.682
<v Andrew Gounardes (NY State Senator, SD 26)>Madison,

7736
07:30:32.080 --> 07:30:36.855
<v Andrew Gounardes (NY State Senator, SD 26)>who wrote in Federalist, November fifty one,
if men were angels, no government would

7737
07:30:36.855 --> 07:30:41.414
<v Andrew Gounardes (NY State Senator, SD 26)>be necessary. In other words, government
exists because humans are imperfect and

7738
07:30:41.414 --> 07:30:44.695
<v Andrew Gounardes (NY State Senator, SD 26)>there are certain problems that the free
market or individuals left to their own

7739
07:30:44.695 --> 07:30:45.416
<v Andrew Gounardes (NY State Senator, SD 26)>devices

7740
07:30:45.654 --> 07:30:46.455
<v Andrew Gounardes (NY State Senator, SD 26)>cannot solve.

7741
07:30:47.371 --> 07:30:52.170
<v Andrew Gounardes (NY State Senator, SD 26)>In these past few weeks, we have seen and
heard headline after terrifying headline

7742
07:30:52.170 --> 07:30:54.010
<v Andrew Gounardes (NY State Senator, SD 26)>after terrifying headline

7743
07:30:54.010 --> 07:30:56.970
<v Andrew Gounardes (NY State Senator, SD 26)>about AI agents hacking websites,

7744
07:30:56.970 --> 07:31:00.250
<v Andrew Gounardes (NY State Senator, SD 26)>catastrophic risks of AI
frontier frontier AI models,

7745
07:31:00.570 --> 07:31:05.275
<v Andrew Gounardes (NY State Senator, SD 26)>and the like. And I just keep coming back
to the imperative of Madison's words that

7746
07:31:05.275 --> 07:31:09.994
<v Andrew Gounardes (NY State Senator, SD 26)>men are not angels. And what I think this
tells us and what I think you have heard

7747
07:31:09.994 --> 07:31:14.234
<v Andrew Gounardes (NY State Senator, SD 26)>very clearly today at this
hearing is that we cannot

7748
07:31:14.314 --> 07:31:15.035
<v Andrew Gounardes (NY State Senator, SD 26)>trust

7749
07:31:15.340 --> 07:31:19.180
<v Andrew Gounardes (NY State Senator, SD 26)>the tech companies who
are racing to the bottom

7750
07:31:19.340 --> 07:31:24.540
<v Andrew Gounardes (NY State Senator, SD 26)>to create and perfect a technology that
they themselves say and have proven can

7751
07:31:24.540 --> 07:31:26.540
<v Andrew Gounardes (NY State Senator, SD 26)>cause catastrophic harms

7752
07:31:26.940 --> 07:31:28.140
<v Andrew Gounardes (NY State Senator, SD 26)>to regulate themselves.

7753
07:31:28.795 --> 07:31:32.234
<v Andrew Gounardes (NY State Senator, SD 26)>That we cannot trust that
they will do the right thing

7754
07:31:32.635 --> 07:31:37.436
<v Andrew Gounardes (NY State Senator, SD 26)>because they have proven time and time
and time again that they will always, when

7755
07:31:37.436 --> 07:31:41.115
<v Andrew Gounardes (NY State Senator, SD 26)>confronted with the choice of doing the
right thing or doing what is in the best

7756
07:31:41.115 --> 07:31:46.729
<v Andrew Gounardes (NY State Senator, SD 26)>interest... Times expired.... Bottom line
that they will always choose to pursue the

7757
07:31:46.729 --> 07:31:47.530
<v Andrew Gounardes (NY State Senator, SD 26)>bottom line.

7758
07:31:47.770 --> 07:31:52.410
<v Andrew Gounardes (NY State Senator, SD 26)>And so my message to the council today
is I thank you all for taking on this

7759
07:31:52.410 --> 07:31:53.209
<v Andrew Gounardes (NY State Senator, SD 26)>initiative,

7760
07:31:53.450 --> 07:31:58.426
<v Andrew Gounardes (NY State Senator, SD 26)>for trying to hold these companies accountable.
I echo the comments of my RAISE co

7761
07:31:58.426 --> 07:32:00.105
<v Andrew Gounardes (NY State Senator, SD 26)>sponsor, Asumam Mambores.

7762
07:32:00.105 --> 07:32:04.744
<v Andrew Gounardes (NY State Senator, SD 26)>What we heard from OpenAI today was
certainly news to me that they endorsed and

7763
07:32:04.744 --> 07:32:08.744
<v Andrew Gounardes (NY State Senator, SD 26)>supported the RAISE Act. Certainly not the
RAISE Act that would have prevented them

7764
07:32:08.980 --> 07:32:09.941
<v Andrew Gounardes (NY State Senator, SD 26)>from releasing

7765
07:32:10.100 --> 07:32:14.420
<v Andrew Gounardes (NY State Senator, SD 26)>models that they know... That they knew
were dangerous and could cause catastrophic

7766
07:32:14.420 --> 07:32:18.660
<v Andrew Gounardes (NY State Senator, SD 26)>harms. Certainly, the RAISE Act version
that required third party auditing

7767
07:32:18.660 --> 07:32:19.621
<v Andrew Gounardes (NY State Senator, SD 26)>capabilities

7768
07:32:19.621 --> 07:32:21.140
<v Andrew Gounardes (NY State Senator, SD 26)>or extended liabilities

7769
07:32:21.140 --> 07:32:21.621
<v Andrew Gounardes (NY State Senator, SD 26)>for

7770
07:32:22.195 --> 07:32:26.834
<v Andrew Gounardes (NY State Senator, SD 26)>for harms caused by by their models.
Clearly, none of that was ever on the table

7771
07:32:26.834 --> 07:32:31.234
<v Andrew Gounardes (NY State Senator, SD 26)>when when they've said... So said that
they had supported this legislation.

7772
07:32:31.475 --> 07:32:34.596
<v Andrew Gounardes (NY State Senator, SD 26)>We cannot trust these companies. The
federal government is failing in its

7773
07:32:34.596 --> 07:32:36.275
<v Andrew Gounardes (NY State Senator, SD 26)>response... In its responsibility

7774
07:32:36.330 --> 07:32:40.810
<v Andrew Gounardes (NY State Senator, SD 26)>to take meaningful action. And so therefore,
it's left to states like New York to be

7775
07:32:40.810 --> 07:32:45.530
<v Andrew Gounardes (NY State Senator, SD 26)>leaders in this space and work with partners
like the New York City Council under

7776
07:32:45.530 --> 07:32:48.970
<v Andrew Gounardes (NY State Senator, SD 26)>all of your collective leadership to make
sure that we are setting in place the

7777
07:32:48.970 --> 07:32:50.490
<v Andrew Gounardes (NY State Senator, SD 26)>right guardrails and protections

7778
07:32:50.936 --> 07:32:55.977
<v Andrew Gounardes (NY State Senator, SD 26)>not only to protect New Yorkers from the
catastrophic risks that are associated with

7779
07:32:56.535 --> 07:32:58.455
<v Andrew Gounardes (NY State Senator, SD 26)>these... This AI technology,

7780
07:32:58.615 --> 07:33:01.576
<v Andrew Gounardes (NY State Senator, SD 26)>but frankly to protect
all of us, all Americans,

7781
07:33:01.576 --> 07:33:05.430
<v Andrew Gounardes (NY State Senator, SD 26)>and everyone worldwide. Because we know
that these harms will not be localized. We

7782
07:33:05.430 --> 07:33:08.549
<v Andrew Gounardes (NY State Senator, SD 26)>know that the scale of what's being
developed will not just limit itself to one

7783
07:33:08.549 --> 07:33:12.950
<v Andrew Gounardes (NY State Senator, SD 26)>block, one borough, one city, one state.
It'll affect everyone the world over. And

7784
07:33:12.950 --> 07:33:16.975
<v Andrew Gounardes (NY State Senator, SD 26)>in the face of an action, it's up to us to
stand up and act. And so I thank you all

7785
07:33:16.975 --> 07:33:20.494
<v Andrew Gounardes (NY State Senator, SD 26)>for your attention to this and your support
of this and look forward to work with

7786
07:33:20.494 --> 07:33:25.215
<v Andrew Gounardes (NY State Senator, SD 26)>you all in the future on strengthening
things like the RAISE Act so we can actually

7787
07:33:25.215 --> 07:33:27.135
<v Andrew Gounardes (NY State Senator, SD 26)>hold these companies
accountable. Thank you.

7788
07:33:27.951 --> 07:33:29.150
<v CM Carmen De La Rosa>Thank you, senator.

7789
07:33:29.791 --> 07:33:33.310
<v CM Carmen De La Rosa>We have a few questions for this panel,
but I wanted to take a moment to thank you

7790
07:33:33.310 --> 07:33:33.871
<v CM Carmen De La Rosa>all.

7791
07:33:34.430 --> 07:33:39.310
<v CM Carmen De La Rosa>You know, we do believe that city and
state government have to work hand in hand,

7792
07:33:39.310 --> 07:33:42.855
<v CM Carmen De La Rosa>and we have to be the first line of defense
in order to make sure that New Yorkers

7793
07:33:42.855 --> 07:33:46.855
<v CM Carmen De La Rosa>are protected. And each and every one of
you has been such an important partner in

7794
07:33:46.855 --> 07:33:49.176
<v CM Carmen De La Rosa>making that accountability a reality

7795
07:33:49.256 --> 07:33:53.336
<v CM Carmen De La Rosa>on the state level. And, to the member of
parliament, thank you for sharing your

7796
07:33:53.336 --> 07:33:55.750
<v CM Carmen De La Rosa>story and your bone... Your vulnerability

7797
07:33:55.750 --> 07:33:56.070
<v CM Carmen De La Rosa>and,

7798
07:33:56.871 --> 07:33:57.510
<v CM Carmen De La Rosa>moving,

7799
07:33:58.150 --> 07:34:02.951
<v CM Carmen De La Rosa>from that experience to action to hold
these count... Companies accountable.

7800
07:34:03.830 --> 07:34:06.710
<v CM Carmen De La Rosa>My colleagues do have some
questions for you all.

7801
07:34:07.110 --> 07:34:08.865
<v CM Carmen De La Rosa>But before we do that

7802
07:34:08.865 --> 07:34:09.666
<v Speaker Julie Menin>I

7803
07:34:10.705 --> 07:34:14.705
<v Speaker Julie Menin>just want to thank all of our
incredible elected officials,

7804
07:34:14.865 --> 07:34:19.904
<v Speaker Julie Menin>for being here today. We know how hard you've
all worked on various legislation, and

7805
07:34:19.904 --> 07:34:23.041
<v Speaker Julie Menin>we really appreciate it so much. And,

7806
07:34:23.760 --> 07:34:28.400
<v Speaker Julie Menin>your expertise on this and, you know,
particularly in terms of the RAISE Act, we

7807
07:34:28.400 --> 07:34:32.000
<v Speaker Julie Menin>really appreciate that. And so it's very
helpful to have you here and to hear

7808
07:34:32.000 --> 07:34:35.906
<v Speaker Julie Menin>directly from you. So thank you. Thank
you for waiting it out with us. We so

7809
07:34:35.906 --> 07:34:36.867
<v Speaker Julie Menin>appreciate that.

7810
07:34:37.586 --> 07:34:41.986
<v CM Carmen De La Rosa>Thank you. We are gonna hear from council
member Jose, followed by Gutierrez,

7811
07:34:41.986 --> 07:34:46.066
<v CM Chi Ossé>William, deputy speaker Williams, and then
council member Hanif. Thank you, chair. I

7812
07:34:46.066 --> 07:34:50.630
<v CM Chi Ossé>have a question for assembly member
Boris. We heard today from OpenAI,

7813
07:34:50.630 --> 07:34:51.512
<v CM Chi Ossé>Anthropic,

7814
07:34:51.512 --> 07:34:56.070
<v CM Chi Ossé>Google, and Meta that they supported
the RAISE Act but felt like they wanted

7815
07:34:56.070 --> 07:34:57.111
<v CM Chi Ossé>stronger

7816
07:34:57.111 --> 07:34:58.150
<v CM Chi Ossé>regulations.

7817
07:34:58.470 --> 07:35:03.176
<v CM Chi Ossé>But we know that during negotiations for
New York State's RAISE Act, there was a

7818
07:35:03.176 --> 07:35:05.816
<v CM Chi Ossé>push to lower penalties from $10,000,000

7819
07:35:05.895 --> 07:35:07.016
<v CM Chi Ossé>to $30,000,000

7820
07:35:07.016 --> 07:35:09.256
<v CM Chi Ossé>to $1,000,000 to $3,000,000,

7821
07:35:09.416 --> 07:35:12.455
<v CM Chi Ossé>which is chump change
for these corporations.

7822
07:35:12.855 --> 07:35:17.100
<v CM Chi Ossé>Why were the fines in the RAISE
Act reduced from 10 to $30,000,000

7823
07:35:17.100 --> 07:35:19.500
<v CM Chi Ossé>to only 1 to $3,000,000?

7824
07:35:20.379 --> 07:35:25.740
<v Alex Bores (NY State Assembly Member, AD 73)>Council council member, thank you for
that question. In the early drafts of the

7825
07:35:25.740 --> 07:35:26.620
<v Alex Bores (NY State Assembly Member, AD 73)>RAISE Act,

7826
07:35:27.260 --> 07:35:31.580
<v Alex Bores (NY State Assembly Member, AD 73)>the penalties were up to 10,000,000 for
the first violation, and, of course, that

7827
07:35:31.580 --> 07:35:36.695
<v Alex Bores (NY State Assembly Member, AD 73)>would be based on the severity of the
violation, and up to 30,000,000 for

7828
07:35:36.695 --> 07:35:38.055
<v Alex Bores (NY State Assembly Member, AD 73)>a second violation.

7829
07:35:38.695 --> 07:35:40.215
<v Alex Bores (NY State Assembly Member, AD 73)>The companies argued

7830
07:35:40.215 --> 07:35:45.414
<v Alex Bores (NY State Assembly Member, AD 73)>that that was too high because they were
predicting that every state and perhaps

7831
07:35:45.414 --> 07:35:46.375
<v Alex Bores (NY State Assembly Member, AD 73)>municipalities

7832
07:35:46.871 --> 07:35:52.310
<v Alex Bores (NY State Assembly Member, AD 73)>would pass similar versions of the bill,
and they would end up paying the fine in 50

7833
07:35:52.310 --> 07:35:54.871
<v Alex Bores (NY State Assembly Member, AD 73)>states plus in additional municipalities.

7834
07:35:55.672 --> 07:35:58.951
<v Alex Bores (NY State Assembly Member, AD 73)>I I disagreed at the time with that
argument. I think that's been borne out since

7835
07:35:58.951 --> 07:36:00.630
<v Alex Bores (NY State Assembly Member, AD 73)>we've seen only three states pass it.

7836
07:36:01.215 --> 07:36:06.895
<v Alex Bores (NY State Assembly Member, AD 73)>But perhaps what's most confusing to me is
that now they turn around and say that no

7837
07:36:06.895 --> 07:36:08.734
<v Alex Bores (NY State Assembly Member, AD 73)>other state should pass

7838
07:36:08.895 --> 07:36:13.855
<v Alex Bores (NY State Assembly Member, AD 73)>bills like this or that no city should pass
bills like this because they're already

7839
07:36:13.855 --> 07:36:18.570
<v Alex Bores (NY State Assembly Member, AD 73)>paying the fines. Their rhetoric, their
testimony was explicit that they are

7840
07:36:18.570 --> 07:36:22.410
<v Alex Bores (NY State Assembly Member, AD 73)>expecting to pay fines in
many, many jurisdictions

7841
07:36:22.490 --> 07:36:26.170
<v Alex Bores (NY State Assembly Member, AD 73)>for the same action. And so I think
that testimony is quite supportive

7842
07:36:27.125 --> 07:36:31.125
<v Alex Bores (NY State Assembly Member, AD 73)>of the bills that are being put forward
today that would hold them accountable and

7843
07:36:31.125 --> 07:36:35.125
<v Alex Bores (NY State Assembly Member, AD 73)>put additional fines on behavior even if
it is covered by the RAISE Act because,

7844
07:36:35.125 --> 07:36:38.725
<v Alex Bores (NY State Assembly Member, AD 73)>again, that was their own testimony as
to why those fines should be lowered.

7845
07:36:40.140 --> 07:36:44.220
<v CM Chi Ossé>Thank you very much. And, senator Gonzalez,
I would love to hear from you about some

7846
07:36:44.220 --> 07:36:47.500
<v CM Chi Ossé>of the organizing that you've
been doing, especially within

7847
07:36:47.740 --> 07:36:52.620
<v CM Chi Ossé>the New York City Democratic Socialists
of America around AI regulation and the

7848
07:36:52.620 --> 07:36:55.020
<v CM Chi Ossé>vision that is being put
forward within that regard.

7849
07:36:56.424 --> 07:37:00.424
<v Kristen Gonzalez (NY State Senator, SD 59)>Thank you so much, council member. As a
fellow member of the Democratic Socialists

7850
07:37:00.424 --> 07:37:05.225
<v Kristen Gonzalez (NY State Senator, SD 59)>of America, one thing I spoke about in my
testimony was prioritizing a rights based

7851
07:37:05.225 --> 07:37:06.025
<v Kristen Gonzalez (NY State Senator, SD 59)>approach.

7852
07:37:06.504 --> 07:37:12.200
<v Kristen Gonzalez (NY State Senator, SD 59)>I think it is so important to put New
Yorkers first by clarifying and setting the

7853
07:37:12.200 --> 07:37:17.318
<v Kristen Gonzalez (NY State Senator, SD 59)>terms for what our digital rights should
be over the profit driven motives of so

7854
07:37:17.318 --> 07:37:20.040
<v Kristen Gonzalez (NY State Senator, SD 59)>many of these tech companies
and tech oligarchs.

7855
07:37:20.200 --> 07:37:24.205
<v Kristen Gonzalez (NY State Senator, SD 59)>That's why we've worked hard on the New
York AI act. We've certainly supported

7856
07:37:24.365 --> 07:37:25.246
<v Kristen Gonzalez (NY State Senator, SD 59)>legislation

7857
07:37:25.246 --> 07:37:29.166
<v Kristen Gonzalez (NY State Senator, SD 59)>on frontier models and reining them in,
but New Yorkers who are marginalized are

7858
07:37:29.166 --> 07:37:33.566
<v Kristen Gonzalez (NY State Senator, SD 59)>feeling the impacts of imperfect tools every
single day, and I'm proud that the tech

7859
07:37:33.566 --> 07:37:37.709
<v Kristen Gonzalez (NY State Senator, SD 59)>action working group has focused on this
issue. In addition, I've worked with a

7860
07:37:37.709 --> 07:37:39.549
<v Kristen Gonzalez (NY State Senator, SD 59)>statewide coalition beyond,

7861
07:37:39.870 --> 07:37:43.549
<v Kristen Gonzalez (NY State Senator, SD 59)>the Democratic Socialists of America,
but that include environmental

7862
07:37:43.549 --> 07:37:48.350
<v Kristen Gonzalez (NY State Senator, SD 59)>justice organizations like Food and Water
Watch and more that are in the fight

7863
07:37:48.350 --> 07:37:52.984
<v Kristen Gonzalez (NY State Senator, SD 59)>against hyperscale data centers. Because as
we're talking about the real world harms

7864
07:37:52.984 --> 07:37:56.346
<v Kristen Gonzalez (NY State Senator, SD 59)>that are being felt and the
potential catastrophic risks,

7865
07:37:56.746 --> 07:37:58.025
<v Kristen Gonzalez (NY State Senator, SD 59)>every single day,

7866
07:37:58.346 --> 07:38:03.602
<v Kristen Gonzalez (NY State Senator, SD 59)>people in Upstate New York, in the capital
region, in the Hudson Valley are feeling

7867
07:38:03.602 --> 07:38:08.800
<v Kristen Gonzalez (NY State Senator, SD 59)>what it means to live near a hyperscale
data center. There are millions

7868
07:38:08.800 --> 07:38:14.960
<v Kristen Gonzalez (NY State Senator, SD 59)>of gallons of water being pulled from our
state's lakes right now. Communities are

7869
07:38:14.960 --> 07:38:19.785
<v Kristen Gonzalez (NY State Senator, SD 59)>dealing with noise pollution and air
pollution right now, And it's really important

7870
07:38:19.785 --> 07:38:24.986
<v Kristen Gonzalez (NY State Senator, SD 59)>to legislate not only the risks, but legislate
those data center driven harms. Thank

7871
07:38:24.986 --> 07:38:25.865
<v Kristen Gonzalez (NY State Senator, SD 59)>you. Both.

7872
07:38:26.266 --> 07:38:28.746
<v CM Carmen De La Rosa>Thanks. Thank you.
Councilmember Gutierrez?

7873
07:38:28.986 --> 07:38:29.705
<v CM Jennifer Gutiérrez>Thank you.

7874
07:38:30.986 --> 07:38:31.385
<v CM Jennifer Gutiérrez>Senator

7875
07:38:32.230 --> 07:38:34.951
<v CM Jennifer Gutiérrez>from from London... I'm sorry.
Didn't catch your name.

7876
07:38:35.910 --> 07:38:37.190
<v CM Jennifer Gutiérrez>I think... I apologize.

7877
07:38:37.590 --> 07:38:40.951
<v CM Jennifer Gutiérrez>Thank you for for your
testimony and for educating

7878
07:38:41.430 --> 07:38:44.630
<v CM Jennifer Gutiérrez>us on the volatility of
of this tech company.

7879
07:38:45.031 --> 07:38:48.164
<v CM Jennifer Gutiérrez>I know you're you're in process right now,

7880
07:38:48.484 --> 07:38:50.965
<v CM Jennifer Gutiérrez>but could you share a little
bit about legislatively

7881
07:38:50.965 --> 07:38:52.004
<v CM Jennifer Gutiérrez>what you think,

7882
07:38:52.484 --> 07:38:53.684
<v CM Jennifer Gutiérrez>if existed,

7883
07:38:53.684 --> 07:38:55.764
<v CM Jennifer Gutiérrez>if in place a guardrail,

7884
07:38:56.965 --> 07:39:02.200
<v CM Jennifer Gutiérrez>what could have in any way prevented
what happened to you? How how can we...

7885
07:39:02.680 --> 07:39:04.760
<v CM Jennifer Gutiérrez>Whether the state or the city legislate

7886
07:39:04.760 --> 07:39:05.639
<v CM Jennifer Gutiérrez>specifically

7887
07:39:05.639 --> 07:39:06.360
<v CM Jennifer Gutiérrez>to

7888
07:39:06.440 --> 07:39:08.040
<v CM Jennifer Gutiérrez>objectifying and and

7889
07:39:08.280 --> 07:39:09.000
<v CM Jennifer Gutiérrez>potentially,

7890
07:39:09.240 --> 07:39:13.080
<v CM Jennifer Gutiérrez>images that include sexual harm to people,
what what have you learned or have you

7891
07:39:13.080 --> 07:39:14.920
<v CM Jennifer Gutiérrez>seen that we can potentially legislate?

7892
07:39:15.215 --> 07:39:18.016
<v CM Jennifer Gutiérrez>And then I just had one more
question for senator Gonzalez.

7893
07:39:18.574 --> 07:39:22.096
<v CM Jennifer Gutiérrez>I love hearing you talk about
your disdain for data centers.

7894
07:39:22.654 --> 07:39:27.951
<v CM Jennifer Gutiérrez>Can you share for us a minute, afterwards
just what you what you've what you were

7895
07:39:27.951 --> 07:39:30.430
<v CM Jennifer Gutiérrez>feeling hearing some of
these tech tech companies,

7896
07:39:30.750 --> 07:39:32.031
<v CM Jennifer Gutiérrez>during their testimony?

7897
07:39:32.031 --> 07:39:32.910
<v CM Jennifer Gutiérrez>Thank you. Thank

7898
07:39:34.670 --> 07:39:37.150
<v Jess Asato (UK Member of Parliament, Lowestoft)>you so much for your question.
I'll try to be brief.

7899
07:39:37.550 --> 07:39:42.256
<v Jess Asato (UK Member of Parliament, Lowestoft)>Within The UK, we already have data protection
and privacy laws, which is the basis

7900
07:39:42.256 --> 07:39:44.096
<v Jess Asato (UK Member of Parliament, Lowestoft)>under which I'm bringing my claim.

7901
07:39:44.574 --> 07:39:47.055
<v Jess Asato (UK Member of Parliament, Lowestoft)>And, of course, we're going
to test that in court.

7902
07:39:47.295 --> 07:39:49.215
<v Jess Asato (UK Member of Parliament, Lowestoft)>But, clearly, if your

7903
07:39:49.295 --> 07:39:52.416
<v Jess Asato (UK Member of Parliament, Lowestoft)>image and your likeness
forms part of your data,

7904
07:39:52.895 --> 07:39:55.215
<v Jess Asato (UK Member of Parliament, Lowestoft)>then making sure that everybody,

7905
07:39:55.775 --> 07:39:56.256
<v Jess Asato (UK Member of Parliament, Lowestoft)>has

7906
07:39:56.640 --> 07:39:57.920
<v Jess Asato (UK Member of Parliament, Lowestoft)>that aspect

7907
07:39:58.000 --> 07:40:00.400
<v Jess Asato (UK Member of Parliament, Lowestoft)>understood by AI companies

7908
07:40:00.720 --> 07:40:05.361
<v Jess Asato (UK Member of Parliament, Lowestoft)>could mean that they would then have
to treat our likeness and and our data

7909
07:40:05.361 --> 07:40:06.800
<v Jess Asato (UK Member of Parliament, Lowestoft)>differently in future.

7910
07:40:07.440 --> 07:40:10.480
<v Jess Asato (UK Member of Parliament, Lowestoft)>But, also, the UK government
has recently made

7911
07:40:11.201 --> 07:40:12.160
<v Jess Asato (UK Member of Parliament, Lowestoft)>the

7912
07:40:12.395 --> 07:40:14.635
<v Jess Asato (UK Member of Parliament, Lowestoft)>sort of generation of nonconsensual

7913
07:40:14.635 --> 07:40:15.914
<v Jess Asato (UK Member of Parliament, Lowestoft)>intimate imagery

7914
07:40:16.074 --> 07:40:18.555
<v Jess Asato (UK Member of Parliament, Lowestoft)>a criminal offense, and as part of that,

7915
07:40:19.115 --> 07:40:20.074
<v Jess Asato (UK Member of Parliament, Lowestoft)>has brought,

7916
07:40:20.395 --> 07:40:20.955
<v Jess Asato (UK Member of Parliament, Lowestoft)>the

7917
07:40:21.275 --> 07:40:26.715
<v Jess Asato (UK Member of Parliament, Lowestoft)>operation of AI chatbots into our online
safety act so that our regulator Ofcom

7918
07:40:27.050 --> 07:40:29.050
<v Jess Asato (UK Member of Parliament, Lowestoft)>can also make sure that if,

7919
07:40:29.691 --> 07:40:32.172
<v Jess Asato (UK Member of Parliament, Lowestoft)>an AI company is allowing users

7920
07:40:32.410 --> 07:40:33.371
<v Jess Asato (UK Member of Parliament, Lowestoft)>to create

7921
07:40:33.531 --> 07:40:35.371
<v Jess Asato (UK Member of Parliament, Lowestoft)>nonconsensual intimate imagery,

7922
07:40:35.531 --> 07:40:37.771
<v Jess Asato (UK Member of Parliament, Lowestoft)>that that act is a is a criminal,

7923
07:40:38.010 --> 07:40:42.895
<v Jess Asato (UK Member of Parliament, Lowestoft)>sort of act happening on their platform.
And so it gives it more of a ability to be

7924
07:40:42.895 --> 07:40:46.414
<v Jess Asato (UK Member of Parliament, Lowestoft)>taken down. But from my own
perspective, it's about consent.

7925
07:40:47.215 --> 07:40:48.174
<v Jess Asato (UK Member of Parliament, Lowestoft)>In the end,

7926
07:40:48.494 --> 07:40:50.494
<v Jess Asato (UK Member of Parliament, Lowestoft)>whether it is nonconsensual

7927
07:40:50.494 --> 07:40:52.814
<v Jess Asato (UK Member of Parliament, Lowestoft)>intimate imagery or just nonconsensual

7928
07:40:52.814 --> 07:40:53.615
<v Jess Asato (UK Member of Parliament, Lowestoft)>imagery,

7929
07:40:54.254 --> 07:40:56.254
<v Jess Asato (UK Member of Parliament, Lowestoft)>if we haven't consented to it,

7930
07:40:56.574 --> 07:40:57.854
<v Jess Asato (UK Member of Parliament, Lowestoft)>then it shouldn't be allowed to do it.

7931
07:40:58.621 --> 07:41:00.380
<v Speaker Julie Menin>I just wanna interject

7932
07:41:00.380 --> 07:41:05.020
<v Speaker Julie Menin>and thank you so much for coming. I mean,
to come, you know, as far as you travel to

7933
07:41:05.020 --> 07:41:07.580
<v Speaker Julie Menin>be here personally to testify,

7934
07:41:07.580 --> 07:41:10.460
<v Speaker Julie Menin>it really means a lot to us,
and we deeply appreciate it.

7935
07:41:14.574 --> 07:41:16.016
<v Kristen Gonzalez (NY State Senator, SD 59)>And thank you so much.

7936
07:41:16.895 --> 07:41:21.775
<v Kristen Gonzalez (NY State Senator, SD 59)>Again, I think this is a great segue to
the fact that there are international

7937
07:41:21.775 --> 07:41:23.055
<v Kristen Gonzalez (NY State Senator, SD 59)>safety standards.

7938
07:41:23.295 --> 07:41:27.960
<v Kristen Gonzalez (NY State Senator, SD 59)>So much of the work that I've done has been
to call out the fact that there are not

7939
07:41:27.960 --> 07:41:30.680
<v Kristen Gonzalez (NY State Senator, SD 59)>only standards like ISO 42,001

7940
07:41:30.680 --> 07:41:35.240
<v Kristen Gonzalez (NY State Senator, SD 59)>or national standards like the risk AI
framework that these companies could comply

7941
07:41:35.240 --> 07:41:39.720
<v Kristen Gonzalez (NY State Senator, SD 59)>with, but actively lobby against
bills that would require them to.

7942
07:41:40.406 --> 07:41:44.967
<v Kristen Gonzalez (NY State Senator, SD 59)>And, to the council member's question and
what I was feeling, I was feeling so much

7943
07:41:44.967 --> 07:41:49.047
<v Kristen Gonzalez (NY State Senator, SD 59)>of the same frustration that my colleagues
on this panel expressed earlier.

7944
07:41:49.285 --> 07:41:53.979
<v Kristen Gonzalez (NY State Senator, SD 59)>These companies are speaking out of both
sides of their mouths. On the one hand,

7945
07:41:53.979 --> 07:41:57.180
<v Kristen Gonzalez (NY State Senator, SD 59)>they'll say that they want
regulation, but anytime

7946
07:41:57.180 --> 07:42:00.700
<v Kristen Gonzalez (NY State Senator, SD 59)>a single bill that would meaningfully hold
them to account is brought in front of

7947
07:42:00.700 --> 07:42:03.340
<v Kristen Gonzalez (NY State Senator, SD 59)>them, they will say, but not like that.

7948
07:42:03.740 --> 07:42:07.340
<v Kristen Gonzalez (NY State Senator, SD 59)>When we talk about catastrophic risks, many
of the bills that have been put forward

7949
07:42:07.340 --> 07:42:09.525
<v Kristen Gonzalez (NY State Senator, SD 59)>and even with the stated
past, for example,

7950
07:42:09.924 --> 07:42:15.205
<v Kristen Gonzalez (NY State Senator, SD 59)>had set definitions at a 100 deaths
or billions of dollars of damage.

7951
07:42:15.525 --> 07:42:17.445
<v Kristen Gonzalez (NY State Senator, SD 59)>Even that was considered

7952
07:42:17.525 --> 07:42:18.324
<v Kristen Gonzalez (NY State Senator, SD 59)>too

7953
07:42:18.564 --> 07:42:21.766
<v Kristen Gonzalez (NY State Senator, SD 59)>low of a threshold for many
of these AI companies.

7954
07:42:22.390 --> 07:42:26.150
<v Kristen Gonzalez (NY State Senator, SD 59)>And bills like the one that I
mentioned that protects our kids,

7955
07:42:26.390 --> 07:42:30.871
<v Kristen Gonzalez (NY State Senator, SD 59)>that seeks to say that a single kid
being coached to commit suicide,

7956
07:42:31.031 --> 07:42:35.350
<v Kristen Gonzalez (NY State Senator, SD 59)>which has actively happened in multiple
cases across the country, is one kid too

7957
07:42:35.350 --> 07:42:38.877
<v Kristen Gonzalez (NY State Senator, SD 59)>many, is considered unacceptable
to these tech companies.

7958
07:42:39.195 --> 07:42:43.516
<v Kristen Gonzalez (NY State Senator, SD 59)>So I feel an incredible amount of frustration
while I feel an incredible amount of

7959
07:42:43.516 --> 07:42:48.630
<v Kristen Gonzalez (NY State Senator, SD 59)>hope seeing you not only have this hearing,
but all of us as legislators step up to

7960
07:42:48.630 --> 07:42:52.070
<v Kristen Gonzalez (NY State Senator, SD 59)>meet the moment and regulate and
rein in these tech companies.

7961
07:42:52.150 --> 07:42:52.791
<v Kristen Gonzalez (NY State Senator, SD 59)>Thank you.

7962
07:42:53.670 --> 07:42:54.710
<v CM Carmen De La Rosa>Thank you.

7963
07:42:55.590 --> 07:42:59.910
<v CM Nantasha Williams>Deputy speaker Williams followed by council
member Hanif and then Maloney. Thank

7964
07:42:59.990 --> 07:43:02.816
<v CM Nantasha Williams>you. Yeah. I also just wanted
to thank you all for coming.

7965
07:43:03.695 --> 07:43:07.377
<v CM Nantasha Williams>So the question I have for the state
members is how can we be helpful?

7966
07:43:07.775 --> 07:43:10.736
<v CM Nantasha Williams>I know that there are a ton of resolutions

7967
07:43:10.816 --> 07:43:12.176
<v CM Nantasha Williams>that are supporting

7968
07:43:12.256 --> 07:43:13.297
<v CM Nantasha Williams>state bills.

7969
07:43:14.176 --> 07:43:18.310
<v CM Nantasha Williams>It would be helpful just to let us how we
can be helpful and how we can work more

7970
07:43:18.310 --> 07:43:19.430
<v CM Nantasha Williams>closely together.

7971
07:43:21.190 --> 07:43:21.830
<v Clyde Vanel (NY State Assembly Member, AD 33)>Well,

7972
07:43:22.550 --> 07:43:25.430
<v Clyde Vanel (NY State Assembly Member, AD 33)>deputy speaker, thank you for that. Thank
you for asking that question. What's

7973
07:43:25.430 --> 07:43:29.270
<v Clyde Vanel (NY State Assembly Member, AD 33)>really important is this step, toward...
This hearing and this step towards this is

7974
07:43:29.270 --> 07:43:32.525
<v Clyde Vanel (NY State Assembly Member, AD 33)>very important. You know, it's it's a
big deal that you guys are having this

7975
07:43:32.525 --> 07:43:33.086
<v Clyde Vanel (NY State Assembly Member, AD 33)>hearing,

7976
07:43:33.484 --> 07:43:37.805
<v Clyde Vanel (NY State Assembly Member, AD 33)>calling the, you know, the industry here
and and and the different panels. We have

7977
07:43:37.805 --> 07:43:41.805
<v Clyde Vanel (NY State Assembly Member, AD 33)>to work on... Now this is hard. This is
difficult, but we have to work on it head

7978
07:43:41.805 --> 07:43:45.084
<v Clyde Vanel (NY State Assembly Member, AD 33)>on. And what's very important is
for us to be able to coordinate our

7979
07:43:45.361 --> 07:43:48.800
<v Clyde Vanel (NY State Assembly Member, AD 33)>to see that, you know, we we were having
a good relationship with many of the the

7980
07:43:48.800 --> 07:43:52.240
<v Clyde Vanel (NY State Assembly Member, AD 33)>members. I have a great relationship with
the with the chair of technology. It's

7981
07:43:52.240 --> 07:43:56.960
<v Clyde Vanel (NY State Assembly Member, AD 33)>very important for us to be able to work
on this. And there are many layers to

7982
07:43:56.960 --> 07:43:58.640
<v Clyde Vanel (NY State Assembly Member, AD 33)>addressing what's, this technology,

7983
07:43:59.416 --> 07:44:01.416
<v Clyde Vanel (NY State Assembly Member, AD 33)>and we touched on a lot of them today,

7984
07:44:01.576 --> 07:44:03.256
<v Clyde Vanel (NY State Assembly Member, AD 33)>but we have to work on this together.

7985
07:44:05.096 --> 07:44:09.096
<v Jordan Wright (NY State Assembly Member, AD 70)>Ditto. And I know how appealing Albany can
be, so please come on up and help us as

7986
07:44:09.096 --> 07:44:11.256
<v Jordan Wright (NY State Assembly Member, AD 70)>we're rallying for different
pieces of legislation

7987
07:44:11.256 --> 07:44:14.871
<v Jordan Wright (NY State Assembly Member, AD 70)>and, hitting the pavement and getting
harder work. Thank you, council member.

7988
07:44:16.070 --> 07:44:19.031
<v Andrew Gounardes (NY State Senator, SD 26)>And I would just chime in there. Think,
you know, what we heard here today... What

7989
07:44:19.031 --> 07:44:23.111
<v Andrew Gounardes (NY State Senator, SD 26)>you all heard today was many of these
companies basically speak out of both sides of

7990
07:44:23.111 --> 07:44:26.710
<v Andrew Gounardes (NY State Senator, SD 26)>their mouths about what they think the
appropriate role of government to be is in.

7991
07:44:27.006 --> 07:44:30.766
<v Andrew Gounardes (NY State Senator, SD 26)>So next year, when we're up in Albany trying
to fight to pass all of these different

7992
07:44:30.766 --> 07:44:36.285
<v Andrew Gounardes (NY State Senator, SD 26)>laws and we're, you know, being bombarded
by tens of millions of dollars from these

7993
07:44:36.285 --> 07:44:40.525
<v Andrew Gounardes (NY State Senator, SD 26)>companies with lobbyists and ads and radio
ads and text messages and everything,

7994
07:44:40.850 --> 07:44:44.530
<v Andrew Gounardes (NY State Senator, SD 26)>Having your voices to go back into your
communities and to come up to talk to our

7995
07:44:44.530 --> 07:44:49.330
<v Andrew Gounardes (NY State Senator, SD 26)>colleagues and say, hey, we held this
hearing. They lied to us about x y and z, a b

7996
07:44:49.330 --> 07:44:53.330
<v Andrew Gounardes (NY State Senator, SD 26)>and c, and, like, here's the proof of it.
This way, we can call these companies out

7997
07:44:53.330 --> 07:44:57.215
<v Andrew Gounardes (NY State Senator, SD 26)>when they're trying to have it both ways.
You all experience that firsthand and you

7998
07:44:57.215 --> 07:45:01.055
<v Andrew Gounardes (NY State Senator, SD 26)>being able to speak to that and remind the
public about that time and time and time

7999
07:45:01.055 --> 07:45:06.174
<v Andrew Gounardes (NY State Senator, SD 26)>again will only further to bolster the
momentum that we have on our side to put in

8000
07:45:06.174 --> 07:45:10.459
<v Andrew Gounardes (NY State Senator, SD 26)>place these guardrails and protections,
whether it's around bias or safety or any of

8001
07:45:10.459 --> 07:45:12.860
<v Andrew Gounardes (NY State Senator, SD 26)>the other issues that we've been
talking about during this hearing.

8002
07:45:13.818 --> 07:45:16.780
<v Alex Bores (NY State Assembly Member, AD 73)>To follow on on on the senator's comments,

8003
07:45:17.020 --> 07:45:20.219
<v Alex Bores (NY State Assembly Member, AD 73)>the company say different things
to different levels of government.

8004
07:45:20.459 --> 07:45:22.700
<v Alex Bores (NY State Assembly Member, AD 73)>A few weeks ago, I was

8005
07:45:23.256 --> 07:45:27.016
<v Alex Bores (NY State Assembly Member, AD 73)>actually in Brussels. I met with a few
members of the European Parliament to talk

8006
07:45:27.016 --> 07:45:29.416
<v Alex Bores (NY State Assembly Member, AD 73)>about the passage of the EUAI Act,

8007
07:45:29.656 --> 07:45:32.295
<v Alex Bores (NY State Assembly Member, AD 73)>and we compared what we were lobbied on.

8008
07:45:32.775 --> 07:45:37.770
<v Alex Bores (NY State Assembly Member, AD 73)>And surprising to me, catastrophic risk
didn't really come up in the EU, but they

8009
07:45:37.770 --> 07:45:40.170
<v Alex Bores (NY State Assembly Member, AD 73)>got a lot of lobbying around copyright.

8010
07:45:40.570 --> 07:45:44.650
<v Alex Bores (NY State Assembly Member, AD 73)>And my assumption is because they thought
the courts are stronger on copyright in

8011
07:45:44.650 --> 07:45:48.490
<v Alex Bores (NY State Assembly Member, AD 73)>Europe, and so that's what they lobbied on
there and said very different things in

8012
07:45:49.176 --> 07:45:49.977
<v Alex Bores (NY State Assembly Member, AD 73)>America,

8013
07:45:50.055 --> 07:45:53.336
<v Alex Bores (NY State Assembly Member, AD 73)>in New York state. I imagine they're saying
different things in Albany as they will

8014
07:45:53.336 --> 07:45:56.295
<v Alex Bores (NY State Assembly Member, AD 73)>say here to you in the city council,

8015
07:45:56.615 --> 07:46:00.135
<v Alex Bores (NY State Assembly Member, AD 73)>and we should all be sharing notes on what's
being said and holding them to account

8016
07:46:00.135 --> 07:46:02.135
<v Alex Bores (NY State Assembly Member, AD 73)>and making sure that it's
a consistent message.

8017
07:46:03.140 --> 07:46:05.541
<v Alex Bores (NY State Assembly Member, AD 73)>The other bit that I would add is,

8018
07:46:05.780 --> 07:46:10.020
<v Alex Bores (NY State Assembly Member, AD 73)>the governor has signaled that she wants
to strengthen the RAISE Act and put new

8019
07:46:10.020 --> 07:46:12.660
<v Alex Bores (NY State Assembly Member, AD 73)>proposals in her budget proposal.

8020
07:46:13.220 --> 07:46:15.380
<v Alex Bores (NY State Assembly Member, AD 73)>Please be part of that conversation.

8021
07:46:15.380 --> 07:46:18.260
<v Alex Bores (NY State Assembly Member, AD 73)>When the city council
submits its budget requests,

8022
07:46:18.654 --> 07:46:22.654
<v Alex Bores (NY State Assembly Member, AD 73)>obviously, you should focus on the fiscal
needs first and foremost, but the state

8023
07:46:22.654 --> 07:46:29.055
<v Alex Bores (NY State Assembly Member, AD 73)>budget tends to be a a broad document.
And if you have policy ideas that for some

8024
07:46:29.055 --> 07:46:33.390
<v Alex Bores (NY State Assembly Member, AD 73)>reason or another need state approval,
think of that as another mechanism,

8025
07:46:33.470 --> 07:46:35.710
<v Alex Bores (NY State Assembly Member, AD 73)>another lever that you have to

8026
07:46:35.870 --> 07:46:38.750
<v Alex Bores (NY State Assembly Member, AD 73)>help ensure that you can do the
best job for your constituents.

8027
07:46:39.709 --> 07:46:40.190
<v CM Nantasha Williams>Just,

8028
07:46:40.750 --> 07:46:45.715
<v CM Nantasha Williams>two more quick questions. I passed the bill
and the chair has a bill on AI's impact

8029
07:46:45.715 --> 07:46:51.314
<v CM Nantasha Williams>on the municipal workforce. Just wondering
if there was any traction on that for the

8030
07:46:51.314 --> 07:46:52.115
<v CM Nantasha Williams>state's

8031
07:46:52.195 --> 07:46:53.074
<v CM Nantasha Williams>workforce.

8032
07:46:53.314 --> 07:46:55.156
<v CM Nantasha Williams>And then just a question for,

8033
07:46:55.475 --> 07:46:59.390
<v CM Nantasha Williams>the member of parliament. My bill that
we're hearing today is to address

8034
07:46:59.390 --> 07:47:04.111
<v CM Nantasha Williams>unauthorized use of deepfakes. So
in addition to the privacy laws,

8035
07:47:04.510 --> 07:47:06.590
<v CM Nantasha Williams>and Britain just wanted to know,

8036
07:47:07.470 --> 07:47:10.830
<v CM Nantasha Williams>if you all have specific
legislation around deepfakes.

8037
07:47:14.055 --> 07:47:16.775
<v CM Nantasha Williams>Oh, and what you... How you envision the

8038
07:47:17.016 --> 07:47:20.215
<v CM Nantasha Williams>cooperation with the international community
because you mentioned that. And what

8039
07:47:20.215 --> 07:47:21.576
<v CM Nantasha Williams>does that look like for you?

8040
07:47:24.295 --> 07:47:25.977
<v Jess Asato (UK Member of Parliament, Lowestoft)>Thank you so much for your question.

8041
07:47:26.615 --> 07:47:28.592
<v Jess Asato (UK Member of Parliament, Lowestoft)>We know that female,

8042
07:47:28.990 --> 07:47:34.752
<v Jess Asato (UK Member of Parliament, Lowestoft)>elected representatives are 33 times more
likely to be targeted by sexual deepfakes

8043
07:47:34.910 --> 07:47:35.871
<v Jess Asato (UK Member of Parliament, Lowestoft)>than men.

8044
07:47:36.592 --> 07:47:40.750
<v Jess Asato (UK Member of Parliament, Lowestoft)>The whole issue around sexualized deepfakes
is deeply gendered, and I think it's

8045
07:47:40.750 --> 07:47:42.271
<v Jess Asato (UK Member of Parliament, Lowestoft)>very important that we acknowledge that.

8046
07:47:43.006 --> 07:47:45.646
<v Jess Asato (UK Member of Parliament, Lowestoft)>Because what it does is it drives women

8047
07:47:45.727 --> 07:47:47.807
<v Jess Asato (UK Member of Parliament, Lowestoft)>from the public space.

8048
07:47:48.285 --> 07:47:53.246
<v Jess Asato (UK Member of Parliament, Lowestoft)>We know that women are less likely to talk
about politics online than men are, and

8049
07:47:53.246 --> 07:47:57.100
<v Jess Asato (UK Member of Parliament, Lowestoft)>that is because they become a target as
soon as they raise their heads in this

8050
07:47:57.100 --> 07:47:58.780
<v Jess Asato (UK Member of Parliament, Lowestoft)>place, which is not social.

8051
07:47:59.660 --> 07:48:04.299
<v Jess Asato (UK Member of Parliament, Lowestoft)>That is why taking action on this is so
important and, you know, to be commended

8052
07:48:04.299 --> 07:48:08.780
<v Jess Asato (UK Member of Parliament, Lowestoft)>in terms of looking at this. But when we're
looking at these deepfakes, it needs to

8053
07:48:08.780 --> 07:48:10.764
<v Jess Asato (UK Member of Parliament, Lowestoft)>be for all of our citizens,

8054
07:48:11.324 --> 07:48:15.244
<v Jess Asato (UK Member of Parliament, Lowestoft)>because it's not just about protecting
those of us who have the honor of being

8055
07:48:15.244 --> 07:48:15.885
<v Jess Asato (UK Member of Parliament, Lowestoft)>elected.

8056
07:48:16.283 --> 07:48:18.604
<v Jess Asato (UK Member of Parliament, Lowestoft)>It is about, you know, coworkers

8057
07:48:18.604 --> 07:48:19.324
<v Jess Asato (UK Member of Parliament, Lowestoft)>who

8058
07:48:19.404 --> 07:48:20.844
<v Jess Asato (UK Member of Parliament, Lowestoft)>may be targeted

8059
07:48:20.924 --> 07:48:21.324
<v Jess Asato (UK Member of Parliament, Lowestoft)>because,

8060
07:48:21.770 --> 07:48:26.250
<v Jess Asato (UK Member of Parliament, Lowestoft)>you know, they didn't smile at the right
time or your last date, he wouldn't go on a

8061
07:48:26.250 --> 07:48:31.450
<v Jess Asato (UK Member of Parliament, Lowestoft)>second date with you or teachers who are
being targeted by their male students.

8062
07:48:31.610 --> 07:48:33.690
<v Jess Asato (UK Member of Parliament, Lowestoft)>All of these people need protection.

8063
07:48:34.010 --> 07:48:36.410
<v Jess Asato (UK Member of Parliament, Lowestoft)>That's why in The UK, we have taken steps

8064
07:48:36.715 --> 07:48:39.275
<v Jess Asato (UK Member of Parliament, Lowestoft)>to make sure that there
is a criminalization

8065
07:48:39.275 --> 07:48:41.834
<v Jess Asato (UK Member of Parliament, Lowestoft)>of the creation of nonconsensual

8066
07:48:41.834 --> 07:48:43.195
<v Jess Asato (UK Member of Parliament, Lowestoft)>sexual deepfakes.

8067
07:48:43.275 --> 07:48:48.395
<v Jess Asato (UK Member of Parliament, Lowestoft)>But my view is that we need to go further
than just the sexualized deepfakes because

8068
07:48:48.875 --> 07:48:51.602
<v Jess Asato (UK Member of Parliament, Lowestoft)>any act could be harmful to women.

8069
07:48:52.000 --> 07:48:54.720
<v Jess Asato (UK Member of Parliament, Lowestoft)>The removal of a woman's hijab

8070
07:48:55.041 --> 07:49:00.320
<v Jess Asato (UK Member of Parliament, Lowestoft)>through AI could put that woman in harm's
way, but there's nothing sexualized about

8071
07:49:00.320 --> 07:49:04.416
<v Jess Asato (UK Member of Parliament, Lowestoft)>it. And so we have to look at this
as an issue, as I said, of consent.

8072
07:49:04.734 --> 07:49:08.975
<v Jess Asato (UK Member of Parliament, Lowestoft)>And that is where data and privacy is so,
so important, and that's where I think

8073
07:49:08.975 --> 07:49:11.295
<v Jess Asato (UK Member of Parliament, Lowestoft)>there could be international agreement.

8074
07:49:11.375 --> 07:49:16.016
<v Jess Asato (UK Member of Parliament, Lowestoft)>We can see, as you said, that things
are moving forward within the EU space.

8075
07:49:16.650 --> 07:49:19.371
<v Jess Asato (UK Member of Parliament, Lowestoft)>Really, though, what we
need is for tech companies

8076
07:49:19.610 --> 07:49:21.451
<v Jess Asato (UK Member of Parliament, Lowestoft)>to be safe by design.

8077
07:49:21.770 --> 07:49:26.330
<v Jess Asato (UK Member of Parliament, Lowestoft)>And every single country has those rules.
We should be looking at product regulation

8078
07:49:26.330 --> 07:49:31.135
<v Jess Asato (UK Member of Parliament, Lowestoft)>just as we do in the offline sphere. The
end of the day, we have cars that have seat

8079
07:49:31.135 --> 07:49:32.975
<v Jess Asato (UK Member of Parliament, Lowestoft)>belts and they have airbags,

8080
07:49:33.055 --> 07:49:37.854
<v Jess Asato (UK Member of Parliament, Lowestoft)>and that's because we expect them to be safe
by design. It is not too much to expect

8081
07:49:37.854 --> 07:49:40.574
<v Jess Asato (UK Member of Parliament, Lowestoft)>that of the tech products
that we use today.

8082
07:49:43.682 --> 07:49:46.160
<v Alex Bores (NY State Assembly Member, AD 73)>On the the question about
the state workforce,

8083
07:49:46.320 --> 07:49:48.080
<v Alex Bores (NY State Assembly Member, AD 73)>I wish senator Gonzalez,

8084
07:49:48.400 --> 07:49:51.121
<v Alex Bores (NY State Assembly Member, AD 73)>could have heard that because she wrote
the Loading Act, which which helps to

8085
07:49:51.121 --> 07:49:52.960
<v Alex Bores (NY State Assembly Member, AD 73)>protect the state workforce.

8086
07:49:53.762 --> 07:49:57.924
<v Alex Bores (NY State Assembly Member, AD 73)>And I think there's an update to it pending
before the governor, which leads me to

8087
07:49:57.924 --> 07:50:00.086
<v Alex Bores (NY State Assembly Member, AD 73)>you asking about deep fakes as well.

8088
07:50:00.564 --> 07:50:03.766
<v Alex Bores (NY State Assembly Member, AD 73)>There's a bill from senator Gennartis and
I pending before the governor on that.

8089
07:50:04.004 --> 07:50:07.445
<v Alex Bores (NY State Assembly Member, AD 73)>Senator Gonzalez also mentioned chatbots
to protect kids. There's a bill pending

8090
07:50:07.445 --> 07:50:11.525
<v Alex Bores (NY State Assembly Member, AD 73)>before the governor. So I should have
also added on ways you can help out is

8091
07:50:12.041 --> 07:50:15.800
<v Alex Bores (NY State Assembly Member, AD 73)>we are really happy to be here and advocate
for the legislation you are putting

8092
07:50:15.800 --> 07:50:19.720
<v Alex Bores (NY State Assembly Member, AD 73)>forward and invite you, if you agree with
some that's pending for the governor, to

8093
07:50:19.720 --> 07:50:21.400
<v Alex Bores (NY State Assembly Member, AD 73)>be a partner in pushing that as well.

8094
07:50:22.760 --> 07:50:27.035
<v Clyde Vanel (NY State Assembly Member, AD 33)>Yeah. We we have a number of deepfake bills,
and we passed actually a number of... A

8095
07:50:27.035 --> 07:50:30.635
<v Clyde Vanel (NY State Assembly Member, AD 33)>couple of my bills became law with respect
to deepfakes, with respect to elections.

8096
07:50:31.195 --> 07:50:34.715
<v Clyde Vanel (NY State Assembly Member, AD 33)>One of the one of the new bills that
just came out was just very recently,

8097
07:50:35.195 --> 07:50:41.459
<v Clyde Vanel (NY State Assembly Member, AD 33)>I I discovered an artist that I fell in
love with on on on on on Apple Music, and I

8098
07:50:41.459 --> 07:50:46.020
<v Clyde Vanel (NY State Assembly Member, AD 33)>downloaded all her music. I've been listening
it for... To her music for a while. I

8099
07:50:46.020 --> 07:50:49.780
<v Clyde Vanel (NY State Assembly Member, AD 33)>actually went to go look for her a couple
of days ago to follow her to go to a

8100
07:50:49.780 --> 07:50:51.621
<v Clyde Vanel (NY State Assembly Member, AD 33)>concert or something like that, found out

8101
07:50:51.940 --> 07:50:52.900
<v Clyde Vanel (NY State Assembly Member, AD 33)>it's not a person.

8102
07:50:54.455 --> 07:50:58.535
<v Clyde Vanel (NY State Assembly Member, AD 33)>So there's no law, there's no regulation
at all for these platforms to tell you

8103
07:50:58.535 --> 07:51:02.295
<v Clyde Vanel (NY State Assembly Member, AD 33)>whether or not you're listening to a real
person or not. Right? So so that, you

8104
07:51:02.295 --> 07:51:05.094
<v Clyde Vanel (NY State Assembly Member, AD 33)>know, the bill... So one of the things
that we have to do is figure out what that

8105
07:51:05.094 --> 07:51:07.734
<v Clyde Vanel (NY State Assembly Member, AD 33)>looks like moving forward. So we're gonna
need your help on that. That's something

8106
07:51:07.734 --> 07:51:10.990
<v Clyde Vanel (NY State Assembly Member, AD 33)>that we're gonna really need your help
on. Yeah. Ask them specifically a question

8107
07:51:10.990 --> 07:51:15.871
<v Clyde Vanel (NY State Assembly Member, AD 33)>about that, and it was very fluffy. It was...
But I literally asked a question about

8108
07:51:15.871 --> 07:51:17.390
<v Clyde Vanel (NY State Assembly Member, AD 33)>that to the companies.

8109
07:51:19.550 --> 07:51:20.912
<v CM Nantasha Williams>Thank you all. Thank you.

8110
07:51:21.550 --> 07:51:26.137
<v Speaker Julie Menin>One last question, from member of parliament,
Asada. Thank you. I wanted to ask you,

8111
07:51:26.137 --> 07:51:27.736
<v Speaker Julie Menin>as as you may know,

8112
07:51:28.295 --> 07:51:29.896
<v Speaker Julie Menin>when we called this hearing,

8113
07:51:29.977 --> 07:51:33.256
<v Speaker Julie Menin>we sent letters to five companies,

8114
07:51:33.336 --> 07:51:35.096
<v Speaker Julie Menin>to Anthropic, OpenAI,

8115
07:51:35.096 --> 07:51:38.600
<v Speaker Julie Menin>Meta, Google, and SpaceX AI. SpaceXAI

8116
07:51:38.600 --> 07:51:39.880
<v Speaker Julie Menin>is the only company

8117
07:51:40.041 --> 07:51:45.000
<v Speaker Julie Menin>that would not appear. We have subpoenaed
them. Given your personal story, I just

8118
07:51:45.000 --> 07:51:46.762
<v Speaker Julie Menin>wanted to get your reaction to that.

8119
07:51:48.680 --> 07:51:53.256
<v Jess Asato (UK Member of Parliament, Lowestoft)>Well, as I said in my opening statements,
it's a huge shame given that they are so

8120
07:51:53.256 --> 07:51:55.256
<v Jess Asato (UK Member of Parliament, Lowestoft)>close and I have come so far.

8121
07:51:55.977 --> 07:52:00.375
<v Jess Asato (UK Member of Parliament, Lowestoft)>But in the end, you've subpoenaed them,
and we look forward to seeing what then

8122
07:52:00.375 --> 07:52:01.016
<v Jess Asato (UK Member of Parliament, Lowestoft)>happens.

8123
07:52:01.416 --> 07:52:05.720
<v Jess Asato (UK Member of Parliament, Lowestoft)>In the end, everybody should be
accountable to elected representatives.

8124
07:52:06.040 --> 07:52:08.200
<v Jess Asato (UK Member of Parliament, Lowestoft)>And you're very powerful here in,

8125
07:52:08.680 --> 07:52:13.160
<v Jess Asato (UK Member of Parliament, Lowestoft)>the city of New York because of the fact
that these companies are based here.

8126
07:52:13.400 --> 07:52:17.559
<v Jess Asato (UK Member of Parliament, Lowestoft)>And therefore, those of us who are in other
countries across the world are looking

8127
07:52:17.559 --> 07:52:19.959
<v Jess Asato (UK Member of Parliament, Lowestoft)>to you to hold these companies to account,

8128
07:52:21.016 --> 07:52:25.574
<v Jess Asato (UK Member of Parliament, Lowestoft)>and we very much look forward to doing
so. I will do my own part in terms of my

8129
07:52:25.574 --> 07:52:27.256
<v Jess Asato (UK Member of Parliament, Lowestoft)>legal action in The UK.

8130
07:52:27.734 --> 07:52:28.535
<v Jess Asato (UK Member of Parliament, Lowestoft)>Thank you.

8131
07:52:29.654 --> 07:52:31.814
<v CM Carmen De La Rosa>Council member Hani followed by Maloney.

8132
07:52:31.975 --> 07:52:35.148
<v CM Shahana Hanif>Thank you, and thank you to
the member of parliament.

8133
07:52:35.549 --> 07:52:38.110
<v CM Shahana Hanif>Your experience really shows how

8134
07:52:38.350 --> 07:52:39.549
<v CM Shahana Hanif>generative AI,

8135
07:52:39.790 --> 07:52:44.510
<v CM Shahana Hanif>can be weaponized against women, queer
people, trans people, and particularly those

8136
07:52:44.510 --> 07:52:49.744
<v CM Shahana Hanif>of us in public life. And I'm super, super
grateful that you're using your platform

8137
07:52:49.744 --> 07:52:51.586
<v CM Shahana Hanif>as a legislator centering,

8138
07:52:51.984 --> 07:52:55.025
<v CM Shahana Hanif>gender justice and tech
surveillance. I think all day,

8139
07:52:55.824 --> 07:53:01.266
<v CM Shahana Hanif>it's in this panel that we're really able
to tease into that particular issue,

8140
07:53:01.426 --> 07:53:02.225
<v CM Shahana Hanif>which is devastating.

8141
07:53:04.590 --> 07:53:06.590
<v CM Shahana Hanif>There are so many survivors

8142
07:53:06.590 --> 07:53:08.430
<v CM Shahana Hanif>of AI generated,

8143
07:53:08.670 --> 07:53:11.389
<v CM Shahana Hanif>sexual abuse who have no recourse.

8144
07:53:11.709 --> 07:53:13.870
<v CM Shahana Hanif>And I know we touched on, several

8145
07:53:13.950 --> 07:53:17.629
<v CM Shahana Hanif>components of this issue in terms of
legislation policy. Could you speak more

8146
07:53:17.629 --> 07:53:22.275
<v CM Shahana Hanif>broadly about gender justice
as it comes to AI systems?

8147
07:53:22.275 --> 07:53:23.715
<v CM Shahana Hanif>I mean, certainly,

8148
07:53:24.195 --> 07:53:25.875
<v CM Shahana Hanif>we're gonna be regulating

8149
07:53:26.195 --> 07:53:27.715
<v CM Shahana Hanif>these systems,

8150
07:53:27.875 --> 07:53:29.875
<v CM Shahana Hanif>but the responsibility

8151
07:53:29.875 --> 07:53:30.516
<v CM Shahana Hanif>that,

8152
07:53:31.234 --> 07:53:32.834
<v CM Shahana Hanif>our communities have

8153
07:53:33.200 --> 07:53:37.120
<v CM Shahana Hanif>in addition to, of course, as
legislators passing stronger

8154
07:53:37.120 --> 07:53:37.920
<v CM Shahana Hanif>regulations?

8155
07:53:39.920 --> 07:53:41.520
<v Jess Asato (UK Member of Parliament, Lowestoft)>Thank you for your question.

8156
07:53:42.000 --> 07:53:46.959
<v Jess Asato (UK Member of Parliament, Lowestoft)>I think that one of the issues here is that
the whole of the online space has been

8157
07:53:46.959 --> 07:53:50.785
<v Jess Asato (UK Member of Parliament, Lowestoft)>developed without the sort of social norms
that we would actually expect of each

8158
07:53:50.785 --> 07:53:53.025
<v Jess Asato (UK Member of Parliament, Lowestoft)>other in our offline world.

8159
07:53:53.904 --> 07:53:58.625
<v Jess Asato (UK Member of Parliament, Lowestoft)>And therefore, when it comes to gender
justice, as you talk about it, that is

8160
07:53:58.625 --> 07:53:59.504
<v Jess Asato (UK Member of Parliament, Lowestoft)>ignored,

8161
07:53:59.984 --> 07:54:00.545
<v Jess Asato (UK Member of Parliament, Lowestoft)>because,

8162
07:54:01.110 --> 07:54:04.390
<v Jess Asato (UK Member of Parliament, Lowestoft)>actually, the Internet and
these platforms were created

8163
07:54:04.790 --> 07:54:05.670
<v Jess Asato (UK Member of Parliament, Lowestoft)>by men

8164
07:54:05.910 --> 07:54:07.910
<v Jess Asato (UK Member of Parliament, Lowestoft)>in the main. They were not shaped,

8165
07:54:08.070 --> 07:54:11.350
<v Jess Asato (UK Member of Parliament, Lowestoft)>by women. They were not
shaped with children in mind.

8166
07:54:11.670 --> 07:54:13.510
<v Jess Asato (UK Member of Parliament, Lowestoft)>And therefore, we're having to

8167
07:54:13.750 --> 07:54:15.110
<v Jess Asato (UK Member of Parliament, Lowestoft)>try to reverse

8168
07:54:15.916 --> 07:54:17.355
<v Jess Asato (UK Member of Parliament, Lowestoft)>decades now

8169
07:54:17.596 --> 07:54:20.477
<v Jess Asato (UK Member of Parliament, Lowestoft)>of a wild west of development.

8170
07:54:21.115 --> 07:54:21.916
<v Jess Asato (UK Member of Parliament, Lowestoft)>But

8171
07:54:21.916 --> 07:54:26.076
<v Jess Asato (UK Member of Parliament, Lowestoft)>that doesn't mean that it's impossible to
do, and I think we're seeing that actually

8172
07:54:26.156 --> 07:54:30.076
<v Jess Asato (UK Member of Parliament, Lowestoft)>our communities are actually fighting
back. And one of the things that has been

8173
07:54:30.076 --> 07:54:30.555
<v Jess Asato (UK Member of Parliament, Lowestoft)>really

8174
07:54:31.521 --> 07:54:35.361
<v Jess Asato (UK Member of Parliament, Lowestoft)>sort of fantastic in terms of my own case
is the number of people coming forward

8175
07:54:35.361 --> 07:54:36.160
<v Jess Asato (UK Member of Parliament, Lowestoft)>saying,

8176
07:54:36.400 --> 07:54:38.560
<v Jess Asato (UK Member of Parliament, Lowestoft)>I'm so glad that somebody,

8177
07:54:38.640 --> 07:54:42.880
<v Jess Asato (UK Member of Parliament, Lowestoft)>somewhere is trying to stand up for
those of us who think that this is wrong.

8178
07:54:44.035 --> 07:54:48.754
<v Jess Asato (UK Member of Parliament, Lowestoft)>And we don't want to live in this space
where so many people are being discriminated

8179
07:54:48.754 --> 07:54:49.555
<v Jess Asato (UK Member of Parliament, Lowestoft)>against.

8180
07:54:49.955 --> 07:54:53.074
<v Jess Asato (UK Member of Parliament, Lowestoft)>Having said that, of course,
there were very, very strong,

8181
07:54:54.514 --> 07:54:55.234
<v Jess Asato (UK Member of Parliament, Lowestoft)>you know,

8182
07:54:55.715 --> 07:54:56.355
<v Jess Asato (UK Member of Parliament, Lowestoft)>sort of

8183
07:54:57.080 --> 07:54:58.040
<v Jess Asato (UK Member of Parliament, Lowestoft)>positions

8184
07:54:58.040 --> 07:55:00.040
<v Jess Asato (UK Member of Parliament, Lowestoft)>taken out there around

8185
07:55:00.040 --> 07:55:03.879
<v Jess Asato (UK Member of Parliament, Lowestoft)>freedom of speech. And what
I would say is is that when

8186
07:55:04.520 --> 07:55:06.840
<v Jess Asato (UK Member of Parliament, Lowestoft)>women and other minorities

8187
07:55:06.840 --> 07:55:09.000
<v Jess Asato (UK Member of Parliament, Lowestoft)>decide not to be in a space,

8188
07:55:09.836 --> 07:55:12.396
<v Jess Asato (UK Member of Parliament, Lowestoft)>they lose their freedom of speech.

8189
07:55:12.875 --> 07:55:17.355
<v Jess Asato (UK Member of Parliament, Lowestoft)>And I think it's something that we simply
aren't recognizing enough because freedom

8190
07:55:17.355 --> 07:55:21.115
<v Jess Asato (UK Member of Parliament, Lowestoft)>of speech should not be
about the loudest voices,

8191
07:55:21.436 --> 07:55:25.620
<v Jess Asato (UK Member of Parliament, Lowestoft)>the ones who have the most money or the
most power. Freedom of speech should be

8192
07:55:25.620 --> 07:55:27.140
<v Jess Asato (UK Member of Parliament, Lowestoft)>about all of us equally

8193
07:55:27.220 --> 07:55:30.900
<v Jess Asato (UK Member of Parliament, Lowestoft)>being able to be in these
spaces on an equal basis.

8194
07:55:31.540 --> 07:55:34.740
<v Jess Asato (UK Member of Parliament, Lowestoft)>So that's what I hope we're
all fighting for today.

8195
07:55:40.895 --> 07:55:43.775
<v CM Carmen De La Rosa>Thank you so much. Councilmember Maloney?

8196
07:55:45.215 --> 07:55:47.936
<v CM Virginia Maloney>Thank you, chair, and thank
you all for being here.

8197
07:55:48.176 --> 07:55:53.799
<v CM Virginia Maloney>A special welcome to my state colleagues
who also represent the constituents of

8198
07:55:53.799 --> 07:55:58.680
<v CM Virginia Maloney>District 4, assembly member Alex Boras
and, state senator Gonzales who was here

8199
07:55:58.680 --> 07:55:59.320
<v CM Virginia Maloney>before.

8200
07:55:59.719 --> 07:56:04.120
<v CM Virginia Maloney>And I also wanna thank you for your work
in Albany, to pass the RAISE Act. Because

8201
07:56:04.120 --> 07:56:09.266
<v CM Virginia Maloney>of you, New York was the first in the nation
to pass a Frontier AI safety bill that

8202
07:56:09.266 --> 07:56:10.227
<v CM Virginia Maloney>went into law.

8203
07:56:10.865 --> 07:56:16.465
<v CM Virginia Maloney>But I wanted to ask about the process. The
original bill had penalties of 10,000,000

8204
07:56:16.465 --> 07:56:17.666
<v CM Virginia Maloney>up to 30,000,000.

8205
07:56:18.400 --> 07:56:22.240
<v CM Virginia Maloney>And the final version is looking
more like 1,000,000 to 3,000,000.

8206
07:56:22.560 --> 07:56:26.960
<v CM Virginia Maloney>These companies are able to spend
millions of dollars training models.

8207
07:56:27.600 --> 07:56:30.775
<v CM Virginia Maloney>What what teeth does this bill

8208
07:56:30.775 --> 07:56:33.896
<v CM Virginia Maloney>actually have now as law?
And can you speak to why,

8209
07:56:34.295 --> 07:56:37.576
<v CM Virginia Maloney>those fines were reduced? Open
to anyone who'd like to comment.

8210
07:56:38.535 --> 07:56:42.135
<v Alex Bores (NY State Assembly Member, AD 73)>Those fines were reduced
because the companies

8211
07:56:42.455 --> 07:56:46.590
<v Alex Bores (NY State Assembly Member, AD 73)>insisted that they would be paying
similar fines in every municipality

8212
07:56:46.590 --> 07:56:51.390
<v Alex Bores (NY State Assembly Member, AD 73)>and every state because, obviously, every
state municipality would pass a copycat

8213
07:56:51.390 --> 07:56:55.950
<v Alex Bores (NY State Assembly Member, AD 73)>law. The fact that here we are and there
are three states that have passed such a

8214
07:56:55.950 --> 07:56:58.750
<v Alex Bores (NY State Assembly Member, AD 73)>law, suggests that they significantly

8215
07:56:58.750 --> 07:57:00.590
<v Alex Bores (NY State Assembly Member, AD 73)>underestimated their own lobbying power,

8216
07:57:01.176 --> 07:57:05.816
<v Alex Bores (NY State Assembly Member, AD 73)>but also I think gives credence to the
effort of what you are doing here today of

8217
07:57:05.816 --> 07:57:09.816
<v Alex Bores (NY State Assembly Member, AD 73)>holding them accountable and that multiple
municipalities and jurisdictions should

8218
07:57:09.816 --> 07:57:13.256
<v Alex Bores (NY State Assembly Member, AD 73)>do that. 1 to $3,000,000 is
nothing for these companies.

8219
07:57:13.810 --> 07:57:17.890
<v Alex Bores (NY State Assembly Member, AD 73)>Absolutely nothing. It's about what
they are gonna pay just to register

8220
07:57:17.890 --> 07:57:19.570
<v Alex Bores (NY State Assembly Member, AD 73)>in New York to,

8221
07:57:19.810 --> 07:57:24.210
<v Alex Bores (NY State Assembly Member, AD 73)>the DIGIT office that will be regulating
them. The... Depending on the number of

8222
07:57:24.210 --> 07:57:26.850
<v Alex Bores (NY State Assembly Member, AD 73)>companies that qualify,
it'll be 1 to $2,000,000.

8223
07:57:27.625 --> 07:57:32.904
<v Alex Bores (NY State Assembly Member, AD 73)>So that is not an amount that provides
an incentive strong enough to change

8224
07:57:32.904 --> 07:57:33.705
<v Alex Bores (NY State Assembly Member, AD 73)>behavior.

8225
07:57:34.584 --> 07:57:35.225
<v Alex Bores (NY State Assembly Member, AD 73)>Maybe

8226
07:57:36.186 --> 07:57:39.064
<v Alex Bores (NY State Assembly Member, AD 73)>private rights of action, maybe
whistleblower protections,

8227
07:57:39.960 --> 07:57:43.960
<v Alex Bores (NY State Assembly Member, AD 73)>maybe requiring outside audits, maybe a
lot of the things that this council is

8228
07:57:43.960 --> 07:57:44.762
<v Alex Bores (NY State Assembly Member, AD 73)>considering

8229
07:57:44.840 --> 07:57:45.400
<v Alex Bores (NY State Assembly Member, AD 73)>will,

8230
07:57:45.800 --> 07:57:48.441
<v Alex Bores (NY State Assembly Member, AD 73)>but certainly a million
to 3,000,000 won't.

8231
07:57:49.720 --> 07:57:53.880
<v Unidentified State Senator (NY State Senate) [?]>I'll just add to that. I totally agree
with what as a member Boris said.

8232
07:57:54.564 --> 07:57:58.564
<v Unidentified State Senator (NY State Senate) [?]>The the one, I think, novel thing
about the DIGIT office that we created

8233
07:57:58.965 --> 07:58:03.365
<v Unidentified State Senator (NY State Senate) [?]>as part of the RAISE Act is that
we empowered the state regulator

8234
07:58:03.523 --> 07:58:05.604
<v Unidentified State Senator (NY State Senate) [?]>to not only receive

8235
07:58:05.604 --> 07:58:09.203
<v Unidentified State Senator (NY State Senate) [?]>the safety reports and the safety
plans of these companies and to,

8236
07:58:09.791 --> 07:58:13.470
<v Unidentified State Senator (NY State Senate) [?]>you know, assess the validity and
the legitimacy of those reports.

8237
07:58:13.550 --> 07:58:18.352
<v Unidentified State Senator (NY State Senate) [?]>But also, we empower them to propose
and and study the issue of continuing

8238
07:58:18.352 --> 07:58:23.230
<v Unidentified State Senator (NY State Senate) [?]>improvement to our AI safety laws so that
we develop the expertise over time about

8239
07:58:23.230 --> 07:58:26.484
<v Unidentified State Senator (NY State Senate) [?]>how we can make our laws
like the RAISE Act stronger.

8240
07:58:26.645 --> 07:58:27.525
<v Unidentified State Senator (NY State Senate) [?]>So today,

8241
07:58:27.686 --> 07:58:31.045
<v Unidentified State Senator (NY State Senate) [?]>that, you know, that function
does not do anything currently,

8242
07:58:31.205 --> 07:58:33.285
<v Unidentified State Senator (NY State Senate) [?]>but it is our hope and our expectation

8243
07:58:33.525 --> 07:58:35.846
<v Unidentified State Senator (NY State Senate) [?]>that as the RAISE Act is implemented,

8244
07:58:36.164 --> 07:58:39.846
<v Unidentified State Senator (NY State Senate) [?]>as the changes... This
technology become more evident,

8245
07:58:40.451 --> 07:58:42.210
<v Unidentified State Senator (NY State Senate) [?]>we develop a deeper

8246
07:58:42.291 --> 07:58:46.531
<v Unidentified State Senator (NY State Senate) [?]>state based, you know, as as a...
An administrative state expertise

8247
07:58:46.932 --> 07:58:51.970
<v Unidentified State Senator (NY State Senate) [?]>in this field to be able to propose far
reaching and ambitious guardrails for the

8248
07:58:51.970 --> 07:58:55.250
<v Unidentified State Senator (NY State Senate) [?]>next iteration and the next iteration
after that and not have to rely

8249
07:58:55.734 --> 07:58:59.895
<v Unidentified State Senator (NY State Senate) [?]>on, as we did early on, you know, hearing
from companies about what what they could

8250
07:58:59.895 --> 07:59:02.775
<v Unidentified State Senator (NY State Senate) [?]>or could not do and not have sufficient,

8251
07:59:03.254 --> 07:59:07.895
<v Unidentified State Senator (NY State Senate) [?]>for lack of a better word, firepower behind
us to always be able to answer back, in

8252
07:59:07.895 --> 07:59:11.980
<v Unidentified State Senator (NY State Senate) [?]>a way that would able to win over our our
critics. And so the digital office in the

8253
07:59:11.980 --> 07:59:16.541
<v Unidentified State Senator (NY State Senate) [?]>future should have this ability to play a
much larger role than it currently does.

8254
07:59:16.861 --> 07:59:21.340
<v Unidentified State Senator (NY State Senate) [?]>And and I I certainly look forward to seeing
that come to fruition as as we see the

8255
07:59:21.340 --> 07:59:22.460
<v Unidentified State Senator (NY State Senate) [?]>law be further implemented.

8256
07:59:26.824 --> 07:59:30.584
<v CM Carmen De La Rosa>Alright. Well, thank you all again for
being here and for taking the time and for

8257
07:59:30.584 --> 07:59:33.545
<v CM Carmen De La Rosa>your patience. We're gonna
call up the next panel.

8258
07:59:34.584 --> 07:59:35.705
<v CM Carmen De La Rosa>Eli Dorkin,

8259
07:59:36.860 --> 07:59:38.459
<v CM Carmen De La Rosa>Corinne Worthington,

8260
07:59:38.540 --> 07:59:39.979
<v CM Carmen De La Rosa>Noel Idalgo,

8261
07:59:40.139 --> 07:59:41.740
<v CM Carmen De La Rosa>Neil Gettnick,

8262
07:59:42.059 --> 07:59:43.420
<v CM Carmen De La Rosa>Gabe Weil,

8263
07:59:43.578 --> 07:59:44.700
<v CM Carmen De La Rosa>and Andrew

8264
07:59:45.260 --> 07:59:46.139
<v CM Carmen De La Rosa>Reich.

8265
07:59:47.180 --> 07:59:50.780
<v CM Carmen De La Rosa>And, again, if I mess up your names,
correct it for the record, and thank you.

8266
08:00:02.566 --> 08:00:04.807
<v CM Carmen De La Rosa>And Matt Henning, tech NYC.

8267
08:00:05.125 --> 08:00:06.566
<v CM Carmen De La Rosa>Please join this panel as well.

8268
08:00:30.205 --> 08:00:31.727
<v CM Carmen De La Rosa>Matt, you can begin.

8269
08:00:32.045 --> 08:00:36.330
<v CM Carmen De La Rosa>Identify yourselves for the record just
so we know who you are, and you can begin

8270
08:00:36.330 --> 08:00:37.930
<v CM Carmen De La Rosa>when you're ready. Yep.

8271
08:00:38.170 --> 08:00:42.250
<v Matt Henning (Tech:NYC)>Good evening, speaker Menon, chair De La
Rosa, and members of the council. My name

8272
08:00:42.250 --> 08:00:45.610
<v Matt Henning (Tech:NYC)>is Matt Henning, and I'm the director
of government affairs at Tech NYC.

8273
08:00:45.850 --> 08:00:47.850
<v Matt Henning (Tech:NYC)>Thank you for the invitation
to testify today.

8274
08:00:48.445 --> 08:00:53.006
<v Matt Henning (Tech:NYC)>Tech NYC's Frontier AI members already
testified today, and we fully support their

8275
08:00:53.006 --> 08:00:55.805
<v Matt Henning (Tech:NYC)>testimony and are proud to
call them Tech NYC members.

8276
08:00:56.125 --> 08:00:59.645
<v Matt Henning (Tech:NYC)>Today, I want to use my time to speak for
the rest of our membership, the startups

8277
08:00:59.645 --> 08:01:03.020
<v Matt Henning (Tech:NYC)>and businesses that build products
on top of frontier models

8278
08:01:03.100 --> 08:01:04.940
<v Matt Henning (Tech:NYC)>rather than create them themselves.

8279
08:01:05.340 --> 08:01:07.980
<v Matt Henning (Tech:NYC)>New York's tech ecosystem
has never been stronger.

8280
08:01:08.220 --> 08:01:12.860
<v Matt Henning (Tech:NYC)>According to a recent CBRE report, New York
is now the largest tech talent market in

8281
08:01:12.860 --> 08:01:13.420
<v Matt Henning (Tech:NYC)>North America,

8282
08:01:13.965 --> 08:01:17.965
<v Matt Henning (Tech:NYC)>passing the Bay Area for the first time
in the report's thirteen year history.

8283
08:01:18.365 --> 08:01:20.205
<v Matt Henning (Tech:NYC)>That momentum isn't guaranteed,

8284
08:01:20.285 --> 08:01:24.285
<v Matt Henning (Tech:NYC)>and the messaging that the city government
sends matters. That's why I was happy to

8285
08:01:24.285 --> 08:01:28.021
<v Matt Henning (Tech:NYC)>hear speaker Menon's comments to that
effect in her opening remarks and hope the

8286
08:01:28.021 --> 08:01:30.342
<v Matt Henning (Tech:NYC)>final language of this packet... Package

8287
08:01:30.580 --> 08:01:32.102
<v Matt Henning (Tech:NYC)>aligns with that intent.

8288
08:01:32.820 --> 08:01:36.820
<v Matt Henning (Tech:NYC)>Our biggest concern is that this package
holds non frontier companies liable

8289
08:01:36.980 --> 08:01:40.262
<v Matt Henning (Tech:NYC)>for things out of their
control. Most AI startups

8290
08:01:40.484 --> 08:01:42.805
<v Matt Henning (Tech:NYC)>and many larger non frontier companies

8291
08:01:42.805 --> 08:01:44.805
<v Matt Henning (Tech:NYC)>have a frontier model under the hood.

8292
08:01:45.125 --> 08:01:48.885
<v Matt Henning (Tech:NYC)>The difference between a deployer and a
developer is something I want to highlight

8293
08:01:48.885 --> 08:01:52.645
<v Matt Henning (Tech:NYC)>when thinking about the various requirements
and liabilities this package imposes.

8294
08:01:53.460 --> 08:01:57.701
<v Matt Henning (Tech:NYC)>Under the third party validation bill, a
startup can't deploy that model unless the

8295
08:01:57.701 --> 08:01:59.941
<v Matt Henning (Tech:NYC)>developer had it validated and assessed,

8296
08:02:00.102 --> 08:02:02.900
<v Matt Henning (Tech:NYC)>which these downstream companies

8297
08:02:02.900 --> 08:02:04.980
<v Matt Henning (Tech:NYC)>can't commission or even verify.

8298
08:02:05.635 --> 08:02:09.395
<v Matt Henning (Tech:NYC)>If the developer is found to not be in
compliance, the startup's product becomes

8299
08:02:09.395 --> 08:02:11.955
<v Matt Henning (Tech:NYC)>illegal overnight at $25,000

8300
08:02:11.955 --> 08:02:12.834
<v Matt Henning (Tech:NYC)>per instance.

8301
08:02:13.154 --> 08:02:14.596
<v Matt Henning (Tech:NYC)>The AI misuse,

8302
08:02:14.674 --> 08:02:18.914
<v Matt Henning (Tech:NYC)>chatbot, and advertising bills follow the
same pattern and the civilian complaint

8303
08:02:18.914 --> 08:02:20.275
<v Matt Henning (Tech:NYC)>bill multiplies the exposure.

8304
08:02:21.100 --> 08:02:21.980
<v Matt Henning (Tech:NYC)>Finally,

8305
08:02:22.139 --> 08:02:25.740
<v Matt Henning (Tech:NYC)>AI safety standards are most
effective at the international,

8306
08:02:25.818 --> 08:02:27.740
<v Matt Henning (Tech:NYC)>federal, or multistate level.

8307
08:02:28.139 --> 08:02:31.818
<v Matt Henning (Tech:NYC)>A city by city patchwork with different
rules in each city would be a compliance

8308
08:02:31.818 --> 08:02:35.180
<v Matt Henning (Tech:NYC)>nightmare that many startups and
small businesses could not survive,

8309
08:02:35.715 --> 08:02:38.436
<v Matt Henning (Tech:NYC)>not only for New York City HQ startups,

8310
08:02:38.516 --> 08:02:41.875
<v Matt Henning (Tech:NYC)>but for non New York City HQ
startups that serve New Yorkers.

8311
08:02:42.115 --> 08:02:46.195
<v Matt Henning (Tech:NYC)>As drafted, we are truly worried this
would result in startups not choosing but

8312
08:02:46.195 --> 08:02:47.955
<v Matt Henning (Tech:NYC)>being forced to leave New York City.

8313
08:02:48.740 --> 08:02:52.660
<v Matt Henning (Tech:NYC)>We strongly urge the council to engage
experts at both frontier and non frontier

8314
08:02:52.660 --> 08:02:57.300
<v Matt Henning (Tech:NYC)>companies on moving forward with this
package to ensure drastic negative unintended

8315
08:02:57.300 --> 08:02:58.580
<v Matt Henning (Tech:NYC)>consequences are in...

8316
08:02:59.700 --> 08:03:03.220
<v Matt Henning (Tech:NYC)>Tech NYC stands out to the ready to
help facilitate those conversations.

8317
08:03:03.846 --> 08:03:04.646
<v Matt Henning (Tech:NYC)>Thank you.

8318
08:03:10.326 --> 08:03:12.166
<v Andrew Rasiej (Civic Hall)>Thank you, madam speaker,

8319
08:03:13.686 --> 08:03:17.125
<v Andrew Rasiej (Civic Hall)>chair Dela Rosa, and members of the
city council. My name is Andrew Rachey.

8320
08:03:17.750 --> 08:03:20.230
<v Andrew Rasiej (Civic Hall)>Nearly thirty years ago,
I founded mouse.org

8321
08:03:20.230 --> 08:03:23.590
<v Andrew Rasiej (Civic Hall)>to bring the Internet to New York City
public schools. More recently, I founded

8322
08:03:23.590 --> 08:03:26.950
<v Andrew Rasiej (Civic Hall)>Civic Hall at Union Square, the city's
largest digital skills training center

8323
08:03:26.950 --> 08:03:28.871
<v Andrew Rasiej (Civic Hall)>serving underestimated New Yorkers.

8324
08:03:29.670 --> 08:03:34.025
<v Andrew Rasiej (Civic Hall)>New Yorkers are scared by the reports that
some AI models evaded human control. I

8325
08:03:34.025 --> 08:03:35.064
<v Andrew Rasiej (Civic Hall)>understand why.

8326
08:03:35.385 --> 08:03:38.105
<v Andrew Rasiej (Civic Hall)>But most people can't tell
you how a chatbot works,

8327
08:03:38.506 --> 08:03:41.865
<v Andrew Rasiej (Civic Hall)>what it does with their data,
or how a deepfake gets made.

8328
08:03:42.186 --> 08:03:45.064
<v Andrew Rasiej (Civic Hall)>They know enough to be alarmed, but
not enough to protect themselves.

8329
08:03:45.791 --> 08:03:50.350
<v Andrew Rasiej (Civic Hall)>And that's the new digital divide. People
who understand AI on one side and people

8330
08:03:50.350 --> 08:03:52.271
<v Andrew Rasiej (Civic Hall)>it happens to on the other.

8331
08:03:52.750 --> 08:03:55.390
<v Andrew Rasiej (Civic Hall)>I'm asking the council
to act on two fronts.

8332
08:03:56.031 --> 08:04:00.975
<v Andrew Rasiej (Civic Hall)>First, pass legislation requiring the city
to launch a free AI literacy campaign for

8333
08:04:00.975 --> 08:04:02.176
<v Andrew Rasiej (Civic Hall)>every New Yorker.

8334
08:04:02.574 --> 08:04:04.654
<v Andrew Rasiej (Civic Hall)>Teach people how to protect their jobs.

8335
08:04:04.895 --> 08:04:08.494
<v Andrew Rasiej (Civic Hall)>AI won't take your job. A person
who knows how to use AI will.

8336
08:04:08.895 --> 08:04:12.334
<v Andrew Rasiej (Civic Hall)>Teach families that when grandma gets a
call from her grandson's voice asking for

8337
08:04:12.334 --> 08:04:13.695
<v Andrew Rasiej (Civic Hall)>money, it may not be him.

8338
08:04:14.371 --> 08:04:18.210
<v Andrew Rasiej (Civic Hall)>And helps small businesses because AI has
made phishing and fraud cheaper and more

8339
08:04:18.210 --> 08:04:18.932
<v Andrew Rasiej (Civic Hall)>convincing,

8340
08:04:19.250 --> 08:04:21.012
<v Andrew Rasiej (Civic Hall)>and the corner store isn't ready.

8341
08:04:21.650 --> 08:04:25.250
<v Andrew Rasiej (Civic Hall)>Finland built a free public course for more
than 2,000,000 people... Which more than

8342
08:04:25.250 --> 08:04:26.291
<v Andrew Rasiej (Civic Hall)>2,000,000 people have taken,

8343
08:04:27.506 --> 08:04:32.385
<v Andrew Rasiej (Civic Hall)>and and and it's reached a 170 countries.
During COVID, the city reached every

8344
08:04:32.385 --> 08:04:35.506
<v Andrew Rasiej (Civic Hall)>resident. We can do it again. Fear of AI

8345
08:04:35.666 --> 08:04:37.906
<v Andrew Rasiej (Civic Hall)>is not a substitute for comprehension.

8346
08:04:38.705 --> 08:04:42.705
<v Andrew Rasiej (Civic Hall)>Second, the only real way to
stop bad AI is to build good AI.

8347
08:04:43.390 --> 08:04:47.230
<v Andrew Rasiej (Civic Hall)>Right now, it's being built by a handful
of companies answerable to answerable

8348
08:04:47.230 --> 08:04:48.590
<v Andrew Rasiej (Civic Hall)>to their shareholders.

8349
08:04:48.910 --> 08:04:52.670
<v Andrew Rasiej (Civic Hall)>The organizations responsible for the public
good are nonprofits, community groups,

8350
08:04:52.670 --> 08:04:54.990
<v Andrew Rasiej (Civic Hall)>and city agencies are years behind.

8351
08:04:55.586 --> 08:04:58.945
<v Andrew Rasiej (Civic Hall)>Most can't afford the computing power, and
their staff haven't been trained to use

8352
08:04:58.945 --> 08:04:59.746
<v Andrew Rasiej (Civic Hall)>these tools.

8353
08:05:00.066 --> 08:05:03.826
<v Andrew Rasiej (Civic Hall)>When the big AI companies offer nonprofits
free computing power, it comes on their

8354
08:05:03.826 --> 08:05:06.945
<v Andrew Rasiej (Civic Hall)>terms with the organization's
data sitting on their servers.

8355
08:05:08.100 --> 08:05:11.940
<v Andrew Rasiej (Civic Hall)>Public good AI needs sovereign computing
power separate from the companies whose

8356
08:05:11.940 --> 08:05:15.780
<v Andrew Rasiej (Civic Hall)>business depends on gathering data. I helped
start the foundation for Civic I, which

8357
08:05:15.780 --> 08:05:18.660
<v Andrew Rasiej (Civic Hall)>offers a supercomputer to the
public interest at no cost.

8358
08:05:18.900 --> 08:05:22.244
<v Andrew Rasiej (Civic Hall)>Few of them have the staff who know how
to use it, so investing in public AI

8359
08:05:22.244 --> 08:05:23.045
<v Andrew Rasiej (Civic Hall)>infrastructure

8360
08:05:23.045 --> 08:05:25.846
<v Andrew Rasiej (Civic Hall)>is a mission that this
council should take up.

8361
08:05:26.564 --> 08:05:27.766
<v Andrew Rasiej (Civic Hall)>Thank you very much.

8362
08:05:28.324 --> 08:05:29.125
<v CM Carmen De La Rosa>Thank you.

8363
08:05:30.324 --> 08:05:30.805
<v CM Carmen De La Rosa>Go ahead.

8364
08:05:33.070 --> 08:05:35.470
<v Neil Getnick (Getnick &amp; Getnick)>Good evening. I'm Neil. Oh, there you go.

8365
08:05:36.590 --> 08:05:38.750
<v Neil Getnick (Getnick &amp; Getnick)>Good evening. I'm Neil Gettnick,

8366
08:05:38.830 --> 08:05:42.430
<v Neil Getnick (Getnick &amp; Getnick)>the managing partner of
Manhattan based Gettnick Law.

8367
08:05:43.150 --> 08:05:48.535
<v Neil Getnick (Getnick &amp; Getnick)>Our whistleblower cases have
recovered over $1,500,000,000

8368
08:05:48.535 --> 08:05:51.096
<v Neil Getnick (Getnick &amp; Getnick)>for federal, state, and local governments,

8369
08:05:51.816 --> 08:05:53.896
<v Neil Getnick (Getnick &amp; Getnick)>including over a $100,000,000

8370
08:05:53.896 --> 08:05:56.615
<v Neil Getnick (Getnick &amp; Getnick)>for New York State and New York City.

8371
08:05:57.336 --> 08:06:02.535
<v Neil Getnick (Getnick &amp; Getnick)>Intro number two six zero five provides
for rewards resulting from recoveries

8372
08:06:03.130 --> 08:06:07.932
<v Neil Getnick (Getnick &amp; Getnick)>arising from whistleblower complaints
alleging that a person or entity

8373
08:06:08.170 --> 08:06:12.890
<v Neil Getnick (Getnick &amp; Getnick)>has violated legal provisions
related to artificial intelligence.

8374
08:06:13.531 --> 08:06:16.730
<v Neil Getnick (Getnick &amp; Getnick)>Speaking from my area of
whistleblower expertise,

8375
08:06:17.295 --> 08:06:18.496
<v Neil Getnick (Getnick &amp; Getnick)>I will highlight

8376
08:06:18.654 --> 08:06:20.336
<v Neil Getnick (Getnick &amp; Getnick)>six key elements

8377
08:06:20.654 --> 08:06:23.295
<v Neil Getnick (Getnick &amp; Getnick)>relevant to these legislative proposals.

8378
08:06:23.775 --> 08:06:24.416
<v Neil Getnick (Getnick &amp; Getnick)>First,

8379
08:06:24.975 --> 08:06:25.695
<v Neil Getnick (Getnick &amp; Getnick)>threshold,

8380
08:06:26.256 --> 08:06:28.814
<v Neil Getnick (Getnick &amp; Getnick)>setting a threshold qualifying

8381
08:06:28.814 --> 08:06:29.615
<v Neil Getnick (Getnick &amp; Getnick)>amount

8382
08:06:29.615 --> 08:06:31.055
<v Neil Getnick (Getnick &amp; Getnick)>for a reward.

8383
08:06:31.940 --> 08:06:32.660
<v Neil Getnick (Getnick &amp; Getnick)>Second,

8384
08:06:32.818 --> 08:06:33.620
<v Neil Getnick (Getnick &amp; Getnick)>floor

8385
08:06:33.700 --> 08:06:36.820
<v Neil Getnick (Getnick &amp; Getnick)>and ceiling, typically
the range of rewards,

8386
08:06:36.979 --> 08:06:38.580
<v Neil Getnick (Getnick &amp; Getnick)>is set between 15

8387
08:06:38.658 --> 08:06:39.940
<v Neil Getnick (Getnick &amp; Getnick)>to 30%.

8388
08:06:40.340 --> 08:06:42.180
<v Neil Getnick (Getnick &amp; Getnick)>Third, no caps.

8389
08:06:42.500 --> 08:06:47.504
<v Neil Getnick (Getnick &amp; Getnick)>Beyond the percentage limits, there
should be no cap on the size of the award

8390
08:06:47.584 --> 08:06:48.785
<v Neil Getnick (Getnick &amp; Getnick)>maximizing

8391
08:06:48.785 --> 08:06:50.225
<v Neil Getnick (Getnick &amp; Getnick)>incentivization.

8392
08:06:50.705 --> 08:06:51.346
<v Neil Getnick (Getnick &amp; Getnick)>Fourth,

8393
08:06:51.744 --> 08:06:52.465
<v Neil Getnick (Getnick &amp; Getnick)>public

8394
08:06:52.545 --> 08:06:54.225
<v Neil Getnick (Getnick &amp; Getnick)>private partnership.

8395
08:06:54.385 --> 08:06:59.344
<v Neil Getnick (Getnick &amp; Getnick)>The government agency and the whistleblower
and council should work together to

8396
08:06:59.344 --> 08:06:59.904
<v Neil Getnick (Getnick &amp; Getnick)>maximize

8397
08:07:00.360 --> 08:07:01.400
<v Neil Getnick (Getnick &amp; Getnick)>the recovery.

8398
08:07:01.799 --> 08:07:02.520
<v Neil Getnick (Getnick &amp; Getnick)>Fifth,

8399
08:07:02.760 --> 08:07:03.879
<v Neil Getnick (Getnick &amp; Getnick)>emphasizing

8400
08:07:03.959 --> 08:07:05.238
<v Neil Getnick (Getnick &amp; Getnick)>anonymity.

8401
08:07:05.238 --> 08:07:06.600
<v Neil Getnick (Getnick &amp; Getnick)>The government should do

8402
08:07:06.920 --> 08:07:10.520
<v Neil Getnick (Getnick &amp; Getnick)>all that it can to prove...
Preserve the whistleblower's

8403
08:07:10.520 --> 08:07:11.639
<v Neil Getnick (Getnick &amp; Getnick)>anonymity.

8404
08:07:12.040 --> 08:07:12.840
<v Neil Getnick (Getnick &amp; Getnick)>Sixth,

8405
08:07:12.840 --> 08:07:14.680
<v Neil Getnick (Getnick &amp; Getnick)>perhaps most importantly

8406
08:07:15.000 --> 08:07:16.896
<v Neil Getnick (Getnick &amp; Getnick)>for this council,

8407
08:07:17.135 --> 08:07:17.936
<v Neil Getnick (Getnick &amp; Getnick)>clarity.

8408
08:07:18.416 --> 08:07:20.736
<v Neil Getnick (Getnick &amp; Getnick)>In passing artificial intelligence

8409
08:07:20.736 --> 08:07:21.535
<v Neil Getnick (Getnick &amp; Getnick)>laws,

8410
08:07:21.695 --> 08:07:23.775
<v Neil Getnick (Getnick &amp; Getnick)>you will be taking on powerful

8411
08:07:23.775 --> 08:07:26.096
<v Neil Getnick (Getnick &amp; Getnick)>and well resourced individuals

8412
08:07:26.096 --> 08:07:27.295
<v Neil Getnick (Getnick &amp; Getnick)>and entities

8413
08:07:27.375 --> 08:07:28.816
<v Neil Getnick (Getnick &amp; Getnick)>who will push back.

8414
08:07:29.215 --> 08:07:31.200
<v Neil Getnick (Getnick &amp; Getnick)>So ask yourself,

8415
08:07:31.440 --> 08:07:33.120
<v Neil Getnick (Getnick &amp; Getnick)>are these laws

8416
08:07:33.200 --> 08:07:35.520
<v Neil Getnick (Getnick &amp; Getnick)>clear? And then do what's necessary

8417
08:07:35.600 --> 08:07:41.200
<v Neil Getnick (Getnick &amp; Getnick)>to accomplish that at the drafting stage
rather than leaving it to the courts.

8418
08:07:41.440 --> 08:07:42.240
<v Neil Getnick (Getnick &amp; Getnick)>In conclusion,

8419
08:07:42.676 --> 08:07:46.195
<v Neil Getnick (Getnick &amp; Getnick)>the legislative proposals
before you are an opportunity

8420
08:07:46.355 --> 08:07:51.236
<v Neil Getnick (Getnick &amp; Getnick)>for the city of New York to lead
the way in safeguarding citizens

8421
08:07:51.316 --> 08:07:53.396
<v Neil Getnick (Getnick &amp; Getnick)>from the potential risks

8422
08:07:53.396 --> 08:07:54.596
<v Neil Getnick (Getnick &amp; Getnick)>of artificial

8423
08:07:54.596 --> 08:07:55.637
<v Neil Getnick (Getnick &amp; Getnick)>intelligence.

8424
08:07:55.875 --> 08:07:56.916
<v Neil Getnick (Getnick &amp; Getnick)>And in doing so,

8425
08:07:57.730 --> 08:08:02.291
<v Neil Getnick (Getnick &amp; Getnick)>this council will light the
way for the nation as a whole.

8426
08:08:03.330 --> 08:08:04.130
<v CM Carmen De La Rosa>Thank you.

8427
08:08:05.810 --> 08:08:08.130
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>Thank you, speaker Menon
and and chair Dilrosa.

8428
08:08:08.130 --> 08:08:12.610
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>I'm Gabriel Weil. I'm a professor at the
University of Houston and a fellow at the

8429
08:08:12.610 --> 08:08:13.570
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>Institute for Law and AI.

8430
08:08:14.186 --> 08:08:15.945
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>And I wanna talk to you about,

8431
08:08:16.744 --> 08:08:20.746
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>what I see is the the largest gap
in the existing liability law,

8432
08:08:21.145 --> 08:08:24.426
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>that is not addressed by
your, legislative package,

8433
08:08:24.744 --> 08:08:26.186
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>which is that AI agents

8434
08:08:26.510 --> 08:08:31.709
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>could engage in conduct that would be a tort
or even a crime for a human, and yet no

8435
08:08:31.709 --> 08:08:34.110
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>one could be liable under current law.

8436
08:08:34.430 --> 08:08:37.950
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>And that is a consequence of two
relevant features of our legal system.

8437
08:08:38.605 --> 08:08:41.406
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>One is that AI agents
are not legal persons.

8438
08:08:41.805 --> 08:08:45.564
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>They do not have tort duties, and that
means not only that they can't be held

8439
08:08:45.564 --> 08:08:46.285
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>liable,

8440
08:08:46.605 --> 08:08:50.926
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>but that their principles cannot be held
vicariously liable in the way that employer

8441
08:08:50.926 --> 08:08:55.170
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>is held liable for the torts of their
employees within the scope of employment.

8442
08:08:55.890 --> 08:08:58.691
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>And second, the regimes that do apply,

8443
08:08:59.010 --> 08:09:00.130
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>to these harms,

8444
08:09:00.451 --> 08:09:03.650
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>negligence and products
liability have important gaps.

8445
08:09:04.210 --> 08:09:08.324
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>They are inadequate to motivate some of
the key precautionary measures that would

8446
08:09:08.324 --> 08:09:09.844
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>prevent these kinds of incidents.

8447
08:09:10.484 --> 08:09:13.844
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>And so I... What I propose
is a is a simple fix,

8448
08:09:14.164 --> 08:09:16.404
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>principle that whenever an AI agent,

8449
08:09:16.645 --> 08:09:20.324
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>engages in conduct that would be
a tort or a crime, for a human,

8450
08:09:20.645 --> 08:09:26.201
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>someone's liable for that. And so, if
neither the user nor any intermediary

8451
08:09:26.201 --> 08:09:30.121
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>that fine tuned or scaffolded the model
either intended or was negligent with

8452
08:09:30.121 --> 08:09:33.640
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>respect to the conducts or no downstream
actor was at fault, then the buck should

8453
08:09:33.640 --> 08:09:37.480
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>stop with the developer, and they should
be liable regardless of the degree of care

8454
08:09:37.775 --> 08:09:41.615
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>they exercised. I have worked on legislation
to this effect. It's been introduced,

8455
08:09:41.775 --> 08:09:45.135
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>in New York State and in Rhode Island,
and I'm talking to several congressional

8456
08:09:45.135 --> 08:09:47.535
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>offices. My written testimony,
which will be submitted,

8457
08:09:47.855 --> 08:09:52.530
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>probably tomorrow, includes proposed
legislative text for New York, and I'm happy to

8458
08:09:52.530 --> 08:09:53.730
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>answer any questions,

8459
08:09:54.129 --> 08:09:57.969
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>about that or about any other, liability
issues including the legislation that's

8460
08:09:57.969 --> 08:09:59.970
<v Gabriel Weil (University of Houston / Institute for Law &amp; AI)>that's in the the package. Thank you.

8461
08:10:00.770 --> 08:10:02.450
<v CM Carmen De La Rosa>Thank you. Noah?

8462
08:10:08.846 --> 08:10:09.566
<v Noel Hidalgo (BetaNYC)>Hi.

8463
08:10:09.725 --> 08:10:14.125
<v Noel Hidalgo (BetaNYC)>My name is Noel Idelgo. I'm the executive
director of Beta NYC. And for the record,

8464
08:10:14.125 --> 08:10:18.125
<v Noel Hidalgo (BetaNYC)>I serve as the public advocate's appointee
to COPIC and to the Internet advisory

8465
08:10:18.125 --> 08:10:20.686
<v Noel Hidalgo (BetaNYC)>board. And today, I speak
on behalf of Beta NYC.

8466
08:10:21.160 --> 08:10:26.040
<v Noel Hidalgo (BetaNYC)>At Beta NYC, AI is a question of power.
Who has the power to consume the world's

8467
08:10:26.040 --> 08:10:27.400
<v Noel Hidalgo (BetaNYC)>knowledge and resources

8468
08:10:27.480 --> 08:10:29.560
<v Noel Hidalgo (BetaNYC)>and who controls the companies that do?

8469
08:10:29.880 --> 08:10:35.135
<v Noel Hidalgo (BetaNYC)>Our research shows four things. First, AI
models are snapshots in time. We asked

8470
08:10:35.135 --> 08:10:40.574
<v Noel Hidalgo (BetaNYC)>five AI agents about the council's own AI
law, local law one eighty eight. Of the

8471
08:10:40.574 --> 08:10:44.574
<v Noel Hidalgo (BetaNYC)>three that answered from memory, none of
them knew it. Two ignored the instructions

8472
08:10:44.574 --> 08:10:49.310
<v Noel Hidalgo (BetaNYC)>and searched the web to find the answer.
Out of the box, AI models are brains with

8473
08:10:49.310 --> 08:10:51.070
<v Noel Hidalgo (BetaNYC)>no eyes and no hands.

8474
08:10:51.390 --> 08:10:54.750
<v Noel Hidalgo (BetaNYC)>Second, access to the city's
information is inconsistent.

8475
08:10:54.910 --> 08:10:56.270
<v Noel Hidalgo (BetaNYC)>Nyc.gov,

8476
08:10:56.830 --> 08:10:58.590
<v Noel Hidalgo (BetaNYC)>nyc.gov's

8477
08:10:58.670 --> 08:11:04.006
<v Noel Hidalgo (BetaNYC)>one published rule welcomes every automated
visitor. Yet when you disclose that you

8478
08:11:04.006 --> 08:11:06.887
<v Noel Hidalgo (BetaNYC)>are who you are, it is refused.

8479
08:11:07.285 --> 08:11:12.086
<v Noel Hidalgo (BetaNYC)>So we had to build open source tools that
go straight to the source and hop around

8480
08:11:12.086 --> 08:11:14.566
<v Noel Hidalgo (BetaNYC)>the city's regulation...
Digital regulation.

8481
08:11:15.190 --> 08:11:18.790
<v Noel Hidalgo (BetaNYC)>Third, people inside a city government
are ready to experiment. Our own

8482
08:11:18.790 --> 08:11:19.750
<v Noel Hidalgo (BetaNYC)>conversations

8483
08:11:19.908 --> 08:11:22.950
<v Noel Hidalgo (BetaNYC)>say that they are ready to
do it safely and reliably,

8484
08:11:23.110 --> 08:11:26.229
<v Noel Hidalgo (BetaNYC)>and they have told us that they lack
the literacy and the resources. And you

8485
08:11:26.229 --> 08:11:30.525
<v Noel Hidalgo (BetaNYC)>actually heard the deputy commissioner say
that that they can't find somebody inside

8486
08:11:30.525 --> 08:11:34.205
<v Noel Hidalgo (BetaNYC)>of New York City government who actually
has the tools and the resources to become

8487
08:11:34.205 --> 08:11:37.006
<v Noel Hidalgo (BetaNYC)>the leader that they are looking
for, which is really disappointing.

8488
08:11:37.645 --> 08:11:40.926
<v Noel Hidalgo (BetaNYC)>Fourth, we have trained nearly
a 100 CUNY undergraduates,

8489
08:11:40.926 --> 08:11:44.110
<v Noel Hidalgo (BetaNYC)>and we interviewed 26 experts
across 18 organizations.

8490
08:11:44.190 --> 08:11:48.590
<v Noel Hidalgo (BetaNYC)>Employers want more than technical skills.
They want... And students want practical

8491
08:11:48.590 --> 08:11:49.310
<v Noel Hidalgo (BetaNYC)>experience.

8492
08:11:50.030 --> 08:11:54.990
<v Noel Hidalgo (BetaNYC)>A century ago, reformers in this chamber
helped bring us clean water and an end to

8493
08:11:54.990 --> 08:11:59.385
<v Noel Hidalgo (BetaNYC)>nepotism in government. A hundred years from
now, how will our descendants look back

8494
08:11:59.385 --> 08:12:04.025
<v Noel Hidalgo (BetaNYC)>at today? So we asked the council to work
with the state on a few fun pieces of

8495
08:12:04.025 --> 08:12:04.826
<v Noel Hidalgo (BetaNYC)>legislation.

8496
08:12:05.146 --> 08:12:10.025
<v Noel Hidalgo (BetaNYC)>One would be an AI wealth fund for public
universities and public education and to

8497
08:12:10.025 --> 08:12:10.906
<v Noel Hidalgo (BetaNYC)>fund enforcement.

8498
08:12:11.500 --> 08:12:13.980
<v Noel Hidalgo (BetaNYC)>We desperately need to revive COPIC,

8499
08:12:14.140 --> 08:12:17.740
<v Noel Hidalgo (BetaNYC)>define New York City's vision of digital
sovereignty starting with the UN open

8500
08:12:17.740 --> 08:12:18.700
<v Noel Hidalgo (BetaNYC)>source principles,

8501
08:12:19.020 --> 08:12:23.660
<v Noel Hidalgo (BetaNYC)>fund the office of algorithmic accountability
and the office of data strategy,

8502
08:12:23.820 --> 08:12:27.994
<v Noel Hidalgo (BetaNYC)>and create an open source programs office
in partnership with CUNY. Our full

8503
08:12:27.994 --> 08:12:29.834
<v Noel Hidalgo (BetaNYC)>testimony is up on our website at

8504
08:12:29.834 --> 08:12:36.475
<v Noel Hidalgo (BetaNYC)>beta.nyc/council-ai-hearings/-2020six.

8505
08:12:36.475 --> 08:12:37.115
<v Noel Hidalgo (BetaNYC)>Thank you.

8506
08:12:37.834 --> 08:12:38.395
<v CM Carmen De La Rosa>Thank you.

8507
08:12:40.041 --> 08:12:41.320
<v David S. Rose (public)>Speaker, chair,

8508
08:12:41.640 --> 08:12:42.842
<v David S. Rose (public)>members of the council,

8509
08:12:43.160 --> 08:12:48.041
<v David S. Rose (public)>my name is David s Rose, and I think my
background is important to understand for

8510
08:12:48.041 --> 08:12:49.002
<v David S. Rose (public)>why I'm here.

8511
08:12:49.400 --> 08:12:54.584
<v David S. Rose (public)>I founded my first technology
company in New York City in 1981,

8512
08:12:54.584 --> 08:12:56.584
<v David S. Rose (public)>more than forty five years ago,

8513
08:12:56.824 --> 08:13:00.023
<v David S. Rose (public)>and I was a founding member of
the New York New Media Association

8514
08:13:00.023 --> 08:13:01.545
<v David S. Rose (public)>in 1994.

8515
08:13:01.785 --> 08:13:05.625
<v David S. Rose (public)>Following the dot com crash, I founded
New York Angels, today one of the world's

8516
08:13:05.625 --> 08:13:09.920
<v David S. Rose (public)>leading business angel networks with
over a 150 members that has funded over

8517
08:13:09.920 --> 08:13:10.879
<v David S. Rose (public)>180,000,000

8518
08:13:10.879 --> 08:13:15.360
<v David S. Rose (public)>into more than 400 start ups, creating
thousands of jobs here in New York.

8519
08:13:15.760 --> 08:13:18.080
<v David S. Rose (public)>I then founded the start up platform Gust,

8520
08:13:18.238 --> 08:13:21.846
<v David S. Rose (public)>which was tapped by first mayor Bloomberg
and then mayor de Blasio to create

8521
08:13:21.846 --> 08:13:23.605
<v David S. Rose (public)>digital.nyc,

8522
08:13:23.686 --> 08:13:27.686
<v David S. Rose (public)>New York City's official online hub for
the city's entrepreneurship ecosystem.

8523
08:13:28.006 --> 08:13:32.086
<v David S. Rose (public)>Most recently, I'm the founder and CEO of
the US real estate market, a lappointed an

8524
08:13:32.086 --> 08:13:36.290
<v David S. Rose (public)>AI based liquidity platform for commercial
real estate. Red Herring magazine has

8525
08:13:36.290 --> 08:13:40.930
<v David S. Rose (public)>described me as patriarch of Silicon
Alley and Forbes as New York's archangel.

8526
08:13:41.410 --> 08:13:46.049
<v David S. Rose (public)>At the same time, I have spent twenty years
working on the future of technology and

8527
08:13:46.049 --> 08:13:50.285
<v David S. Rose (public)>artificial intelligence as associate
founder of Singularity University with Ray

8528
08:13:50.285 --> 08:13:52.205
<v David S. Rose (public)>Kurzweil and Peter Diamantis.

8529
08:13:52.205 --> 08:13:55.645
<v David S. Rose (public)>And today, I serve as chairman
of two applied AI companies.

8530
08:13:55.885 --> 08:13:59.004
<v David S. Rose (public)>My testimony today is gonna be
brief, and it's about three

8531
08:13:59.164 --> 08:14:03.762
<v David S. Rose (public)>fundamental facts that are the key takeaways
I think we all need to have. Number

8532
08:14:03.762 --> 08:14:09.842
<v David S. Rose (public)>one, artificial intelligence is today
baked into the core of every single company

8533
08:14:09.842 --> 08:14:15.361
<v David S. Rose (public)>starting up in New York City. No investor
today will invest in any company that does

8534
08:14:15.361 --> 08:14:19.605
<v David S. Rose (public)>not explicitly show how they are using
it as an essential foundation of their

8535
08:14:19.605 --> 08:14:20.244
<v David S. Rose (public)>business.

8536
08:14:20.725 --> 08:14:23.205
<v David S. Rose (public)>Number two, AI absolutely

8537
08:14:23.205 --> 08:14:24.564
<v David S. Rose (public)>needs to be regulated

8538
08:14:24.564 --> 08:14:28.805
<v David S. Rose (public)>because the power it has already unlocked,
not to mention what is coming in the

8539
08:14:28.805 --> 08:14:33.389
<v David S. Rose (public)>immediate future, is literally the biggest
development in the history of our planet

8540
08:14:33.389 --> 08:14:35.389
<v David S. Rose (public)>since the emergence of humans.

8541
08:14:36.110 --> 08:14:39.709
<v David S. Rose (public)>But really critical for the
hearing today is number three.

8542
08:14:40.030 --> 08:14:43.754
<v David S. Rose (public)>The most important thing for you to
consider in your deliberations is that

8543
08:14:43.754 --> 08:14:44.635
<v David S. Rose (public)>regulation

8544
08:14:44.635 --> 08:14:49.434
<v David S. Rose (public)>absolutely needs to be done either at the
federal level or at a coordinated state

8545
08:14:49.434 --> 08:14:53.834
<v David S. Rose (public)>level as New York state has done with
California in passing the RAISE Act. The

8546
08:14:53.834 --> 08:14:58.059
<v David S. Rose (public)>challenge with imposing additional requirements
at the city level, especially such

8547
08:14:58.059 --> 08:14:59.900
<v David S. Rose (public)>as those considered in the bill here,

8548
08:15:00.219 --> 08:15:04.700
<v David S. Rose (public)>will have the effect of literally destroying
New York City's role as a tech leader.

8549
08:15:04.940 --> 08:15:09.695
<v CM Carmen De La Rosa>No company could afford to offer conclude
and submit the full testimony. We have

8550
08:15:09.695 --> 08:15:14.016
<v CM Carmen De La Rosa>about seven more panels to go, so
we appreciate your time. Thank you.

8551
08:15:14.496 --> 08:15:18.416
<v Speaker Julie Menin>Thank you. Speaker Mena? Okay. Thank
you. Just a couple quick comments slash

8552
08:15:18.416 --> 08:15:23.880
<v Speaker Julie Menin>questions for Tech NYC. Thank you
for your, comments. We appreciate

8553
08:15:23.880 --> 08:15:27.240
<v Speaker Julie Menin>that. And as I said at the
beginning of the hearing,

8554
08:15:28.040 --> 08:15:34.121
<v Speaker Julie Menin>we very much want to ensure that New York
City stays the tech capital of the world.

8555
08:15:34.360 --> 08:15:38.654
<v Speaker Julie Menin>We wanna promote innovation. We want these
tech companies to locate here. We want

8556
08:15:38.654 --> 08:15:40.254
<v Speaker Julie Menin>startups to locate here.

8557
08:15:40.654 --> 08:15:41.535
<v Speaker Julie Menin>And so

8558
08:15:42.334 --> 08:15:48.494
<v Speaker Julie Menin>I I I understand your comment, and this is
why we have been very focused on hearing

8559
08:15:48.494 --> 08:15:53.432
<v Speaker Julie Menin>from the frontier companies as we did
today, which I think was really important.

8560
08:15:53.432 --> 08:15:53.990
<v Speaker Julie Menin>So,

8561
08:15:55.432 --> 08:16:00.230
<v Speaker Julie Menin>no... Noel? Is it Noel? Okay. Thank you.
You mentioned the office of algorithmic

8562
08:16:00.230 --> 08:16:01.271
<v Speaker Julie Menin>accountability.

8563
08:16:01.750 --> 08:16:05.385
<v Speaker Julie Menin>You know, as you might have
heard earlier when we had the

8564
08:16:05.385 --> 08:16:08.025
<v Speaker Julie Menin>city administration here in OTI,

8565
08:16:08.186 --> 08:16:13.145
<v Speaker Julie Menin>you know, I've been concerned that
that office is not up and running yet.

8566
08:16:14.346 --> 08:16:15.305
<v Speaker Julie Menin>Thoughts on

8567
08:16:15.545 --> 08:16:19.550
<v Speaker Julie Menin>what needs to be done there and how we can
quickly make sure that that office office

8568
08:16:19.550 --> 08:16:21.871
<v Speaker Julie Menin>is doing the critical
work we need it to do.

8569
08:16:22.990 --> 08:16:28.031
<v Noel Hidalgo (BetaNYC)>Thank you for that question. And thank
you for those point... Questions at OTI.

8570
08:16:28.590 --> 08:16:29.470
<v Noel Hidalgo (BetaNYC)>Inside...

8571
08:16:29.791 --> 08:16:33.230
<v Noel Hidalgo (BetaNYC)>So in the wonderful chaos that is OTI,

8572
08:16:33.470 --> 08:16:37.994
<v Noel Hidalgo (BetaNYC)>there is the office of data strategy that
help maintains the city's open data portal

8573
08:16:37.994 --> 08:16:40.395
<v Noel Hidalgo (BetaNYC)>and have been helping
maintain the data standards.

8574
08:16:40.635 --> 08:16:45.754
<v Noel Hidalgo (BetaNYC)>They're the team that right now that hold
the keys to the city's GitHub copilot,

8575
08:16:46.234 --> 08:16:46.795
<v Noel Hidalgo (BetaNYC)>capability.

8576
08:16:47.430 --> 08:16:52.070
<v Noel Hidalgo (BetaNYC)>So every single developer or technical
person in the city that wants to use

8577
08:16:52.310 --> 08:16:56.629
<v Noel Hidalgo (BetaNYC)>an authorized and approved
AI system already has

8578
08:16:56.629 --> 08:17:01.110
<v Noel Hidalgo (BetaNYC)>a team that's been trained and that's willing
to go and work and more or less do the

8579
08:17:01.416 --> 08:17:05.256
<v Noel Hidalgo (BetaNYC)>missionary work across the city to ensure
that the city's developers and the

8580
08:17:05.256 --> 08:17:10.137
<v Noel Hidalgo (BetaNYC)>technical teams have the the skills and
the capability to use AI safely so that way

8581
08:17:10.137 --> 08:17:12.695
<v Noel Hidalgo (BetaNYC)>they can start Yeah.
Growing their capability.

8582
08:17:13.256 --> 08:17:14.375
<v Noel Hidalgo (BetaNYC)>When it comes down to,

8583
08:17:16.129 --> 08:17:21.570
<v Noel Hidalgo (BetaNYC)>so so they have that framework in place.
The Office of Algorithmic Accountability is

8584
08:17:21.570 --> 08:17:23.570
<v Noel Hidalgo (BetaNYC)>supposed to be kind of that grander

8585
08:17:23.570 --> 08:17:25.170
<v Noel Hidalgo (BetaNYC)>scaled up version.

8586
08:17:25.410 --> 08:17:29.330
<v Noel Hidalgo (BetaNYC)>The problem that we have is that this is
about booting up a brand new office and a

8587
08:17:29.330 --> 08:17:31.890
<v Noel Hidalgo (BetaNYC)>brand new administration when priorities
have have been been changing. Changing.

8588
08:17:32.424 --> 08:17:34.025
<v Noel Hidalgo (BetaNYC)>And so the frustration

8589
08:17:34.104 --> 08:17:37.945
<v Noel Hidalgo (BetaNYC)>that I continue to have and
that I've heard from inside

8590
08:17:37.945 --> 08:17:39.625
<v Noel Hidalgo (BetaNYC)>is the fact that OMB

8591
08:17:39.705 --> 08:17:45.625
<v Noel Hidalgo (BetaNYC)>is holding back the ability for OTI to make
those hires and then to actually start

8592
08:17:45.780 --> 08:17:50.260
<v Noel Hidalgo (BetaNYC)>that office. But there's still gonna be
time. You've worked in government. You know

8593
08:17:50.260 --> 08:17:54.500
<v Noel Hidalgo (BetaNYC)>that it takes at least six months to get
those people kind of doing all those first

8594
08:17:54.578 --> 08:17:56.180
<v Noel Hidalgo (BetaNYC)>introductory conversations.

8595
08:17:56.500 --> 08:18:00.236
<v Speaker Julie Menin>We can skip that. There's already a team
in place that knows how to do that. Great.

8596
08:18:00.236 --> 08:18:04.236
<v Speaker Julie Menin>Thank you. And then lastly, sir, you
mentioned about the tort law, which I thought

8597
08:18:04.236 --> 08:18:05.275
<v Speaker Julie Menin>was very interesting.

8598
08:18:05.596 --> 08:18:07.436
<v Speaker Julie Menin>So on the question of

8599
08:18:07.756 --> 08:18:11.836
<v Speaker Julie Menin>your idea of expanding the tort law, I
did wanna say council member Maloney has a

8600
08:18:11.836 --> 08:18:16.000
<v Speaker Julie Menin>bill that would create a private right of
action. So we do have that bill as part of

8601
08:18:16.000 --> 08:18:17.521
<v Speaker Julie Menin>this legislative package.

8602
08:18:17.682 --> 08:18:21.121
<v Speaker Julie Menin>Yes. Not amending the tort law, but
creating that private right of action.

8603
08:18:22.080 --> 08:18:26.480
<v Neil Getnick (Getnick &amp; Getnick)>So a little different, but I appreciated
your comments. So... Before we adjourn, may

8604
08:18:26.480 --> 08:18:30.684
<v Neil Getnick (Getnick &amp; Getnick)>I say a brief word about the
validator question you raised earlier?

8605
08:18:31.324 --> 08:18:34.684
<v Speaker Julie Menin>Sure. And then I think we're gonna have to
move on because I know we have several be

8606
08:18:34.684 --> 08:18:37.645
<v Speaker Julie Menin>very quick. You already
have your mechanism.

8607
08:18:38.125 --> 08:18:39.484
<v Neil Getnick (Getnick &amp; Getnick)>It's the independent

8608
08:18:39.484 --> 08:18:42.203
<v Neil Getnick (Getnick &amp; Getnick)>private sector inspector general model

8609
08:18:42.480 --> 08:18:44.720
<v Neil Getnick (Getnick &amp; Getnick)>that this council used

8610
08:18:44.960 --> 08:18:47.041
<v Neil Getnick (Getnick &amp; Getnick)>Mhmm. To basically eradicate

8611
08:18:47.041 --> 08:18:48.800
<v Neil Getnick (Getnick &amp; Getnick)>organized crime infiltration

8612
08:18:48.800 --> 08:18:50.240
<v Neil Getnick (Getnick &amp; Getnick)>of many industries,

8613
08:18:50.320 --> 08:18:52.080
<v Neil Getnick (Getnick &amp; Getnick)>including the trade waste industry.

8614
08:18:52.240 --> 08:18:52.880
<v Neil Getnick (Getnick &amp; Getnick)>It

8615
08:18:53.520 --> 08:18:56.080
<v Neil Getnick (Getnick &amp; Getnick)>has been recognized by
the National Law Journal

8616
08:18:56.240 --> 08:18:58.115
<v Neil Getnick (Getnick &amp; Getnick)>Mhmm. As the methodology

8617
08:18:58.115 --> 08:18:58.996
<v Neil Getnick (Getnick &amp; Getnick)>involving

8618
08:18:59.154 --> 08:19:00.596
<v Neil Getnick (Getnick &amp; Getnick)>multidisciplinary

8619
08:19:01.395 --> 08:19:02.355
<v Neil Getnick (Getnick &amp; Getnick)>independent

8620
08:19:02.596 --> 08:19:03.475
<v Neil Getnick (Getnick &amp; Getnick)>monitors

8621
08:19:03.635 --> 08:19:08.436
<v Neil Getnick (Getnick &amp; Getnick)>Okay. Which is immediately adaptable. I'd
be happy to confer confer with your office

8622
08:19:08.436 --> 08:19:11.790
<v Neil Getnick (Getnick &amp; Getnick)>Thank you. And provide that information.
Thank you very much. We greatly appreciate

8623
08:19:11.790 --> 08:19:15.148
<v Neil Getnick (Getnick &amp; Getnick)>appreciate this whole panel. Thank you.
Thank you. Councilmember Hanif, you had a

8624
08:19:15.148 --> 08:19:17.629
<v Neil Getnick (Getnick &amp; Getnick)>question? I had a question for Noel.

8625
08:19:17.950 --> 08:19:22.990
<v CM Shahana Hanif>Could you just speak a little bit more
about what a fully functioning COPIC

8626
08:19:23.709 --> 08:19:28.744
<v CM Shahana Hanif>could do today that other city agents,
no other city entity is currently doing?

8627
08:19:30.264 --> 08:19:33.225
<v Noel Hidalgo (BetaNYC)>Yeah. So thank you for
asking that question.

8628
08:19:33.545 --> 08:19:37.465
<v Noel Hidalgo (BetaNYC)>Since its inception, COPIC, as the
commission of public information and

8629
08:19:37.465 --> 08:19:41.762
<v Noel Hidalgo (BetaNYC)>communications, has the unique
power that brings council,

8630
08:19:41.840 --> 08:19:45.441
<v Noel Hidalgo (BetaNYC)>the executive, public advocate,
as well as public designee...

8631
08:19:45.441 --> 08:19:46.400
<v Noel Hidalgo (BetaNYC)>Designees

8632
08:19:46.400 --> 08:19:47.602
<v Noel Hidalgo (BetaNYC)>to oversee,

8633
08:19:48.160 --> 08:19:49.201
<v Noel Hidalgo (BetaNYC)>communications,

8634
08:19:49.281 --> 08:19:51.281
<v Noel Hidalgo (BetaNYC)>terrestrial TV, radio,

8635
08:19:51.441 --> 08:19:52.400
<v Noel Hidalgo (BetaNYC)>and the web.

8636
08:19:53.121 --> 08:19:57.496
<v Noel Hidalgo (BetaNYC)>KOPIC's first responsibility was to create
the very first data dictionary that the

8637
08:19:57.496 --> 08:20:00.695
<v Noel Hidalgo (BetaNYC)>city has ever seen. We now
have the city's open data law.

8638
08:20:00.936 --> 08:20:05.895
<v Noel Hidalgo (BetaNYC)>COPIK has the unique capability existing
in the city charter to do the oversight of

8639
08:20:05.895 --> 08:20:10.535
<v Noel Hidalgo (BetaNYC)>websites and to help define how
the public uses those websites,

8640
08:20:10.535 --> 08:20:13.550
<v Noel Hidalgo (BetaNYC)>and that provides a very unique capability

8641
08:20:13.550 --> 08:20:16.670
<v Noel Hidalgo (BetaNYC)>inside of the charter that
isn't being exercised.

8642
08:20:16.830 --> 08:20:21.630
<v Noel Hidalgo (BetaNYC)>And so what I would love to see is COPIK
have a series of hearings explicitly

8643
08:20:21.630 --> 08:20:23.390
<v Noel Hidalgo (BetaNYC)>focused on public information.

8644
08:20:23.790 --> 08:20:28.664
<v Noel Hidalgo (BetaNYC)>AI is public information. It is how
we... How these models have used public

8645
08:20:28.664 --> 08:20:32.025
<v Noel Hidalgo (BetaNYC)>information to essentially turn around
and distill public information.

8646
08:20:32.504 --> 08:20:33.465
<v Noel Hidalgo (BetaNYC)>So revamp

8647
08:20:33.465 --> 08:20:35.305
<v Noel Hidalgo (BetaNYC)>COPIC with resources

8648
08:20:35.305 --> 08:20:39.545
<v Noel Hidalgo (BetaNYC)>to actually host hearings and to
continue this existing conversation

8649
08:20:39.625 --> 08:20:43.390
<v Noel Hidalgo (BetaNYC)>can help bring that power. And COPIC's

8650
08:20:43.390 --> 08:20:44.190
<v Noel Hidalgo (BetaNYC)>advice,

8651
08:20:44.510 --> 08:20:48.430
<v Noel Hidalgo (BetaNYC)>when it writes a report, then comes to
counsel and essentially to introduce new

8652
08:20:48.430 --> 08:20:49.791
<v Noel Hidalgo (BetaNYC)>pieces of legislation,

8653
08:20:49.950 --> 08:20:53.230
<v Noel Hidalgo (BetaNYC)>which you then, as the speaker,
have other pieces of power

8654
08:20:53.470 --> 08:20:54.190
<v Noel Hidalgo (BetaNYC)>to execute.

8655
08:20:54.756 --> 08:20:56.275
<v Noel Hidalgo (BetaNYC)>I'll leave that to the lawyers.

8656
08:20:57.635 --> 08:21:01.156
<v CM Carmen De La Rosa>Thank you all so much, and thank you to
this panel. We really appreciate your

8657
08:21:01.156 --> 08:21:02.115
<v CM Carmen De La Rosa>testimony

8658
08:21:02.195 --> 08:21:04.916
<v CM Carmen De La Rosa>and look forward to
reading the full, extent

8659
08:21:04.916 --> 08:21:06.275
<v CM Carmen De La Rosa>of your testimony.

8660
08:21:06.516 --> 08:21:08.420
<v CM Carmen De La Rosa>The next panel

8661
08:21:08.818 --> 08:21:13.780
<v CM Carmen De La Rosa>is a hybrid panel. You had a question? Oh,
okay. The next panel is a hybrid panel.

8662
08:21:13.780 --> 08:21:15.780
<v CM Carmen De La Rosa>We have Scott Wizzer

8663
08:21:15.940 --> 08:21:22.635
<v CM Carmen De La Rosa>and Zach Praes in person. And then on
Zoom, we have Shay Brown, Gary Marcus, and

8664
08:21:22.635 --> 08:21:23.516
<v CM Carmen De La Rosa>David Krueger.

8665
08:21:42.690 --> 08:21:46.129
<v CM Carmen De La Rosa>We're gonna begin with the in person panel.
So if you could just identify yourself

8666
08:21:46.129 --> 08:21:47.330
<v CM Carmen De La Rosa>and begin whenever you're ready.

8667
08:21:56.465 --> 08:21:57.586
<v Zach Price (Reset Tech)>Madam speaker,

8668
08:21:57.986 --> 08:22:02.785
<v Zach Price (Reset Tech)>council members, my name is Zach Price.
I work with Reset Tech, a nonprofit

8669
08:22:02.785 --> 08:22:06.781
<v Zach Price (Reset Tech)>advocating for technology that protects
children, consumers, and public safety.

8670
08:22:07.020 --> 08:22:08.781
<v Zach Price (Reset Tech)>Thank you for convening this hearing.

8671
08:22:09.020 --> 08:22:12.780
<v Zach Price (Reset Tech)>While president Trump advocates for self
regulation and congress stalls, the New

8672
08:22:12.780 --> 08:22:17.100
<v Zach Price (Reset Tech)>York City Council is questioning AI companies
on the harms of their technologies and

8673
08:22:17.100 --> 08:22:19.596
<v Zach Price (Reset Tech)>leading on legislation
to hold them to account.

8674
08:22:20.396 --> 08:22:23.996
<v Zach Price (Reset Tech)>I'll focus on the harms New
Yorkers face today from AI.

8675
08:22:24.717 --> 08:22:30.076
<v Zach Price (Reset Tech)>AI powered scams are surging. Experts
estimate New Yorkers lost $8,700,000,000

8676
08:22:30.076 --> 08:22:32.156
<v Zach Price (Reset Tech)>to online scams in 2025.

8677
08:22:32.396 --> 08:22:34.640
<v Zach Price (Reset Tech)>That's about 1,000 per household.

8678
08:22:35.040 --> 08:22:38.160
<v Zach Price (Reset Tech)>Meta's internal documents
projected about 16,000,000,000

8679
08:22:38.160 --> 08:22:39.120
<v Zach Price (Reset Tech)>in revenue

8680
08:22:39.520 --> 08:22:43.280
<v Zach Price (Reset Tech)>from ads for scams and
banned goods in 2025...

8681
08:22:43.280 --> 08:22:44.400
<v Zach Price (Reset Tech)>2024,

8682
08:22:44.719 --> 08:22:47.200
<v Zach Price (Reset Tech)>that's 10% of the
company's global revenue.

8683
08:22:48.156 --> 08:22:51.115
<v Zach Price (Reset Tech)>72% of teens have used an AI companion.

8684
08:22:51.195 --> 08:22:55.195
<v Zach Price (Reset Tech)>Families alleged these products
have contributed to teen suicides.

8685
08:22:55.516 --> 08:22:59.836
<v Zach Price (Reset Tech)>Meta allowed romantic role play chatbots
for minors despite warnings from its own

8686
08:22:59.836 --> 08:23:00.635
<v Zach Price (Reset Tech)>internal staff.

8687
08:23:02.219 --> 08:23:04.860
<v Zach Price (Reset Tech)>AI is increasing cost for New Yorkers.

8688
08:23:04.940 --> 08:23:09.818
<v Zach Price (Reset Tech)>A recent analysis found that AI tools
increased health care cost for insurers by

8689
08:23:09.818 --> 08:23:10.780
<v Zach Price (Reset Tech)>$1,000,000,000

8690
08:23:10.780 --> 08:23:11.580
<v Zach Price (Reset Tech)>over two years.

8691
08:23:12.795 --> 08:23:15.996
<v Zach Price (Reset Tech)>AI is generating harmful
digital forgeries.

8692
08:23:16.234 --> 08:23:19.676
<v Zach Price (Reset Tech)>Researchers found that x AI's
Grock produced an estimated

8693
08:23:19.756 --> 08:23:21.756
<v Zach Price (Reset Tech)>3,000,000 sexualized images,

8694
08:23:21.994 --> 08:23:23.195
<v Zach Price (Reset Tech)>including 23,000

8695
08:23:23.195 --> 08:23:25.355
<v Zach Price (Reset Tech)>of children in an eleven day period.

8696
08:23:26.580 --> 08:23:29.300
<v Zach Price (Reset Tech)>These harms result from
business and design decisions.

8697
08:23:29.700 --> 08:23:33.140
<v Zach Price (Reset Tech)>When AI developers create harm, they
should bear the responsibility.

8698
08:23:33.459 --> 08:23:37.620
<v Zach Price (Reset Tech)>Holding AI companies accountable for these
harms will not only yield safer products

8699
08:23:37.620 --> 08:23:38.020
<v Zach Price (Reset Tech)>today,

8700
08:23:38.465 --> 08:23:42.625
<v Zach Price (Reset Tech)>but also create the internal incentive
structures at these companies to prevent harm

8701
08:23:42.625 --> 08:23:43.506
<v Zach Price (Reset Tech)>in the future.

8702
08:23:43.984 --> 08:23:49.105
<v Zach Price (Reset Tech)>The council can act to pass legislation
today to ensure and strengthen liability for

8703
08:23:49.105 --> 08:23:52.305
<v Zach Price (Reset Tech)>AI developers and privacy
protections for AI users,

8704
08:23:53.000 --> 08:23:56.520
<v Zach Price (Reset Tech)>prohibit AI from generating
harmful digital forgeries,

8705
08:23:56.840 --> 08:24:00.600
<v Zach Price (Reset Tech)>require third party validation and
incident reporting for AI companies,

8706
08:24:01.240 --> 08:24:03.880
<v Zach Price (Reset Tech)>support and expand
whistleblower protections.

8707
08:24:04.120 --> 08:24:07.080
<v Zach Price (Reset Tech)>Thank you for your leadership
on this matter. Thank you.

8708
08:24:10.195 --> 08:24:13.717
<v Scott Weiser (Secure AI Project)>Speaker chair, thank you for having us.
My name is Scott Weiser, policy director

8709
08:24:13.717 --> 08:24:16.996
<v Scott Weiser (Secure AI Project)>from the Secure AI Project.
We're nonprofit, nonpartisan.

8710
08:24:16.996 --> 08:24:20.115
<v Scott Weiser (Secure AI Project)>Don't take corporate or government
funding. And we supported,

8711
08:24:20.275 --> 08:24:23.990
<v Scott Weiser (Secure AI Project)>state AI safety legislation over the last
couple of years. We supported the RAISE

8712
08:24:23.990 --> 08:24:28.070
<v Scott Weiser (Secure AI Project)>Act in New York, s b 53 in California,
s b three fifteen in Illinois,

8713
08:24:28.310 --> 08:24:31.670
<v Scott Weiser (Secure AI Project)>and pending AI safety legislation that's
likely to pass conference committee in

8714
08:24:31.670 --> 08:24:33.111
<v Scott Weiser (Secure AI Project)>Massachusetts at the moment.

8715
08:24:33.670 --> 08:24:37.203
<v Scott Weiser (Secure AI Project)>I was really pleased to hear elected
officials earlier call out what the companies

8716
08:24:37.203 --> 08:24:40.484
<v Scott Weiser (Secure AI Project)>had said. I don't know if it's possible to
follow-up, but I think a very good thing

8717
08:24:40.484 --> 08:24:42.965
<v Scott Weiser (Secure AI Project)>for your committee to do would be to ask

8718
08:24:43.283 --> 08:24:45.125
<v Scott Weiser (Secure AI Project)>what specific legislation

8719
08:24:45.125 --> 08:24:46.805
<v Scott Weiser (Secure AI Project)>do you and did you support

8720
08:24:46.965 --> 08:24:50.725
<v Scott Weiser (Secure AI Project)>and at what time. It's really easy to
support the legislation after it's become

8721
08:24:50.725 --> 08:24:51.523
<v Scott Weiser (Secure AI Project)>legally binding.

8722
08:24:51.890 --> 08:24:55.570
<v Scott Weiser (Secure AI Project)>The other thing that I heard the company
say a lot is that they... Oh, yes. We very

8723
08:24:55.570 --> 08:24:59.810
<v Scott Weiser (Secure AI Project)>much like independent evaluation. We
sign up for that. There's a mechanism

8724
08:24:59.810 --> 08:25:03.012
<v Scott Weiser (Secure AI Project)>which they can use today to
make things legally binding.

8725
08:25:03.250 --> 08:25:07.695
<v Scott Weiser (Secure AI Project)>The compliance framework that they submit
that's required under s p 53 and the raise

8726
08:25:07.695 --> 08:25:09.055
<v Scott Weiser (Secure AI Project)>act on January 1,

8727
08:25:09.535 --> 08:25:12.416
<v Scott Weiser (Secure AI Project)>if they put their commitments
in that framework,

8728
08:25:12.496 --> 08:25:14.256
<v Scott Weiser (Secure AI Project)>then it's legally obligatory

8729
08:25:14.256 --> 08:25:17.695
<v Scott Weiser (Secure AI Project)>that they follow it. So I think that would
be a very good question to ask after this

8730
08:25:17.695 --> 08:25:18.016
<v Scott Weiser (Secure AI Project)>hearing.

8731
08:25:19.121 --> 08:25:24.320
<v Scott Weiser (Secure AI Project)>As far as what we think good public policy
means going forward to address the very

8732
08:25:24.320 --> 08:25:28.800
<v Scott Weiser (Secure AI Project)>severe risks from both advanced cyber
attacks, the possibility of bioweapons,

8733
08:25:28.880 --> 08:25:32.320
<v Scott Weiser (Secure AI Project)>we think we need much greater transparency
into what's happening at the companies.

8734
08:25:32.320 --> 08:25:35.236
<v Scott Weiser (Secure AI Project)>That means continuous
oversight by embedded

8735
08:25:35.316 --> 08:25:36.996
<v Scott Weiser (Secure AI Project)>third party evaluators,

8736
08:25:37.156 --> 08:25:38.916
<v Scott Weiser (Secure AI Project)>continued whistleblower protections,

8737
08:25:39.076 --> 08:25:42.916
<v Scott Weiser (Secure AI Project)>and we need the power for the government
upon getting information either through

8738
08:25:42.916 --> 08:25:46.570
<v Scott Weiser (Secure AI Project)>whistleblowers, independent evaluators,
the company itself to be able to halt

8739
08:25:46.570 --> 08:25:47.850
<v Scott Weiser (Secure AI Project)>dangerous activities.

8740
08:25:48.010 --> 08:25:51.850
<v Scott Weiser (Secure AI Project)>So we've been supporting, those steps in
other states and will continue to be a

8741
08:25:51.850 --> 08:25:54.410
<v Scott Weiser (Secure AI Project)>resource for, the council
as you're going forward.

8742
08:25:54.729 --> 08:25:59.129
<v Scott Weiser (Secure AI Project)>We think that it's an extraordinarily
important time for doing this, that the risks

8743
08:25:59.129 --> 08:26:02.266
<v Scott Weiser (Secure AI Project)>that we talked about six
months ago are now here,

8744
08:26:02.664 --> 08:26:06.504
<v Scott Weiser (Secure AI Project)>and we think that the risks six months from
now are much greater as the first panel

8745
08:26:06.504 --> 08:26:10.344
<v Scott Weiser (Secure AI Project)>today talked about recursive self improvement.
That's a very dangerous position to

8746
08:26:10.344 --> 08:26:12.424
<v Scott Weiser (Secure AI Project)>be in, and we're grateful
for your leadership on this.

8747
08:26:13.740 --> 08:26:18.541
<v CM Carmen De La Rosa>Thank you so much. We're on Zoom.
We're going to hear from Shay Brown.

8748
08:26:22.140 --> 08:26:25.020
<v CM Carmen De La Rosa>Shay, if you're on and ready,
please unmute yourself.

8749
08:26:29.395 --> 08:26:30.195
<v Unidentified [?]>Shay?

8750
08:26:32.914 --> 08:26:36.035
<v Shay Brown (Babble AI) [?]>Can you hear me now? Yes. We can hear you.

8751
08:26:36.914 --> 08:26:41.154
<v Shay Brown (Babble AI)>Thank you, speaker Menon and members of the
committee for this opportunity to speak.

8752
08:26:41.234 --> 08:26:45.932
<v Shay Brown (Babble AI)>My name is Shay Brown. I'm the CEO of Babble
AI, a company that audits high risk AI

8753
08:26:45.932 --> 08:26:48.332
<v Shay Brown (Babble AI)>systems for safety and
effective governance.

8754
08:26:48.730 --> 08:26:53.050
<v Shay Brown (Babble AI)>And we've been independent bias auditors
under local law one four four since that

8755
08:26:53.050 --> 08:26:54.650
<v Shay Brown (Babble AI)>law first took effect.

8756
08:26:54.970 --> 08:26:58.810
<v Shay Brown (Babble AI)>I'd like to speak today in support
of proposal two six zero two,

8757
08:26:59.414 --> 08:27:03.574
<v Shay Brown (Babble AI)>an AI system that affects the public
should be examined by a qualified party,

8758
08:27:04.055 --> 08:27:08.215
<v Shay Brown (Babble AI)>outside the company that built it before it
reaches the public. I further think that

8759
08:27:08.215 --> 08:27:11.895
<v Shay Brown (Babble AI)>New York City is the right place for this
to start, and we'd like to see it done in

8760
08:27:11.895 --> 08:27:16.129
<v Shay Brown (Babble AI)>a way that is effective and promotes the
public's well-being. I'd like to raise two

8761
08:27:16.129 --> 08:27:19.010
<v Shay Brown (Babble AI)>points, and I offer more
in my written testimony.

8762
08:27:19.330 --> 08:27:23.650
<v Shay Brown (Babble AI)>First, the bill asked the outside validator
to certify that a model is appropriately

8763
08:27:23.650 --> 08:27:25.250
<v Shay Brown (Babble AI)>positioned for deployment.

8764
08:27:25.570 --> 08:27:29.570
<v Shay Brown (Babble AI)>From our experience, we believe this is
more than an outside party can reasonably

8765
08:27:29.570 --> 08:27:33.855
<v Shay Brown (Babble AI)>say given the third party dynamics
and access to the system in question.

8766
08:27:34.336 --> 08:27:38.816
<v Shay Brown (Babble AI)>Instead, we suggest potentially requiring
the developer to make written assertions

8767
08:27:38.816 --> 08:27:40.016
<v Shay Brown (Babble AI)>about its compliance

8768
08:27:40.256 --> 08:27:44.336
<v Shay Brown (Babble AI)>over specific topics and to a minimum
standard that is set by the city,

8769
08:27:44.960 --> 08:27:49.201
<v Shay Brown (Babble AI)>and an independent practitioner validates
those assertions through careful testing,

8770
08:27:49.281 --> 08:27:51.041
<v Shay Brown (Babble AI)>verification, and inspection.

8771
08:27:51.361 --> 08:27:55.281
<v Shay Brown (Babble AI)>This puts the burden of proof back on the
developer and does not force the auditor

8772
08:27:55.281 --> 08:27:58.000
<v Shay Brown (Babble AI)>to make a judgment call that it's
not in the best position to make.

8773
08:27:58.734 --> 08:28:02.574
<v Shay Brown (Babble AI)>My second point involves the
validator's independence and competence.

8774
08:28:02.895 --> 08:28:06.654
<v Shay Brown (Babble AI)>The only requirement for a validator in
the current draft is that it not be an

8775
08:28:06.654 --> 08:28:08.174
<v Shay Brown (Babble AI)>affiliate of the developer.

8776
08:28:08.654 --> 08:28:13.310
<v Shay Brown (Babble AI)>We ask that qualification be more
closely tied to existing standards and or

8777
08:28:13.310 --> 08:28:17.630
<v Shay Brown (Babble AI)>accreditation for independent assurance
and certification bodies that exist today.

8778
08:28:18.110 --> 08:28:22.110
<v Shay Brown (Babble AI)>While the subject matter is unique and
requires careful thought, the frameworks for

8779
08:28:22.110 --> 08:28:26.350
<v Shay Brown (Babble AI)>professional conduct of third party audit
assurance and validation professionals

8780
08:28:26.506 --> 08:28:29.227
<v Shay Brown (Babble AI)>already exists, and we
should leverage that.

8781
08:28:29.467 --> 08:28:32.105
<v Shay Brown (Babble AI)>Thank you for your time, and I'm
happy to answer any questions.

8782
08:28:32.586 --> 08:28:37.787
<v CM Carmen De La Rosa>Thank you so much. Up next, we'll
hear from Gary Marcus on Zoom. Gary,

8783
08:28:38.506 --> 08:28:39.865
<v CM Carmen De La Rosa>when you're ready, please unmute.

8784
08:28:56.145 --> 08:28:57.506
<v Committee Counsel>Gary, you may begin.

8785
08:29:02.785 --> 08:29:05.346
<v CM Carmen De La Rosa>Alright. We're gonna move
on to David Krueger.

8786
08:29:07.550 --> 08:29:10.512
<v CM Carmen De La Rosa>David, if you're on the Zoom,
please unmute and begin.

8787
08:29:15.070 --> 08:29:15.710
<v Unidentified (public) [?]>Hi.

8788
08:29:15.951 --> 08:29:16.672
<v David Krueger (University of Montreal / Mila)>Yep.

8789
08:29:18.031 --> 08:29:22.035
<v David Krueger (University of Montreal / Mila)>Thanks so much for having this hearing,
and thanks for inviting me. So my name is

8790
08:29:22.035 --> 08:29:24.834
<v David Krueger (University of Montreal / Mila)>David Krueger. I'm a machine learning
professor at the University of Montreal in

8791
08:29:24.834 --> 08:29:28.754
<v David Krueger (University of Montreal / Mila)>Mila, and I was previously an intern on
the DeepMind AI safety team and a founding

8792
08:29:28.754 --> 08:29:31.635
<v David Krueger (University of Montreal / Mila)>research director at the
UK AI Security Institute.

8793
08:29:32.115 --> 08:29:35.555
<v David Krueger (University of Montreal / Mila)>It was great hearing all of your
questioning of the representatives of the AI

8794
08:29:35.555 --> 08:29:38.490
<v David Krueger (University of Montreal / Mila)>companies. I think there are a lot of
really important questions that are being

8795
08:29:38.490 --> 08:29:41.530
<v David Krueger (University of Montreal / Mila)>asked here about the ability to,

8796
08:29:42.010 --> 08:29:44.570
<v David Krueger (University of Montreal / Mila)>actually provide meaningful
assurances for these systems.

8797
08:29:44.809 --> 08:29:48.648
<v David Krueger (University of Montreal / Mila)>So I think we're quite probably on the verge
of a catastrophe at this moment, and we

8798
08:29:48.648 --> 08:29:53.176
<v David Krueger (University of Montreal / Mila)>should not proceed building more powerful
AI systems Even if there was only a 10%

8799
08:29:53.176 --> 08:29:57.096
<v David Krueger (University of Montreal / Mila)>risk within the next ten years of losing
control, there'd be more than enough reason

8800
08:29:57.096 --> 08:30:01.256
<v David Krueger (University of Montreal / Mila)>to stop building these systems, but I
believe the risk is significantly higher and

8801
08:30:01.256 --> 08:30:01.656
<v David Krueger (University of Montreal / Mila)>sooner.

8802
08:30:04.420 --> 08:30:07.940
<v David Krueger (University of Montreal / Mila)>I've been doing deep learning and AI safety
for over a dozen years, and I've seen

8803
08:30:07.940 --> 08:30:11.219
<v David Krueger (University of Montreal / Mila)>these concerns go from fringe to mainstream.
So we're now at a watershed moment

8804
08:30:11.219 --> 08:30:14.898
<v David Krueger (University of Montreal / Mila)>where society is finally facing up to this
problem. As a result, I'm more optimistic

8805
08:30:14.898 --> 08:30:18.340
<v David Krueger (University of Montreal / Mila)>than I've been since I started worrying
about AI risk over fifteen years ago,

8806
08:30:18.996 --> 08:30:22.676
<v David Krueger (University of Montreal / Mila)>but we still need more awareness and
understanding of the situation we're in, how

8807
08:30:22.676 --> 08:30:24.115
<v David Krueger (University of Montreal / Mila)>bad the situation is.

8808
08:30:24.355 --> 08:30:28.195
<v David Krueger (University of Montreal / Mila)>And the discussion is currently not where
it needs to be around this. So people

8809
08:30:28.195 --> 08:30:31.076
<v David Krueger (University of Montreal / Mila)>really need to have an understanding of the
scope, the depth, and the urgency of the

8810
08:30:31.076 --> 08:30:31.236
<v David Krueger (University of Montreal / Mila)>problem.

8811
08:30:32.041 --> 08:30:36.201
<v David Krueger (University of Montreal / Mila)>We are talking about things up to and
including human extinction or unseen levels of

8812
08:30:36.201 --> 08:30:37.880
<v David Krueger (University of Montreal / Mila)>unemployment concentration of power.

8813
08:30:38.281 --> 08:30:41.480
<v David Krueger (University of Montreal / Mila)>We are talking about recursive self improvement
that companies say they're beginning

8814
08:30:41.480 --> 08:30:46.281
<v David Krueger (University of Montreal / Mila)>to do right now, which could, at virtually
any moment, lead to the creation of AI

8815
08:30:46.281 --> 08:30:50.656
<v David Krueger (University of Montreal / Mila)>systems much smarter than people, which
we have no, plan for how to control,

8816
08:30:50.975 --> 08:30:52.256
<v David Krueger (University of Montreal / Mila)>or knowledge of how to.

8817
08:30:53.535 --> 08:30:56.016
<v David Krueger (University of Montreal / Mila)>And most importantly,
these problems go deep.

8818
08:30:56.416 --> 08:30:59.135
<v David Krueger (University of Montreal / Mila)>The science of AI is extremely immature.

8819
08:30:59.791 --> 08:31:02.990
<v David Krueger (University of Montreal / Mila)>All of the technical approaches
to understanding AI systems,

8820
08:31:03.310 --> 08:31:07.390
<v David Krueger (University of Montreal / Mila)>controlling them, and telling if they're
safe are are not where they need to be.

8821
08:31:07.390 --> 08:31:10.432
<v David Krueger (University of Montreal / Mila)>They're far from where they need to be
despite over a decade of research that I and

8822
08:31:10.432 --> 08:31:13.230
<v David Krueger (University of Montreal / Mila)>other, like many other researchers,
have spent a long time on this.

8823
08:31:13.945 --> 08:31:18.025
<v David Krueger (University of Montreal / Mila)>So, you know, we are having discussions
about policy and including some more

8824
08:31:18.025 --> 08:31:19.145
<v David Krueger (University of Montreal / Mila)>substantive policy.

8825
08:31:19.465 --> 08:31:20.105
<v David Krueger (University of Montreal / Mila)>Sorry?

8826
08:31:20.904 --> 08:31:23.545
<v CM Carmen De La Rosa>Your time is up. Thank you so much for

8827
08:31:24.266 --> 08:31:27.545
<v CM Carmen De La Rosa>coming in virtually. And if you could
just submit the rest of your testimony,

8828
08:31:27.980 --> 08:31:32.701
<v CM Carmen De La Rosa>we'd really appreciate it. And and thank
you to this panel for being on. We're gonna

8829
08:31:32.701 --> 08:31:35.900
<v CM Carmen De La Rosa>call one more time Gary Marcus
if Gary's on Zoom... On mute.

8830
08:31:39.945 --> 08:31:43.305
<v CM Carmen De La Rosa>Alright. We're gonna move on to the next
panel. Thank you all for being here.

8831
08:31:44.186 --> 08:31:45.545
<v CM Carmen De La Rosa>The next panel

8832
08:31:45.705 --> 08:31:47.385
<v CM Carmen De La Rosa>is Nicholas Arture,

8833
08:31:47.865 --> 08:31:49.305
<v CM Carmen De La Rosa>Mary Vaccaro,

8834
08:31:49.545 --> 08:31:50.986
<v CM Carmen De La Rosa>Dawn DeCosta,

8835
08:31:51.064 --> 08:31:52.346
<v CM Carmen De La Rosa>Manny Pastrich,

8836
08:31:52.666 --> 08:31:54.105
<v CM Carmen De La Rosa>Carrie Faye Ibar,

8837
08:31:54.941 --> 08:31:55.980
<v CM Carmen De La Rosa>Jimmy Pan,

8838
08:31:56.140 --> 08:31:58.060
<v CM Carmen De La Rosa>and Jessica Timo on Zoom.

8839
08:32:21.330 --> 08:32:24.050
<v CM Carmen De La Rosa>Okay. Let's start with the in person panel

8840
08:32:24.050 --> 08:32:26.852
<v CM Carmen De La Rosa>now. If you could just
identify yourself and begin.

8841
08:32:28.291 --> 08:32:30.930
<v Carrie Fallhaber (Jobs First NYC)>Good evening, speaker Menon, chairperson

8842
08:32:30.930 --> 08:32:34.531
<v Carrie Fallhaber (Jobs First NYC)>de la Rosa, and members of the
council. My name is Carrie Fallhaber.

8843
08:32:34.531 --> 08:32:37.535
<v Carrie Fallhaber (Jobs First NYC)>I'm the senior vice
president at Jobs First NYC.

8844
08:32:37.936 --> 08:32:40.574
<v Carrie Fallhaber (Jobs First NYC)>Thank you for the invitation
to testify today.

8845
08:32:40.895 --> 08:32:46.094
<v Carrie Fallhaber (Jobs First NYC)>Jobs First NYC works with more than 200
partners to strengthen workforce systems and

8846
08:32:46.094 --> 08:32:47.775
<v Carrie Fallhaber (Jobs First NYC)>expand economic opportunity.

8847
08:32:48.670 --> 08:32:50.030
<v Carrie Fallhaber (Jobs First NYC)>AI is reshaping

8848
08:32:50.430 --> 08:32:51.230
<v Carrie Fallhaber (Jobs First NYC)>jobs,

8849
08:32:51.389 --> 08:32:52.190
<v Carrie Fallhaber (Jobs First NYC)>skills,

8850
08:32:52.510 --> 08:32:55.390
<v Carrie Fallhaber (Jobs First NYC)>and how employers find and deploy talent

8851
08:32:55.549 --> 08:32:58.030
<v Carrie Fallhaber (Jobs First NYC)>in ways that are difficult to quantify.

8852
08:32:58.670 --> 08:33:01.870
<v Carrie Fallhaber (Jobs First NYC)>The impact of a trial and
error approach to safety

8853
08:33:02.404 --> 08:33:07.445
<v Carrie Fallhaber (Jobs First NYC)>is a risk that AI companies should
bear, not New York City's young adults.

8854
08:33:08.324 --> 08:33:12.244
<v Carrie Fallhaber (Jobs First NYC)>Our immediate concern is what happens
when the first rung of the career ladder

8855
08:33:12.244 --> 08:33:15.764
<v Carrie Fallhaber (Jobs First NYC)>changes faster than our workforce
system is unable to respond.

8856
08:33:16.621 --> 08:33:19.980
<v Carrie Fallhaber (Jobs First NYC)>That matters especially
for the more than 110,000

8857
08:33:19.980 --> 08:33:25.500
<v Carrie Fallhaber (Jobs First NYC)>young adults who are out of school
and out of work and more than 177,000

8858
08:33:25.580 --> 08:33:27.182
<v Carrie Fallhaber (Jobs First NYC)>who are underemployed,

8859
08:33:27.420 --> 08:33:30.861
<v Carrie Fallhaber (Jobs First NYC)>more than thirty five percent of all
sixteen to twenty four year olds.

8860
08:33:31.525 --> 08:33:35.445
<v Carrie Fallhaber (Jobs First NYC)>As entry level work changes, they could
face even greater barriers to getting their

8861
08:33:35.445 --> 08:33:36.885
<v Carrie Fallhaber (Jobs First NYC)>foothold in the workforce.

8862
08:33:37.445 --> 08:33:39.045
<v Carrie Fallhaber (Jobs First NYC)>Workforce preparedness

8863
08:33:39.045 --> 08:33:40.805
<v Carrie Fallhaber (Jobs First NYC)>is an AI safeguard.

8864
08:33:41.205 --> 08:33:44.885
<v Carrie Fallhaber (Jobs First NYC)>AI readiness should be part of New York
City's workforce infrastructure from

8865
08:33:44.885 --> 08:33:47.205
<v Carrie Fallhaber (Jobs First NYC)>education and training through
hiring and advancement.

8866
08:33:47.980 --> 08:33:49.980
<v Carrie Fallhaber (Jobs First NYC)>For more than a decade, JobSourceMIC

8867
08:33:49.980 --> 08:33:53.740
<v Carrie Fallhaber (Jobs First NYC)>has worked to understand how
technology is changing access to work.

8868
08:33:53.980 --> 08:33:55.260
<v Carrie Fallhaber (Jobs First NYC)>Today, we are developing

8869
08:33:55.500 --> 08:33:58.460
<v Carrie Fallhaber (Jobs First NYC)>decision making tools to reduce AI risk,

8870
08:33:58.700 --> 08:34:01.740
<v Carrie Fallhaber (Jobs First NYC)>scale what works, and shrink
the distance between education,

8871
08:34:01.740 --> 08:34:06.494
<v Carrie Fallhaber (Jobs First NYC)>training, and getting a job. We recommend
that the city invest in sector based AI

8872
08:34:06.494 --> 08:34:08.895
<v Carrie Fallhaber (Jobs First NYC)>fluency paid pathways into work.

8873
08:34:09.135 --> 08:34:11.535
<v Carrie Fallhaber (Jobs First NYC)>AI is changing each sector differently.

8874
08:34:11.855 --> 08:34:16.014
<v Carrie Fallhaber (Jobs First NYC)>Employers, education, and workforce partners
should define sector skills with city

8875
08:34:16.014 --> 08:34:17.535
<v Carrie Fallhaber (Jobs First NYC)>investment in internships,

8876
08:34:17.535 --> 08:34:18.174
<v Carrie Fallhaber (Jobs First NYC)>apprenticeships,

8877
08:34:18.360 --> 08:34:21.799
<v Carrie Fallhaber (Jobs First NYC)>and a civic core where young people
build those skills through work.

8878
08:34:22.120 --> 08:34:26.840
<v Carrie Fallhaber (Jobs First NYC)>Every young person divert... Deserves a
first chance at a good job. As AI changes

8879
08:34:26.840 --> 08:34:31.320
<v Carrie Fallhaber (Jobs First NYC)>work, New York City has to make sure that
first chance is still there. Thank you.

8880
08:34:32.040 --> 08:34:32.920
<v CM Carmen De La Rosa>Thank you so much.

8881
08:34:37.865 --> 08:34:42.426
<v Nicholas Artur (Consortium for Worker Education)>Good evening, speaker Menon, technology
chair De La Rosa, and members of the

8882
08:34:42.426 --> 08:34:46.426
<v Nicholas Artur (Consortium for Worker Education)>committee of the whole. My name is
Nicholas Artur, and I serve as director of

8883
08:34:46.426 --> 08:34:50.870
<v Nicholas Artur (Consortium for Worker Education)>external affairs and strategic partnerships
at consortium the for worker education,

8884
08:34:51.030 --> 08:34:51.990
<v Nicholas Artur (Consortium for Worker Education)>CWE.

8885
08:34:52.309 --> 08:34:56.389
<v Nicholas Artur (Consortium for Worker Education)>Rooted in the values of the labor movement,
CWE empowers workers with the skills to

8886
08:34:56.389 --> 08:35:00.389
<v Nicholas Artur (Consortium for Worker Education)>thrive, support their families, and strengthen
their communities through education,

8887
08:35:00.389 --> 08:35:01.030
<v Nicholas Artur (Consortium for Worker Education)>training,

8888
08:35:01.270 --> 08:35:03.270
<v Nicholas Artur (Consortium for Worker Education)>childcare, and job placement services.

8889
08:35:03.756 --> 08:35:04.955
<v Nicholas Artur (Consortium for Worker Education)>For CWE,

8890
08:35:04.955 --> 08:35:07.996
<v Nicholas Artur (Consortium for Worker Education)>worker protections must
be a part of AI safety.

8891
08:35:08.715 --> 08:35:12.557
<v Nicholas Artur (Consortium for Worker Education)>This is why we strongly support the council's
efforts through intro number one sixty

8892
08:35:12.557 --> 08:35:16.236
<v Nicholas Artur (Consortium for Worker Education)>one to understand how algorithmic
tools are changing city employment.

8893
08:35:16.940 --> 08:35:21.660
<v Nicholas Artur (Consortium for Worker Education)>Measuring what happened to city workers
jobs because of AI is just the first step.

8894
08:35:21.898 --> 08:35:25.180
<v Nicholas Artur (Consortium for Worker Education)>We also need to be able to prepare
workers for what comes next.

8895
08:35:25.420 --> 08:35:29.660
<v Nicholas Artur (Consortium for Worker Education)>CWE is developing a framework to inform how
we can support workers through education

8896
08:35:29.660 --> 08:35:32.620
<v Nicholas Artur (Consortium for Worker Education)>and policy to assist them during
this point of transition.

8897
08:35:33.156 --> 08:35:37.637
<v Nicholas Artur (Consortium for Worker Education)>When AI technology risks changing the nature
of someone's jobs, a few things should

8898
08:35:37.637 --> 08:35:41.877
<v Nicholas Artur (Consortium for Worker Education)>happen before that technology gets rolled
out. Employers should assess the impact,

8899
08:35:42.115 --> 08:35:46.020
<v Nicholas Artur (Consortium for Worker Education)>tell workers what's coming, sit down
with those workers and their unions, and

8900
08:35:46.020 --> 08:35:50.020
<v Nicholas Artur (Consortium for Worker Education)>provide training that actually... That's
actually relevant to the work people do.

8901
08:35:50.100 --> 08:35:53.219
<v Nicholas Artur (Consortium for Worker Education)>That means asking more than
how many jobs did AI eliminate.

8902
08:35:53.379 --> 08:35:57.940
<v Nicholas Artur (Consortium for Worker Education)>It also means asking which jobs are changing
and how. What skills will workers need?

8903
08:35:57.940 --> 08:35:59.860
<v Nicholas Artur (Consortium for Worker Education)>Can they be retrained, redeployed?

8904
08:36:00.020 --> 08:36:03.334
<v Nicholas Artur (Consortium for Worker Education)>And do workers have any say in
this? Their voices are paramount.

8905
08:36:03.975 --> 08:36:08.295
<v Nicholas Artur (Consortium for Worker Education)>This matters more now because AI isn't just
a tool sitting on our desktops anymore.

8906
08:36:08.295 --> 08:36:11.574
<v Nicholas Artur (Consortium for Worker Education)>It's changing how work gets assigned, how
people are evaluated, and how schedules

8907
08:36:11.574 --> 08:36:16.455
<v Nicholas Artur (Consortium for Worker Education)>are built, and how work gets done. That's
why we see intro one sixty one and CWE's

8908
08:36:16.455 --> 08:36:18.890
<v Nicholas Artur (Consortium for Worker Education)>framework as two sides of the same effort.

8909
08:36:19.210 --> 08:36:23.451
<v Nicholas Artur (Consortium for Worker Education)>One sixty one tells us what's changed. CWE
wants to help the city get ahead of that

8910
08:36:23.451 --> 08:36:27.371
<v Nicholas Artur (Consortium for Worker Education)>change and make sure workers are ready
for it. We're not asking the city to pick

8911
08:36:27.371 --> 08:36:31.064
<v Nicholas Artur (Consortium for Worker Education)>between innovation and workers. We're
asking the city to make sure workers don't

8912
08:36:31.064 --> 08:36:33.465
<v Nicholas Artur (Consortium for Worker Education)>carry the cost of
innovation by themselves.

8913
08:36:33.865 --> 08:36:37.545
<v Nicholas Artur (Consortium for Worker Education)>CWE is ready to work with the
council, organized labor, employers,

8914
08:36:37.625 --> 08:36:40.984
<v Nicholas Artur (Consortium for Worker Education)>and educational institutions to
build a workforce transition system

8915
08:36:41.240 --> 08:36:45.240
<v Nicholas Artur (Consortium for Worker Education)>that makes technological changes safer,
fairer, and more beneficial for working New

8916
08:36:45.240 --> 08:36:48.840
<v Nicholas Artur (Consortium for Worker Education)>Yorkers, and above all, build
worker power. Thank you.

8917
08:36:49.400 --> 08:36:50.201
<v CM Carmen De La Rosa>Thank you.

8918
08:36:52.600 --> 08:36:53.000
<v Jimmy Pan (NYC Equal Employment Practices Commission)>Council,

8919
08:36:53.984 --> 08:36:57.824
<v Jimmy Pan (NYC Equal Employment Practices Commission)>speaker, chair, thank you very much for
this hearing today. My name is Jimmy Pan.

8920
08:36:57.824 --> 08:37:01.824
<v Jimmy Pan (NYC Equal Employment Practices Commission)>I'm the executive director of a small city
agency called the New York City Equal

8921
08:37:01.824 --> 08:37:03.424
<v Jimmy Pan (NYC Equal Employment Practices Commission)>Employment Practices Commission.

8922
08:37:03.984 --> 08:37:07.184
<v Jimmy Pan (NYC Equal Employment Practices Commission)>I wasn't planning on testifying today, to
be honest, even though I think our work

8923
08:37:07.184 --> 08:37:10.510
<v Jimmy Pan (NYC Equal Employment Practices Commission)>touches on a lot of the issues that
came up today and on the panel.

8924
08:37:10.990 --> 08:37:14.990
<v Jimmy Pan (NYC Equal Employment Practices Commission)>I decided to testify because what a lot of
what I heard today actually reminds me of

8925
08:37:14.990 --> 08:37:19.470
<v Jimmy Pan (NYC Equal Employment Practices Commission)>the types of issues I was hearing two
decades ago. So before I became a lawyer, I

8926
08:37:19.470 --> 08:37:21.310
<v Jimmy Pan (NYC Equal Employment Practices Commission)>was in another frontier technology.

8927
08:37:21.470 --> 08:37:23.070
<v Jimmy Pan (NYC Equal Employment Practices Commission)>I was a biotechnology researcher.

8928
08:37:23.705 --> 08:37:27.064
<v Jimmy Pan (NYC Equal Employment Practices Commission)>And then the claims that companies were
putting forward as to why we need the

8929
08:37:27.064 --> 08:37:28.984
<v Jimmy Pan (NYC Equal Employment Practices Commission)>biotechnology despite the risks

8930
08:37:29.305 --> 08:37:33.945
<v Jimmy Pan (NYC Equal Employment Practices Commission)>where that could solve issues like world
hunger. The implication being if you're

8931
08:37:33.945 --> 08:37:35.465
<v Jimmy Pan (NYC Equal Employment Practices Commission)>opposed to biotechnology,

8932
08:37:35.545 --> 08:37:38.025
<v Jimmy Pan (NYC Equal Employment Practices Commission)>then you're dooming kids to
hunger around the world.

8933
08:37:38.600 --> 08:37:43.238
<v Jimmy Pan (NYC Equal Employment Practices Commission)>But, of course, the part that goes unsaid
is that hunger, for example, is not a

8934
08:37:43.238 --> 08:37:45.639
<v Jimmy Pan (NYC Equal Employment Practices Commission)>technology issue. It's
a distribution issue.

8935
08:37:46.680 --> 08:37:48.120
<v Jimmy Pan (NYC Equal Employment Practices Commission)>To to analogize

8936
08:37:48.280 --> 08:37:49.080
<v Jimmy Pan (NYC Equal Employment Practices Commission)>it today,

8937
08:37:49.398 --> 08:37:54.625
<v Jimmy Pan (NYC Equal Employment Practices Commission)>we don't have a hunger issue in The US. We
choose not to, for example, give school

8938
08:37:54.625 --> 08:37:55.504
<v Jimmy Pan (NYC Equal Employment Practices Commission)>lunches

8939
08:37:55.504 --> 08:37:56.385
<v Jimmy Pan (NYC Equal Employment Practices Commission)>for free.

8940
08:37:57.584 --> 08:38:01.424
<v Jimmy Pan (NYC Equal Employment Practices Commission)>Let's fast forward a little bit. I became
a lawyer, took on the pharmaceutical

8941
08:38:01.424 --> 08:38:02.225
<v Jimmy Pan (NYC Equal Employment Practices Commission)>industry.

8942
08:38:02.625 --> 08:38:06.064
<v Jimmy Pan (NYC Equal Employment Practices Commission)>The pharmaceutical industry is another one
of these frontier industries that says,

8943
08:38:06.541 --> 08:38:10.701
<v Jimmy Pan (NYC Equal Employment Practices Commission)>you have to allow us to innovate. You
can't regulate us. Otherwise, people won't

8944
08:38:10.701 --> 08:38:12.220
<v Jimmy Pan (NYC Equal Employment Practices Commission)>have these lifesaving drugs.

8945
08:38:12.460 --> 08:38:17.420
<v Jimmy Pan (NYC Equal Employment Practices Commission)>Again, presenting this narrative that if
you regulate this industry, if you cap

8946
08:38:17.420 --> 08:38:18.541
<v Jimmy Pan (NYC Equal Employment Practices Commission)>prices, you're,

8947
08:38:19.166 --> 08:38:20.926
<v Jimmy Pan (NYC Equal Employment Practices Commission)>as a regulator, responsible

8948
08:38:21.246 --> 08:38:23.967
<v Jimmy Pan (NYC Equal Employment Practices Commission)>for people who are gonna
suffer from these diseases.

8949
08:38:23.967 --> 08:38:26.207
<v Jimmy Pan (NYC Equal Employment Practices Commission)>I think we heard some
remnants of that today.

8950
08:38:26.686 --> 08:38:29.566
<v Jimmy Pan (NYC Equal Employment Practices Commission)>But, again, I wanna emphasize
to this regulatory body

8951
08:38:29.887 --> 08:38:34.205
<v Jimmy Pan (NYC Equal Employment Practices Commission)>that if we care about innovation and if
we care about improving lives for people,

8952
08:38:34.621 --> 08:38:36.781
<v Jimmy Pan (NYC Equal Employment Practices Commission)>there are many, many paths to innovation.

8953
08:38:37.021 --> 08:38:41.422
<v Jimmy Pan (NYC Equal Employment Practices Commission)>There are policy choices that we make as
a nation that put people's lives at risk.

8954
08:38:41.660 --> 08:38:45.900
<v Jimmy Pan (NYC Equal Employment Practices Commission)>We are currently defunding health systems.
We are currently not providing good

8955
08:38:45.900 --> 08:38:50.545
<v Jimmy Pan (NYC Equal Employment Practices Commission)>education. We're making choices on a policy
framework. And I think a lot of these

8956
08:38:50.545 --> 08:38:53.584
<v Jimmy Pan (NYC Equal Employment Practices Commission)>companies are saying, well, look,
all these issues you have today,

8957
08:38:53.824 --> 08:38:57.184
<v Jimmy Pan (NYC Equal Employment Practices Commission)>AI can solve that, and you need us there
at the table. I think that's a false

8958
08:38:57.184 --> 08:38:57.824
<v Jimmy Pan (NYC Equal Employment Practices Commission)>narrative.

8959
08:39:00.064 --> 08:39:02.400
<v CM Carmen De La Rosa>Thank you so much for your testimony.

8960
08:39:03.120 --> 08:39:07.440
<v CM Carmen De La Rosa>We're gonna call on Zoom
Jessica Timu and Sam Wheeler.

8961
08:39:07.520 --> 08:39:11.440
<v CM Carmen De La Rosa>Jessica, if you're on, please
unmute yourself and begin.

8962
08:39:12.640 --> 08:39:16.160
<v Jessica Timo (OPEIU Local 153)>Good evening, speaker Menon and members
of the city council. Thank you for the

8963
08:39:16.160 --> 08:39:19.975
<v Jessica Timo (OPEIU Local 153)>opportunity to testify on the state of
artificial intelligence and its sweeping

8964
08:39:19.975 --> 08:39:21.014
<v Jessica Timo (OPEIU Local 153)>consequences

8965
08:39:21.254 --> 08:39:24.295
<v Jessica Timo (OPEIU Local 153)>on our community, workplace,
and the labor movement.

8966
08:39:24.535 --> 08:39:26.295
<v Jessica Timo (OPEIU Local 153)>My name is Jessica Timo,

8967
08:39:26.295 --> 08:39:30.215
<v Jessica Timo (OPEIU Local 153)>and I am the president of OPIU
Local one five three AFL CIO

8968
08:39:30.530 --> 08:39:32.930
<v Jessica Timo (OPEIU Local 153)>based here in New York City representing

8969
08:39:32.930 --> 08:39:34.690
<v Jessica Timo (OPEIU Local 153)>over 11,000 workers.

8970
08:39:35.568 --> 08:39:39.648
<v Jessica Timo (OPEIU Local 153)>We represent a broad spectrum of workers
and roles across industries including

8971
08:39:39.648 --> 08:39:40.450
<v Jessica Timo (OPEIU Local 153)>clerical,

8972
08:39:40.450 --> 08:39:41.330
<v Jessica Timo (OPEIU Local 153)>administrative,

8973
08:39:41.330 --> 08:39:42.209
<v Jessica Timo (OPEIU Local 153)>nonprofit,

8974
08:39:42.209 --> 08:39:45.836
<v Jessica Timo (OPEIU Local 153)>tech, health care, credit
unions, and higher education.

8975
08:39:46.234 --> 08:39:50.475
<v Jessica Timo (OPEIU Local 153)>The future of artificial intelligence
regulation is a core question for the labor

8976
08:39:50.475 --> 08:39:52.314
<v Jessica Timo (OPEIU Local 153)>movement here in New York City.

8977
08:39:52.795 --> 08:39:56.795
<v Jessica Timo (OPEIU Local 153)>Drawing on the council's attention towards
the issues of artificial intelligence in

8978
08:39:56.795 --> 08:39:57.355
<v Jessica Timo (OPEIU Local 153)>the workplace,

8979
08:39:57.990 --> 08:40:02.950
<v Jessica Timo (OPEIU Local 153)>Local one five three has heard concerns from
workers across the city and encountered

8980
08:40:02.950 --> 08:40:04.230
<v Jessica Timo (OPEIU Local 153)>workplace navigating

8981
08:40:04.230 --> 08:40:05.750
<v Jessica Timo (OPEIU Local 153)>AI implementation

8982
08:40:05.830 --> 08:40:08.951
<v Jessica Timo (OPEIU Local 153)>during our bargaining of
hundreds of workers' contract.

8983
08:40:09.430 --> 08:40:12.470
<v Jessica Timo (OPEIU Local 153)>As a labor union, we're extremely alarmed

8984
08:40:12.805 --> 08:40:18.244
<v Jessica Timo (OPEIU Local 153)>the way the integration of AI tools and
technology have negatively impacted workers

8985
08:40:18.244 --> 08:40:19.365
<v Jessica Timo (OPEIU Local 153)>we represent

8986
08:40:19.525 --> 08:40:22.164
<v Jessica Timo (OPEIU Local 153)>and cascading effect that it will have

8987
08:40:26.010 --> 08:40:31.210
<v Jessica Timo (OPEIU Local 153)>workers who are forced to use AI with no
autonomy regarding the implementation.

8988
08:40:31.610 --> 08:40:36.410
<v Jessica Timo (OPEIU Local 153)>Local We know that this trend is not only
affecting our local, but workers all

8989
08:40:36.410 --> 08:40:37.450
<v Jessica Timo (OPEIU Local 153)>across the city,

8990
08:40:37.770 --> 08:40:39.610
<v Jessica Timo (OPEIU Local 153)>unionized and nonunionized.

8991
08:40:40.135 --> 08:40:43.895
<v Jessica Timo (OPEIU Local 153)>We want workers to control the
effects that AI have in the workplace.

8992
08:40:44.375 --> 08:40:49.174
<v Jessica Timo (OPEIU Local 153)>Local one five three's position is clear.
We want workers to have control over the

8993
08:40:49.174 --> 08:40:51.975
<v Jessica Timo (OPEIU Local 153)>effect that AI and the right to negotiate

8994
08:40:51.975 --> 08:40:53.254
<v Jessica Timo (OPEIU Local 153)>over its implementation.

8995
08:40:53.900 --> 08:40:56.781
<v Jessica Timo (OPEIU Local 153)>We call on the city
council to in... And enact

8996
08:40:57.020 --> 08:40:57.900
<v Jessica Timo (OPEIU Local 153)>legislator

8997
08:40:57.900 --> 08:41:01.340
<v Jessica Timo (OPEIU Local 153)>to strengthen workers'
position across the sectors

8998
08:41:01.580 --> 08:41:06.300
<v Jessica Timo (OPEIU Local 153)>to have ownership and autonomy over the
impact that AI has in the workplace.

8999
08:41:06.846 --> 08:41:11.086
<v Jessica Timo (OPEIU Local 153)>In this spirit, we urge the council to
work with the Merrell administration to

9000
08:41:11.086 --> 08:41:13.727
<v Jessica Timo (OPEIU Local 153)>listen to workers over
the tech billionaires

9001
08:41:13.805 --> 08:41:15.887
<v Jessica Timo (OPEIU Local 153)>and be on the right side of history.

9002
08:41:16.445 --> 08:41:20.525
<v Jessica Timo (OPEIU Local 153)>Thank you again, New York City Council
speakers and Julia Menon. We look forward to

9003
08:41:20.525 --> 08:41:24.560
<v Jessica Timo (OPEIU Local 153)>more action, and workers across
the city is depending on you.

9004
08:41:26.080 --> 08:41:30.879
<v CM Carmen De La Rosa>Thank you so much. We'll now hear from Sam
Wheeler. Sam, please unmute yourself and

9005
08:41:30.879 --> 08:41:31.200
<v CM Carmen De La Rosa>begin.

9006
08:41:34.775 --> 08:41:38.295
<v Sam Wheeler (Writers Guild of America East)>Good evening. My name is Sam Wheeler, and
I'm the executive director of the Writers

9007
08:41:38.295 --> 08:41:42.936
<v Sam Wheeler (Writers Guild of America East)>Guild of America East. We are a union
of more than 7,000 writers and creators

9008
08:41:42.936 --> 08:41:47.254
<v Sam Wheeler (Writers Guild of America East)>working in film, television, broadcast
news, nonfiction television, and digital

9009
08:41:47.254 --> 08:41:51.549
<v Sam Wheeler (Writers Guild of America East)>media. Most of the film and television you
watch and much of the news you consume is

9010
08:41:51.549 --> 08:41:53.229
<v Sam Wheeler (Writers Guild of America East)>likely produced by guild members,

9011
08:41:53.469 --> 08:41:56.750
<v Sam Wheeler (Writers Guild of America East)>either our members or the members of our
sibling union, the Raiders Guild of America

9012
08:41:56.750 --> 08:41:57.309
<v Sam Wheeler (Writers Guild of America East)>West.

9013
08:41:57.790 --> 08:42:01.775
<v Sam Wheeler (Writers Guild of America East)>At the outset, I would like to thank speaker
Menon along with chair De La Rosa for

9014
08:42:01.775 --> 08:42:04.256
<v Sam Wheeler (Writers Guild of America East)>convening this whole council
hearing on this issue.

9015
08:42:04.736 --> 08:42:07.217
<v Sam Wheeler (Writers Guild of America East)>Our members are
fundamentally storytellers,

9016
08:42:07.217 --> 08:42:08.656
<v Sam Wheeler (Writers Guild of America East)>journalists, reporters.

9017
08:42:08.816 --> 08:42:12.896
<v Sam Wheeler (Writers Guild of America East)>They are the people that others turn to
to help them understand our increasingly

9018
08:42:12.896 --> 08:42:13.695
<v Sam Wheeler (Writers Guild of America East)>complex

9019
08:42:13.775 --> 08:42:17.670
<v Sam Wheeler (Writers Guild of America East)>and frightening world. Guild
members take that responsibility

9020
08:42:17.670 --> 08:42:18.389
<v Sam Wheeler (Writers Guild of America East)>seriously.

9021
08:42:18.549 --> 08:42:21.110
<v Sam Wheeler (Writers Guild of America East)>For most, their work is
both a craft and a calling.

9022
08:42:21.350 --> 08:42:25.750
<v Sam Wheeler (Writers Guild of America East)>AI is of significant concern to our members.
It is a subject... A key subject in

9023
08:42:25.750 --> 08:42:28.148
<v Sam Wheeler (Writers Guild of America East)>almost every negotiation we
have with our employers.

9024
08:42:28.506 --> 08:42:33.227
<v Sam Wheeler (Writers Guild of America East)>With very limited exceptions, AI is almost
always the last point of contention in

9025
08:42:33.227 --> 08:42:33.945
<v Sam Wheeler (Writers Guild of America East)>bargaining.

9026
08:42:34.186 --> 08:42:38.186
<v Sam Wheeler (Writers Guild of America East)>Making gains almost always requires a
strike vote, and we rarely get any...

9027
08:42:38.186 --> 08:42:42.266
<v Sam Wheeler (Writers Guild of America East)>Everything that we need. In 2023, it took
a hundred and forty eight day strike to

9028
08:42:42.266 --> 08:42:46.470
<v Sam Wheeler (Writers Guild of America East)>win the first AI protections in Hollywood,
protections that plainly state that AI

9029
08:42:46.470 --> 08:42:48.470
<v Sam Wheeler (Writers Guild of America East)>cannot be a writer under our contract.

9030
08:42:48.951 --> 08:42:53.191
<v Sam Wheeler (Writers Guild of America East)>When our members engage in collective
activity to to address AI, they're fighting

9031
08:42:53.191 --> 08:42:57.031
<v Sam Wheeler (Writers Guild of America East)>not just for themselves, but for their
audiences, their readers, and our democracy.

9032
08:42:57.916 --> 08:43:02.076
<v Sam Wheeler (Writers Guild of America East)>Because we believe that the public
wants news written by human journalists

9033
08:43:02.076 --> 08:43:03.436
<v Sam Wheeler (Writers Guild of America East)>who can contextualize

9034
08:43:03.436 --> 08:43:04.156
<v Sam Wheeler (Writers Guild of America East)>events,

9035
08:43:04.316 --> 08:43:07.756
<v Sam Wheeler (Writers Guild of America East)>who can verify facts and check
that sources are credible,

9036
08:43:07.996 --> 08:43:11.275
<v Sam Wheeler (Writers Guild of America East)>who live in the same physical
world as their readers.

9037
08:43:11.790 --> 08:43:15.469
<v Sam Wheeler (Writers Guild of America East)>They want movies and television shows
written by a diverse group of other human

9038
08:43:15.469 --> 08:43:19.950
<v Sam Wheeler (Writers Guild of America East)>beings who can take their real lived
experience and turn those experiences

9039
08:43:20.030 --> 08:43:22.590
<v Sam Wheeler (Writers Guild of America East)>into stories that tell universal truths.

9040
08:43:22.988 --> 08:43:27.105
<v Sam Wheeler (Writers Guild of America East)>We have a fundamental disagreement with
the tech industry on this point. They seem

9041
08:43:27.105 --> 08:43:30.465
<v Sam Wheeler (Writers Guild of America East)>to think that what people
want is AI generated slop,

9042
08:43:30.705 --> 08:43:35.586
<v Sam Wheeler (Writers Guild of America East)>content that is so generic that no one would
ever bother asking who conceived of it.

9043
08:43:35.744 --> 08:43:37.346
<v Sam Wheeler (Writers Guild of America East)>Because even if it's bad... Times expired.

9044
08:43:39.490 --> 08:43:44.129
<v CM Carmen De La Rosa>Thank you so much for your testimony
today. It's it's an important testimony.

9045
08:43:44.370 --> 08:43:46.129
<v CM Carmen De La Rosa>Please submit it for the record.

9046
08:43:46.530 --> 08:43:51.809
<v CM Carmen De La Rosa>We're gonna call Gary Marcus who is back
on Zoom. Gary, if you're on, please unmute

9047
08:43:51.809 --> 08:43:53.010
<v CM Carmen De La Rosa>yourself and begin.

9048
08:43:54.666 --> 08:43:55.867
<v Gary Marcus (public)>Can you hear me?

9049
08:43:57.227 --> 08:44:02.266
<v Gary Marcus (public)>You hear me? Yes. We hear you, Gary. My
name is Gary Marcus. I'm a scientist,

9050
08:44:02.266 --> 08:44:06.986
<v Gary Marcus (public)>author, and NYU professor emeritus who
cofounded an AI startup that was acquired by

9051
08:44:06.986 --> 08:44:10.750
<v Gary Marcus (public)>Uber. My substack, Marcus
on AI, has about a 120,000

9052
08:44:10.750 --> 08:44:13.150
<v Gary Marcus (public)>subscribers. On x, I have about 250,000

9053
08:44:13.150 --> 08:44:17.150
<v Gary Marcus (public)>followers. I'm frequently in the media
often described as a leading AI expert and

9054
08:44:17.150 --> 08:44:18.030
<v Gary Marcus (public)>voice of reason.

9055
08:44:18.510 --> 08:44:22.110
<v Gary Marcus (public)>We stand at a pivotal moment in history.
New York City has a chance to improve our

9056
08:44:22.110 --> 08:44:26.645
<v Gary Marcus (public)>chances both for itself and the world.
Contemporary AI is like a bull in a China

9057
08:44:26.645 --> 08:44:29.365
<v Gary Marcus (public)>shop, powerful and impressive
but difficult to control.

9058
08:44:29.604 --> 08:44:31.045
<v Gary Marcus (public)>Already, the advocate...

9059
08:44:31.203 --> 08:44:35.443
<v Gary Marcus (public)>The advent of general purpose agents with
broad Internet access has led to tens of

9060
08:44:35.443 --> 08:44:36.725
<v Gary Marcus (public)>thousands of hacking incidents.

9061
08:44:37.320 --> 08:44:40.280
<v Gary Marcus (public)>Other related technologies like deepfakes,

9062
08:44:41.240 --> 08:44:46.680
<v Gary Marcus (public)>and, chatbots in in... Implicated in suicides
and even mass murder have caused other

9063
08:44:46.680 --> 08:44:51.234
<v Gary Marcus (public)>problems. It's clear by now that AI comes
with serious risks. It's also clear that

9064
08:44:51.234 --> 08:44:54.836
<v Gary Marcus (public)>the current administration in DC,
deeply tied to the big tech companies,

9065
08:44:54.996 --> 08:45:00.035
<v Gary Marcus (public)>is unwilling to do anything about it. What
we need above all else is an FDA like

9066
08:45:00.035 --> 08:45:04.600
<v Gary Marcus (public)>regulatory regime for AI in which AI
developers bring their software to independent

9067
08:45:04.600 --> 08:45:09.720
<v Gary Marcus (public)>review, much as drug manufacturers bring
their their new drugs to FDA for review.

9068
08:45:09.960 --> 08:45:13.400
<v Gary Marcus (public)>AI developers should be obligated to
demonstrate that the benefits of their new

9069
08:45:13.400 --> 08:45:16.281
<v Gary Marcus (public)>products outweigh the risks
and not have Marcus...

9070
08:45:16.440 --> 08:45:21.836
<v Gary Marcus (public)>Market access otherwise. I said the same
to Louisiana senator Kennedy in May 2023

9071
08:45:21.996 --> 08:45:26.156
<v Gary Marcus (public)>at the US senate, and it's still true, but
the US senate did nothing. New York City

9072
08:45:26.156 --> 08:45:30.557
<v Gary Marcus (public)>is thus in a position to do what the US
senate failed to do, demand independent

9073
08:45:30.557 --> 08:45:33.035
<v Gary Marcus (public)>regulatory review for
AI used in its borders.

9074
08:45:33.560 --> 08:45:37.880
<v Gary Marcus (public)>The bill on third party validation proposes
to do just that, and it would protect

9075
08:45:37.880 --> 08:45:39.000
<v Gary Marcus (public)>New York citizens

9076
08:45:39.080 --> 08:45:42.680
<v Gary Marcus (public)>from some of the harms of AI and could
allow New York to be a global leader much as

9077
08:45:42.680 --> 08:45:47.240
<v Gary Marcus (public)>California has been a global reader...
Leader around clean energy standards.

9078
08:45:47.645 --> 08:45:52.045
<v Gary Marcus (public)>In the strongest possible terms, I urge
counsel to support this bill. Finally,

9079
08:45:52.205 --> 08:45:55.885
<v Gary Marcus (public)>events of recent days suggest OpenAI
may have fired employees who were

9080
08:45:55.885 --> 08:45:56.844
<v Gary Marcus (public)>whistleblowing.

9081
08:45:56.924 --> 08:45:59.164
<v Gary Marcus (public)>Without strong protections
from whistleblowing,

9082
08:45:59.244 --> 08:46:01.885
<v Gary Marcus (public)>like New York City is considered...
Time's expired.... Trouble.

9083
08:46:03.980 --> 08:46:07.500
<v CM Carmen De La Rosa>Thank you, mister Martin.... 1%.
There. I'll just stop. Please,

9084
08:46:07.660 --> 08:46:10.860
<v CM Carmen De La Rosa>submit the rest of your testimony
for our consideration. And

9085
08:46:11.420 --> 08:46:14.860
<v CM Carmen De La Rosa>I wanna thank this panel again for
taking the time for your patience

9086
08:46:15.020 --> 08:46:18.620
<v CM Carmen De La Rosa>and understanding. It's been a long day,
and we appreciate your testimony today.

9087
08:46:20.184 --> 08:46:22.184
<v CM Carmen De La Rosa>The next panel that we're calling

9088
08:46:22.424 --> 08:46:25.943
<v CM Carmen De La Rosa>is also a hybrid panel. We have Mayu Mayu

9089
08:46:25.943 --> 08:46:26.824
<v CM Carmen De La Rosa>Tobin

9090
08:46:26.824 --> 08:46:27.785
<v CM Carmen De La Rosa>Mayai.

9091
08:46:27.785 --> 08:46:30.424
<v CM Carmen De La Rosa>Sorry. If I messed up your
name, please correct it.

9092
08:46:30.824 --> 08:46:32.184
<v CM Carmen De La Rosa>Irene Tung,

9093
08:46:32.424 --> 08:46:33.545
<v CM Carmen De La Rosa>Bryce Rogers,

9094
08:46:34.000 --> 08:46:35.200
<v CM Carmen De La Rosa>Sophie Cope,

9095
08:46:35.760 --> 08:46:36.640
<v CM Carmen De La Rosa>Brianna

9096
08:46:36.640 --> 08:46:37.600
<v CM Carmen De La Rosa>Laser

9097
08:46:38.000 --> 08:46:39.041
<v CM Carmen De La Rosa>on Zoom,

9098
08:46:39.280 --> 08:46:42.160
<v CM Carmen De La Rosa>Brian Romero on Zoom, David Carter,

9099
08:46:42.800 --> 08:46:44.480
<v CM Carmen De La Rosa>and Randall Fox.

9100
08:46:45.680 --> 08:46:49.355
<v CM Carmen De La Rosa>If your name was called, please come up to
the desk. We'll begin with the in person

9101
08:46:49.355 --> 08:46:50.795
<v CM Carmen De La Rosa>panel, and you may begin.

9102
08:46:53.275 --> 08:46:54.555
<v Mayu Toviniagi (EPIC)>Is it... Okay.

9103
08:46:54.875 --> 08:46:59.674
<v Mayu Toviniagi (EPIC)>Good evening, speaker Menin and members of
the council. My name is Mayu Toviniagi,

9104
08:47:00.260 --> 08:47:05.780
<v Mayu Toviniagi (EPIC)>and I'm an attorney at the Electronic Privacy
Information Center or EPIC. EPIC is an

9105
08:47:05.780 --> 08:47:09.540
<v Mayu Toviniagi (EPIC)>independent nonprofit
organization established in 1994

9106
08:47:09.540 --> 08:47:12.020
<v Mayu Toviniagi (EPIC)>to secure the right to
privacy in a digital age.

9107
08:47:12.885 --> 08:47:16.805
<v Mayu Toviniagi (EPIC)>Thank you to the council for organizing this
hearing and putting the AI companies on

9108
08:47:16.805 --> 08:47:17.525
<v Mayu Toviniagi (EPIC)>the spot.

9109
08:47:18.244 --> 08:47:22.805
<v Mayu Toviniagi (EPIC)>There is no reason that AI companies should
be able to release products that they

9110
08:47:22.805 --> 08:47:25.205
<v Mayu Toviniagi (EPIC)>cannot control and then avoid liability.

9111
08:47:26.010 --> 08:47:28.970
<v Mayu Toviniagi (EPIC)>Despite the hypothetical
harms that receive attention,

9112
08:47:29.129 --> 08:47:32.970
<v Mayu Toviniagi (EPIC)>there are ongoing harms that are
currently affecting New Yorkers.

9113
08:47:33.209 --> 08:47:37.690
<v Mayu Toviniagi (EPIC)>The council should focus on holding AI
companies liable beyond transparency

9114
08:47:37.690 --> 08:47:38.010
<v Mayu Toviniagi (EPIC)>measures.

9115
08:47:39.064 --> 08:47:43.465
<v Mayu Toviniagi (EPIC)>AI companies are deploying AI agents that
seek as much sensitive data about its

9116
08:47:43.465 --> 08:47:44.666
<v Mayu Toviniagi (EPIC)>users as possible,

9117
08:47:44.826 --> 08:47:49.545
<v Mayu Toviniagi (EPIC)>training its AI models by default with
questionable privacy practices and cyber

9118
08:47:49.545 --> 08:47:50.266
<v Mayu Toviniagi (EPIC)>security risks.

9119
08:47:50.830 --> 08:47:55.230
<v Mayu Toviniagi (EPIC)>Other harms include AI chatbots that
produce false or dangerous outputs,

9120
08:47:55.469 --> 08:47:57.470
<v Mayu Toviniagi (EPIC)>businesses using surveillance pricing,

9121
08:47:57.709 --> 08:48:00.910
<v Mayu Toviniagi (EPIC)>or employers surveilling its
workers for union campaigns.

9122
08:48:01.469 --> 08:48:05.150
<v Mayu Toviniagi (EPIC)>Some of the bills considered today
would address some of those real harms.

9123
08:48:06.176 --> 08:48:07.775
<v Mayu Toviniagi (EPIC)>Intro 2599

9124
08:48:07.775 --> 08:48:11.295
<v Mayu Toviniagi (EPIC)>would implement a version of
the people first chatbot bill,

9125
08:48:11.295 --> 08:48:15.615
<v Mayu Toviniagi (EPIC)>which Epic developed with Consumer
Federation of America and Fair Play.

9126
08:48:16.096 --> 08:48:19.135
<v Mayu Toviniagi (EPIC)>The bill would treat chatbots
for what they are, products,

9127
08:48:19.580 --> 08:48:23.340
<v Mayu Toviniagi (EPIC)>and hold accountable the companies
that make them when they cause harm.

9128
08:48:23.900 --> 08:48:27.740
<v Mayu Toviniagi (EPIC)>Intro 2600 was similarly
imposed common sense liability

9129
08:48:27.740 --> 08:48:30.300
<v Mayu Toviniagi (EPIC)>on AI companies with a
private right of action.

9130
08:48:30.620 --> 08:48:34.300
<v Mayu Toviniagi (EPIC)>We support these bills and urge the
council to ensure that there are sufficient

9131
08:48:34.395 --> 08:48:35.195
<v Mayu Toviniagi (EPIC)>enforcement

9132
08:48:35.195 --> 08:48:36.314
<v Mayu Toviniagi (EPIC)>resources.

9133
08:48:36.635 --> 08:48:40.475
<v Mayu Toviniagi (EPIC)>We urge the council to prioritize these
bills to address these harms that are

9134
08:48:40.475 --> 08:48:42.154
<v Mayu Toviniagi (EPIC)>already affecting New Yorkers.

9135
08:48:42.475 --> 08:48:46.154
<v Mayu Toviniagi (EPIC)>Thank you for the opportunity to testify
today and happy to answer any questions.

9136
08:48:46.635 --> 08:48:47.195
<v CM Carmen De La Rosa>Thank you.

9137
08:48:52.041 --> 08:48:56.521
<v Bryce Rogers (RAD Collective)>I thank the council for leading the charge
on local AI legislation by convening this

9138
08:48:56.521 --> 08:48:57.640
<v Bryce Rogers (RAD Collective)>historic hearing.

9139
08:48:57.960 --> 08:49:02.441
<v Bryce Rogers (RAD Collective)>My name is Bryce Rogers. I'm a resident
of the 34th Council District. I thank

9140
08:49:02.441 --> 08:49:04.441
<v Bryce Rogers (RAD Collective)>council member Gutierrez
for representing me.

9141
08:49:05.266 --> 08:49:09.186
<v Bryce Rogers (RAD Collective)>I'm co lead of the Resisting
Automated Disempowerment Collective

9142
08:49:09.186 --> 08:49:10.705
<v Bryce Rogers (RAD Collective)>or RAD Collective,

9143
08:49:10.865 --> 08:49:15.266
<v Bryce Rogers (RAD Collective)>a New York City grassroots community
advocating for city and state AI policy

9144
08:49:15.346 --> 08:49:17.986
<v Bryce Rogers (RAD Collective)>that protects human
life, joy, and democracy.

9145
08:49:20.430 --> 08:49:24.191
<v Bryce Rogers (RAD Collective)>The great strength of humanity is our
ability to transfer information to one

9146
08:49:24.191 --> 08:49:28.031
<v Bryce Rogers (RAD Collective)>another. We are blessed with
expressive hands, faces, and voices,

9147
08:49:28.270 --> 08:49:32.510
<v Bryce Rogers (RAD Collective)>but have also constantly sought to augment
our inherent communicative capabilities

9148
08:49:32.510 --> 08:49:33.390
<v Bryce Rogers (RAD Collective)>via technology.

9149
08:49:34.416 --> 08:49:38.977
<v Bryce Rogers (RAD Collective)>To collaborate on complex and abstract
topics, we developed language and systems of

9150
08:49:38.977 --> 08:49:39.617
<v Bryce Rogers (RAD Collective)>measurement.

9151
08:49:40.016 --> 08:49:44.656
<v Bryce Rogers (RAD Collective)>To convey information farther than our
voice can reach, we built postal services,

9152
08:49:44.656 --> 08:49:46.256
<v Bryce Rogers (RAD Collective)>telegraphs, and telephones.

9153
08:49:46.736 --> 08:49:51.270
<v Bryce Rogers (RAD Collective)>And to transmit messages across time,
we've developed oral histories, library

9154
08:49:51.270 --> 08:49:52.070
<v Bryce Rogers (RAD Collective)>collections,

9155
08:49:52.070 --> 08:49:54.070
<v Bryce Rogers (RAD Collective)>and recently digital archives,

9156
08:49:54.230 --> 08:49:58.150
<v Bryce Rogers (RAD Collective)>the evolving means by which each
generation inherits their predecessors'

9157
08:49:58.150 --> 08:49:59.510
<v Bryce Rogers (RAD Collective)>infrastructure and wisdom.

9158
08:50:00.390 --> 08:50:04.545
<v Bryce Rogers (RAD Collective)>Through these technologies, humanity has
accrued the wealth of information upon

9159
08:50:04.545 --> 08:50:06.785
<v Bryce Rogers (RAD Collective)>which artificial
intelligence is predicated.

9160
08:50:07.424 --> 08:50:11.584
<v Bryce Rogers (RAD Collective)>AI distills the meaningful patterns from
very large bodies of training data,

9161
08:50:11.824 --> 08:50:15.824
<v Bryce Rogers (RAD Collective)>not just navigating the information as
a search engine does, but digesting it.

9162
08:50:16.590 --> 08:50:18.110
<v Bryce Rogers (RAD Collective)>Converting our species'

9163
08:50:18.190 --> 08:50:22.590
<v Bryce Rogers (RAD Collective)>informatic endowment into an
inquirerable oracle like machine.

9164
08:50:23.389 --> 08:50:26.030
<v Bryce Rogers (RAD Collective)>I can't shake the feeling that
this could have been beautiful,

9165
08:50:26.430 --> 08:50:27.790
<v Bryce Rogers (RAD Collective)>but it just doesn't feel like us.

9166
08:50:29.195 --> 08:50:33.035
<v Bryce Rogers (RAD Collective)>These companies trained their products
on millennia of our predecessors dogged

9167
08:50:33.035 --> 08:50:37.834
<v Bryce Rogers (RAD Collective)>achievement only to deploy their tech
to out compete us, to spy on us, and to

9168
08:50:37.834 --> 08:50:38.715
<v Bryce Rogers (RAD Collective)>replace us.

9169
08:50:39.514 --> 08:50:43.275
<v Bryce Rogers (RAD Collective)>All of this is to say, democracy is
the only righteous path forward.

9170
08:50:43.951 --> 08:50:48.432
<v Bryce Rogers (RAD Collective)>AI is built on the work of many, and its
fate oughtn't be dictated by a few self

9171
08:50:48.432 --> 08:50:49.871
<v Bryce Rogers (RAD Collective)>regulating executives

9172
08:50:49.871 --> 08:50:52.832
<v Bryce Rogers (RAD Collective)>regardless of how their
spokespeople reassure us over Zoom.

9173
08:50:53.630 --> 08:50:56.990
<v Bryce Rogers (RAD Collective)>In this spirit, we again thank the
council for convening this forum of local

9174
08:50:56.990 --> 08:50:57.630
<v Bryce Rogers (RAD Collective)>democracy

9175
08:50:57.855 --> 08:51:01.535
<v Bryce Rogers (RAD Collective)>and implore you to foreground the diverse
constituent voices of New York City in

9176
08:51:01.535 --> 08:51:02.576
<v Bryce Rogers (RAD Collective)>your legislation.

9177
08:51:03.135 --> 08:51:07.295
<v Bryce Rogers (RAD Collective)>Rad wrote a letter to the council, that I'd
love for you to check out if you haven't

9178
08:51:07.295 --> 08:51:08.416
<v Bryce Rogers (RAD Collective)>already, and,

9179
08:51:09.295 --> 08:51:13.852
<v Bryce Rogers (RAD Collective)>we look forward to advancing the moral
frontier of AI policy together. Thank you so

9180
08:51:13.852 --> 08:51:14.890
<v Bryce Rogers (RAD Collective)>much. Thank you.

9181
08:51:18.650 --> 08:51:22.650
<v Sophie Cody (NYC DSA tech action working group)>Hi. Thank you, counsel. Although, I think
I'm the first without the the speaker in

9182
08:51:22.650 --> 08:51:23.932
<v Sophie Cody (NYC DSA tech action working group)>the audience to hear.

9183
08:51:24.810 --> 08:51:26.250
<v Sophie Cody (NYC DSA tech action working group)>Should we wait for her to come back?

9184
08:51:26.945 --> 08:51:29.346
<v CM Carmen De La Rosa>You can go ahead. You
can go ahead. Thank you.

9185
08:51:30.145 --> 08:51:33.904
<v Sophie Cody (NYC DSA tech action working group)>So thank you to the council for hosting
this important hearing. My name is Sophie

9186
08:51:33.904 --> 08:51:38.385
<v Sophie Cody (NYC DSA tech action working group)>Cody. I am a member of the NYC DSA tech
action working group, although I'm not

9187
08:51:38.385 --> 08:51:39.744
<v Sophie Cody (NYC DSA tech action working group)>speaking on their behalf tonight.

9188
08:51:40.400 --> 08:51:44.160
<v Sophie Cody (NYC DSA tech action working group)>Instead, I wanna talk about the two men
whose comments were the impetus of this

9189
08:51:44.160 --> 08:51:46.800
<v Sophie Cody (NYC DSA tech action working group)>hearing, Sam Altman and Dario Amadeh,

9190
08:51:46.800 --> 08:51:49.361
<v Sophie Cody (NYC DSA tech action working group)>AI CEOs at the helm of this crisis.

9191
08:51:49.760 --> 08:51:52.960
<v Sophie Cody (NYC DSA tech action working group)>As this hearing has continued, I've
heard a lot of anthropomorphic

9192
08:51:52.960 --> 08:51:54.880
<v Sophie Cody (NYC DSA tech action working group)>language used to describe AI models.

9193
08:51:55.336 --> 08:51:56.535
<v Sophie Cody (NYC DSA tech action working group)>They understand.

9194
08:51:56.535 --> 08:52:01.096
<v Sophie Cody (NYC DSA tech action working group)>They think. They can break containment.
They have agency. They grow. Dario Amade

9195
08:52:01.096 --> 08:52:03.416
<v Sophie Cody (NYC DSA tech action working group)>even implied that Claude may be conscious.

9196
08:52:03.896 --> 08:52:07.656
<v Sophie Cody (NYC DSA tech action working group)>These men show a cultish concern for
the unproven claims of superintelligence

9197
08:52:07.656 --> 08:52:11.920
<v Sophie Cody (NYC DSA tech action working group)>and the consciousness of their own models,
and they do not care about women or

9198
08:52:11.920 --> 08:52:16.480
<v Sophie Cody (NYC DSA tech action working group)>children that are harmed by their products
at all. Stories abound, and we heard them

9199
08:52:16.480 --> 08:52:20.240
<v Sophie Cody (NYC DSA tech action working group)>tonight of women and girls being victims
of AI notification tools and deepfake

9200
08:52:20.240 --> 08:52:21.680
<v Sophie Cody (NYC DSA tech action working group)>pornographic images and videos.

9201
08:52:22.314 --> 08:52:26.475
<v Sophie Cody (NYC DSA tech action working group)>AI is used to generate childhood... Child
sexual abuse material. We also heard about

9202
08:52:26.475 --> 08:52:27.674
<v Sophie Cody (NYC DSA tech action working group)>tonight already.

9203
08:52:27.914 --> 08:52:31.354
<v Sophie Cody (NYC DSA tech action working group)>AI sex slaves are advertised
to young boys online.

9204
08:52:31.674 --> 08:52:35.834
<v Sophie Cody (NYC DSA tech action working group)>Sam Altman and Dario Amadeh both actively
collaborate with our rapist in chief,

9205
08:52:35.834 --> 08:52:36.395
<v Sophie Cody (NYC DSA tech action working group)>Donald Trump.

9206
08:52:36.910 --> 08:52:40.110
<v Sophie Cody (NYC DSA tech action working group)>These men claim they need help
controlling themselves, their technology.

9207
08:52:40.270 --> 08:52:44.190
<v Sophie Cody (NYC DSA tech action working group)>At work online and in our personal lives,
we hear phrases like we need to embrace

9208
08:52:44.190 --> 08:52:49.470
<v Sophie Cody (NYC DSA tech action working group)>Claude and use phrases like AI is being
shoved down our throats and forced upon us.

9209
08:52:49.550 --> 08:52:51.230
<v Sophie Cody (NYC DSA tech action working group)>We heard that it's inevitable.

9210
08:52:51.310 --> 08:52:52.430
<v Sophie Cody (NYC DSA tech action working group)>Does that sound familiar?

9211
08:52:53.285 --> 08:52:56.965
<v Sophie Cody (NYC DSA tech action working group)>I am appalled that it takes statements of
these men to call this hearing. What I

9212
08:52:56.965 --> 08:52:58.645
<v Sophie Cody (NYC DSA tech action working group)>listed above should have been enough.

9213
08:52:58.965 --> 08:53:01.285
<v Sophie Cody (NYC DSA tech action working group)>Sam Altman is a deranged sociopath.

9214
08:53:01.365 --> 08:53:06.084
<v Sophie Cody (NYC DSA tech action working group)>Dario Amade is in a cult. These men are
drunk drivers behind the wheels of this

9215
08:53:06.084 --> 08:53:10.580
<v Sophie Cody (NYC DSA tech action working group)>runaway bus that nobody asked to be on.
Right now, New Yorkers have the opportunity

9216
08:53:10.580 --> 08:53:13.860
<v Sophie Cody (NYC DSA tech action working group)>to be the anchor in this tug of war for
power between the people and these tech

9217
08:53:13.860 --> 08:53:17.460
<v Sophie Cody (NYC DSA tech action working group)>billionaires who believe more in the
consciousness of their own AI than the humanity

9218
08:53:17.460 --> 08:53:23.734
<v Sophie Cody (NYC DSA tech action working group)>of women. I close and appeal to you, the
city council, listen to us, not them.

9219
08:53:23.975 --> 08:53:24.695
<v Sophie Cody (NYC DSA tech action working group)>Thank you.

9220
08:53:25.494 --> 08:53:27.416
<v CM Carmen De La Rosa>Thank you so much for your testimony.

9221
08:53:28.695 --> 08:53:32.455
<v David Carter (Ecommerce Innovation Alliance)>Good evening, chair De La Rosa, members of
the council. My name is David Carter. I'm

9222
08:53:32.455 --> 08:53:37.070
<v David Carter (Ecommerce Innovation Alliance)>the president and CEO of the Ecommerce
Innovation Alliance. I represent about 20,000

9223
08:53:37.070 --> 08:53:38.990
<v David Carter (Ecommerce Innovation Alliance)>ecommerce companies across the country,

9224
08:53:39.390 --> 08:53:42.352
<v David Carter (Ecommerce Innovation Alliance)>including many that are based in your
districts right here in New York.

9225
08:53:42.750 --> 08:53:45.791
<v David Carter (Ecommerce Innovation Alliance)>And today, I just wanna take a couple
minutes of your time to speak to you about a

9226
08:53:45.791 --> 08:53:49.791
<v David Carter (Ecommerce Innovation Alliance)>few unintended consequences. I've worked
with legislatures across the country when

9227
08:53:49.791 --> 08:53:53.453
<v David Carter (Ecommerce Innovation Alliance)>they've been implementing legislation,
particularly around telemarketing,

9228
08:53:53.453 --> 08:53:57.295
<v David Carter (Ecommerce Innovation Alliance)>to try to make sure legislation gets to
the heart of the issues you care about

9229
08:53:57.533 --> 08:53:59.215
<v David Carter (Ecommerce Innovation Alliance)>without unintended

9230
08:53:59.215 --> 08:54:01.934
<v David Carter (Ecommerce Innovation Alliance)>consequences for small
and mid sized businesses.

9231
08:54:02.174 --> 08:54:05.660
<v David Carter (Ecommerce Innovation Alliance)>And in particular, I wanted to speak to
you about twenty five ninety nine, the

9232
08:54:05.660 --> 08:54:06.700
<v David Carter (Ecommerce Innovation Alliance)>chatbot The

9233
08:54:07.500 --> 08:54:10.060
<v David Carter (Ecommerce Innovation Alliance)>chatbot bill uses vague language

9234
08:54:10.060 --> 08:54:12.300
<v David Carter (Ecommerce Innovation Alliance)>to define what is a chatbot.

9235
08:54:12.620 --> 08:54:14.220
<v David Carter (Ecommerce Innovation Alliance)>Anything that stimulates

9236
08:54:14.220 --> 08:54:16.060
<v David Carter (Ecommerce Innovation Alliance)>or simulates conversation

9237
08:54:16.060 --> 08:54:20.795
<v David Carter (Ecommerce Innovation Alliance)>is a chatbot, and anyone that makes
a chatbot available is subject to the

9238
08:54:20.795 --> 08:54:22.875
<v David Carter (Ecommerce Innovation Alliance)>regulations in 2599.

9239
08:54:23.035 --> 08:54:27.115
<v David Carter (Ecommerce Innovation Alliance)>So I wanna talk to you for a moment about
ecommerce companies and how they work. You

9240
08:54:27.115 --> 08:54:29.115
<v David Carter (Ecommerce Innovation Alliance)>go to their website. You volunteer

9241
08:54:29.115 --> 08:54:33.080
<v David Carter (Ecommerce Innovation Alliance)>to sign up for text text messages that you
get your discount codes. Most people like

9242
08:54:33.080 --> 08:54:35.639
<v David Carter (Ecommerce Innovation Alliance)>the discount codes. They
don't want unintended

9243
08:54:35.639 --> 08:54:36.520
<v David Carter (Ecommerce Innovation Alliance)>consequences.

9244
08:54:36.520 --> 08:54:37.959
<v David Carter (Ecommerce Innovation Alliance)>They don't want a lot of messages.

9245
08:54:38.440 --> 08:54:43.080
<v David Carter (Ecommerce Innovation Alliance)>Under this bill, a company that uses AI
to communicate with consumers that have

9246
08:54:43.080 --> 08:54:44.760
<v David Carter (Ecommerce Innovation Alliance)>asked to receive text messages

9247
08:54:45.055 --> 08:54:47.936
<v David Carter (Ecommerce Innovation Alliance)>would be required to remind the consumer

9248
08:54:48.094 --> 08:54:49.295
<v David Carter (Ecommerce Innovation Alliance)>every hour

9249
08:54:49.375 --> 08:54:53.615
<v David Carter (Ecommerce Innovation Alliance)>on the hour that they're engaging
in a conversation with an AI.

9250
08:54:53.934 --> 08:54:58.654
<v David Carter (Ecommerce Innovation Alliance)>That works if you're on a website that you
open and close the chatbot conversation.

9251
08:54:59.200 --> 08:55:03.440
<v David Carter (Ecommerce Innovation Alliance)>It doesn't work when you're engaged in text
message conversations that can last for

9252
08:55:03.440 --> 08:55:05.040
<v David Carter (Ecommerce Innovation Alliance)>days or weeks at a time.

9253
08:55:05.520 --> 08:55:10.000
<v David Carter (Ecommerce Innovation Alliance)>Under this legislation though, if a
company uses AI just to communicate with

9254
08:55:10.000 --> 08:55:14.875
<v David Carter (Ecommerce Innovation Alliance)>consumers and they don't send out that
reminder every hour on the hour, they're

9255
08:55:14.875 --> 08:55:16.875
<v David Carter (Ecommerce Innovation Alliance)>subject not only to $25,000

9256
08:55:16.875 --> 08:55:17.676
<v David Carter (Ecommerce Innovation Alliance)>penalties,

9257
08:55:17.836 --> 08:55:19.195
<v David Carter (Ecommerce Innovation Alliance)>but they're subject to,

9258
08:55:19.756 --> 08:55:21.996
<v David Carter (Ecommerce Innovation Alliance)>private litigation and other enforcement.

9259
08:55:22.236 --> 08:55:26.875
<v David Carter (Ecommerce Innovation Alliance)>This is a recipe to put businesses out
of business, unfortunately. We have an

9260
08:55:26.875 --> 08:55:28.611
<v David Carter (Ecommerce Innovation Alliance)>industry of professional

9261
08:55:28.611 --> 08:55:30.531
<v David Carter (Ecommerce Innovation Alliance)>plaintiffs and plaintiffs' attorneys

9262
08:55:30.691 --> 08:55:32.130
<v David Carter (Ecommerce Innovation Alliance)>that would be incentivized

9263
08:55:32.130 --> 08:55:32.691
<v David Carter (Ecommerce Innovation Alliance)>to

9264
08:55:33.170 --> 08:55:37.730
<v David Carter (Ecommerce Innovation Alliance)>to crowd the courts with repetitive litigation
about these issues. So I just wanna

9265
08:55:37.730 --> 08:55:42.756
<v David Carter (Ecommerce Innovation Alliance)>help the the counsel refine the language
to not have these unintended consequences.

9266
08:55:42.756 --> 08:55:44.516
<v David Carter (Ecommerce Innovation Alliance)>Thank you. If you could submit,

9267
08:55:44.676 --> 08:55:46.355
<v CM Carmen De La Rosa>the language of your testimony,

9268
08:55:46.756 --> 08:55:50.275
<v CM Carmen De La Rosa>we'd appreciate that. We have on Zoom,

9269
08:55:50.996 --> 08:55:52.436
<v CM Carmen De La Rosa>Brianna Laser.

9270
08:55:54.490 --> 08:55:57.930
<v CM Carmen De La Rosa>If you're on, Brianna, if you
could unmute yourself and begin.

9271
08:55:58.490 --> 08:55:59.129
<v Brenna Lieser (Common Sense Media)>Yes.

9272
08:55:59.450 --> 08:56:03.530
<v Brenna Lieser (Common Sense Media)>Thank you. Good evening, madam speaker
and council members. My name is Brenna

9273
08:56:03.530 --> 08:56:07.266
<v Brenna Lieser (Common Sense Media)>Lieser, and I serve as a tech policy
adviser at Common Sense Media, which is the

9274
08:56:07.266 --> 08:56:11.266
<v Brenna Lieser (Common Sense Media)>largest nonprofit organization in the
country that works to keep kids safe in a

9275
08:56:11.266 --> 08:56:12.145
<v Brenna Lieser (Common Sense Media)>digital world.

9276
08:56:12.625 --> 08:56:16.465
<v Brenna Lieser (Common Sense Media)>In New York State alone, we have 80,000
parent and teacher members and are active in

9277
08:56:16.465 --> 08:56:18.785
<v Brenna Lieser (Common Sense Media)>the legislature as well fighting
for kids online safety.

9278
08:56:19.510 --> 08:56:23.590
<v Brenna Lieser (Common Sense Media)>On our and their behalf, thank you for
considering measures to enhance AI safety for

9279
08:56:23.590 --> 08:56:24.550
<v Brenna Lieser (Common Sense Media)>kids in the city.

9280
08:56:25.110 --> 08:56:29.510
<v Brenna Lieser (Common Sense Media)>AI is a powerful technology that demands
careful consideration before any use.

9281
08:56:29.670 --> 08:56:33.110
<v Brenna Lieser (Common Sense Media)>Parents and educators specifically need
allies on the council to keep their kids

9282
08:56:33.110 --> 08:56:33.350
<v Brenna Lieser (Common Sense Media)>safe.

9283
08:56:34.025 --> 08:56:38.266
<v Brenna Lieser (Common Sense Media)>Common Sense Media's testing found that many
of the most popular AI chatbots are not

9284
08:56:38.266 --> 08:56:43.145
<v Brenna Lieser (Common Sense Media)>safe for kids. Our recommendation is that
no one 18 should use, AI for companionship

9285
08:56:43.145 --> 08:56:44.426
<v Brenna Lieser (Common Sense Media)>or mental health support.

9286
08:56:44.586 --> 08:56:48.186
<v Brenna Lieser (Common Sense Media)>And, tragically, there have been many cases
tying AI chatbots to teen suicides and

9287
08:56:48.186 --> 08:56:52.559
<v Brenna Lieser (Common Sense Media)>mental health harms. Yet, AI chatbots are
increasingly embedded in our children's

9288
08:56:52.559 --> 08:56:55.200
<v Brenna Lieser (Common Sense Media)>lives through search tools,
social media platforms,

9289
08:56:55.280 --> 08:56:58.080
<v Brenna Lieser (Common Sense Media)>or through pressure to use AI
chatbots to support their education.

9290
08:56:58.398 --> 08:57:02.879
<v Brenna Lieser (Common Sense Media)>Our research found that seventy percent of
teens use AI companion chatbots, and one

9291
08:57:02.879 --> 08:57:06.217
<v Brenna Lieser (Common Sense Media)>in three prefer to interact with
them as much or more than humans.

9292
08:57:06.535 --> 08:57:10.455
<v Brenna Lieser (Common Sense Media)>This year, senator Kristin Gonzalez and
assembly member Alex Boris, with strong

9293
08:57:10.455 --> 08:57:14.137
<v Brenna Lieser (Common Sense Media)>support from attorney general James,
everyday New Yorkers, Common Sense Media,

9294
08:57:14.137 --> 08:57:17.190
<v Brenna Lieser (Common Sense Media)>NYSET, and others got the
legislature to unanimously

9295
08:57:17.190 --> 08:57:19.670
<v Brenna Lieser (Common Sense Media)>pass senate bill nine zero five one,

9296
08:57:19.910 --> 08:57:23.590
<v Brenna Lieser (Common Sense Media)>their bill to prohibit unsafe features
and AI chatbots offered to minors.

9297
08:57:24.150 --> 08:57:27.590
<v Brenna Lieser (Common Sense Media)>The bill now awaits governor Hochul's
signature, and if signed, it will be the

9298
08:57:27.590 --> 08:57:32.004
<v Brenna Lieser (Common Sense Media)>strongest youth AI safety chatbot law in
the country. It would prohibit chatbots

9299
08:57:32.004 --> 08:57:35.525
<v Brenna Lieser (Common Sense Media)>from providing unsafe features that
promote self harm, suicide, or disordered

9300
08:57:35.525 --> 08:57:39.525
<v Brenna Lieser (Common Sense Media)>eating, from encouraging minors to keep
interaction secret or to self isolate,

9301
08:57:39.525 --> 08:57:41.766
<v Brenna Lieser (Common Sense Media)>and from prioritizing
engagement over safety.

9302
08:57:42.484 --> 08:57:46.324
<v Brenna Lieser (Common Sense Media)>So we urge the city council to pass a
resolution supporting senate bill nine zero

9303
08:57:46.324 --> 08:57:50.120
<v Brenna Lieser (Common Sense Media)>five one to ensure that New York's kids
are protected from unsafe AI chatbots.

9304
08:57:50.280 --> 08:57:53.080
<v Brenna Lieser (Common Sense Media)>We thank you for this hearing and
your attention to our concerns.

9305
08:57:53.240 --> 08:57:57.080
<v CM Carmen De La Rosa>Thank you. Thank you so much.
Up next, we have Brian Romero

9306
08:57:57.080 --> 08:58:00.360
<v CM Carmen De La Rosa>on Zoom. Brian, if you're
ready, Emile, and begin.

9307
08:58:03.266 --> 08:58:05.746
<v Brian Romero (public)>Hi. Can you all hear me? We hear you.

9308
08:58:06.465 --> 08:58:11.025
<v Brian Romero (public)>Great. Good morning, council speaker Menon
and members of the council. My name is

9309
08:58:11.025 --> 08:58:14.545
<v Brian Romero (public)>Brian Romero, and I am the democratic
nominee for New York State Assembly District

9310
08:58:14.545 --> 08:58:15.025
<v Brian Romero (public)>34.

9311
08:58:15.740 --> 08:58:20.220
<v Brian Romero (public)>As the former chief of staff to senator
Kristin Gonzalez, I have front row seat as

9312
08:58:20.220 --> 08:58:23.420
<v Brian Romero (public)>New York tried to keep regulation
moving alongside a technology

9313
08:58:23.580 --> 08:58:25.900
<v Brian Romero (public)>that is evolving at an
extraordinary speed.

9314
08:58:26.380 --> 08:58:28.380
<v Brian Romero (public)>Thank you for the
opportunity to testify today.

9315
08:58:29.385 --> 08:58:33.785
<v Brian Romero (public)>As AI rapidly reshapes public services
and life as we know it, the council and

9316
08:58:33.785 --> 08:58:35.145
<v Brian Romero (public)>subsequently the state

9317
08:58:35.385 --> 08:58:39.305
<v Brian Romero (public)>must lead with both enforceable
oversight that protects workers,

9318
08:58:39.465 --> 08:58:43.545
<v Brian Romero (public)>tenants, and immigrant communities
like the ones in a D 34.

9319
08:58:44.580 --> 08:58:47.700
<v Brian Romero (public)>First, I appreciate the goal
of intro to six zero two.

9320
08:58:48.260 --> 08:58:52.420
<v Brian Romero (public)>Responsible AI governance requires
clear boundaries and transparency.

9321
08:58:52.420 --> 08:58:56.900
<v Brian Romero (public)>However, I will note that proposals around
emergency shutdown mechanisms or kill

9322
08:58:56.900 --> 08:59:01.164
<v Brian Romero (public)>switches and then general enforcement of
that scale is certainly best suited for the

9323
08:59:01.164 --> 08:59:02.686
<v Brian Romero (public)>state level and federal level.

9324
08:59:03.244 --> 08:59:06.924
<v Brian Romero (public)>It is a mechanism legislation that I
have drafted to introduce when we resume

9325
08:59:06.924 --> 08:59:08.045
<v Brian Romero (public)>session next year.

9326
08:59:08.445 --> 08:59:11.885
<v Brian Romero (public)>I'm glad that the governor has already
signaled openness to these short term kinds

9327
08:59:11.885 --> 08:59:14.190
<v Brian Romero (public)>of statewide mitigation frameworks,

9328
08:59:14.430 --> 08:59:18.430
<v Brian Romero (public)>but, obviously, we need more long term
solutions to the lack of guardrails.

9329
08:59:18.990 --> 08:59:23.310
<v Brian Romero (public)>I support the intent of two six zero one,
but it could use some more enforcement

9330
08:59:23.310 --> 08:59:27.790
<v Brian Romero (public)>teeth. Two six zero one can be amended
to include strict financial penalties.

9331
08:59:28.145 --> 08:59:31.904
<v Brian Romero (public)>I don't think that they suffice as they
currently stand, and a mechanism for being

9332
08:59:31.904 --> 08:59:36.705
<v Brian Romero (public)>terminated from Citi contracts if vendors
repeatedly fail compliance audits or

9333
08:59:36.705 --> 08:59:38.385
<v Brian Romero (public)>deploy harmful discriminatory

9334
08:59:38.385 --> 08:59:39.504
<v Brian Romero (public)>AI systems.

9335
08:59:40.145 --> 08:59:41.664
<v Brian Romero (public)>Without real consequences,

9336
08:59:41.744 --> 08:59:42.545
<v Brian Romero (public)>oversight

9337
08:59:42.545 --> 08:59:43.664
<v Brian Romero (public)>is simply a recommendation.

9338
08:59:44.760 --> 08:59:49.160
<v Brian Romero (public)>Meaningful consequences for repeated
violations can help to improve that oversight.

9339
08:59:50.040 --> 08:59:53.559
<v Brian Romero (public)>Finally, effective tech policy should
certainly include the council moving forward

9340
08:59:53.559 --> 08:59:57.799
<v Brian Romero (public)>on bold and progressive legislation that
protects New Yorkers from automated harm,

9341
08:59:58.120 --> 08:59:59.158
<v Brian Romero (public)>including in our workplaces.

9342
08:59:59.855 --> 09:00:04.574
<v Brian Romero (public)>So I urge the council to therefore discharge
and bring to a full vote intro two one

9343
09:00:04.574 --> 09:00:07.855
<v Brian Romero (public)>three or ban the scan and
the delivery protection act.

9344
09:00:08.574 --> 09:00:12.016
<v Brian Romero (public)>I also look forward to exploring how to
expand with some global experts at state

9345
09:00:12.016 --> 09:00:12.336
<v Brian Romero (public)>level.

9346
09:00:14.510 --> 09:00:18.830
<v Brian Romero (public)>Thank you so much. We need bold action from
the council and the state to show that

9347
09:00:18.830 --> 09:00:22.270
<v Brian Romero (public)>we are giving this the scrutiny the more
clarity and the bold action that it

9348
09:00:22.270 --> 09:00:25.790
<v Brian Romero (public)>deserves. Thank you. Thank you, Brian.
And we look forward to continuing the

9349
09:00:25.790 --> 09:00:28.750
<v Brian Romero (public)>conversation, and we appreciate
you taking the time and

9350
09:00:29.125 --> 09:00:33.125
<v CM Carmen De La Rosa>and look forward to reading the rest of
your testimony. I wanna... Thank you,

9351
09:00:33.125 --> 09:00:36.885
<v CM Carmen De La Rosa>Joanne. You're welcome.
I wanna see David Carter.

9352
09:00:37.285 --> 09:00:40.004
<v CM Carmen De La Rosa>Oh, you already testified.
Perfect. And Randall Fox,

9353
09:00:41.140 --> 09:00:45.140
<v CM Carmen De La Rosa>if you're in the room. Alright. Well, then
we thank this panel for being here, and

9354
09:00:45.140 --> 09:00:49.460
<v CM Carmen De La Rosa>we appreciate you taking the time and your
thoughtful testimony this this evening.

9355
09:00:49.460 --> 09:00:50.420
<v CM Carmen De La Rosa>Thank you so much.

9356
09:00:50.900 --> 09:00:51.861
<v CM Carmen De La Rosa>We're

9357
09:00:51.861 --> 09:00:52.900
<v CM Carmen De La Rosa>gonna call

9358
09:00:53.285 --> 09:00:57.045
<v CM Carmen De La Rosa>what I think is the last panel, but I'm
gonna call all of these names. And if you're

9359
09:00:57.045 --> 09:00:57.686
<v CM Carmen De La Rosa>here

9360
09:00:58.166 --> 09:01:01.926
<v CM Carmen De La Rosa>In person. In person,
please come up to the dais.

9361
09:01:02.006 --> 09:01:03.365
<v CM Carmen De La Rosa>There's a Crystal,

9362
09:01:03.605 --> 09:01:05.926
<v CM Carmen De La Rosa>no last name, Jamal Bowman,

9363
09:01:06.760 --> 09:01:08.360
<v CM Carmen De La Rosa>Elizabeth Crotty,

9364
09:01:08.920 --> 09:01:09.879
<v CM Carmen De La Rosa>Niza

9365
09:01:10.280 --> 09:01:12.520
<v CM Carmen De La Rosa>Faye... Faye...

9366
09:01:12.520 --> 09:01:13.240
<v CM Carmen De La Rosa>Faye... Faye... Faye...

9367
09:01:14.360 --> 09:01:15.799
<v CM Carmen De La Rosa>Wes Rizen,

9368
09:01:16.760 --> 09:01:17.879
<v CM Carmen De La Rosa>Elliot Liv,

9369
09:01:19.096 --> 09:01:20.535
<v CM Carmen De La Rosa>Matthew Trump,

9370
09:01:21.574 --> 09:01:22.455
<v CM Carmen De La Rosa>Robbie

9371
09:01:22.455 --> 09:01:23.496
<v CM Carmen De La Rosa>Roshmal,

9372
09:01:25.016 --> 09:01:26.615
<v CM Carmen De La Rosa>Jacqueline Payne,

9373
09:01:27.494 --> 09:01:28.535
<v CM Carmen De La Rosa>Shuruti

9374
09:01:29.096 --> 09:01:29.814
<v CM Carmen De La Rosa>Velidi,

9375
09:01:30.560 --> 09:01:31.281
<v CM Carmen De La Rosa>Sorry.

9376
09:01:31.680 --> 09:01:32.800
<v CM Carmen De La Rosa>Ambarine

9377
09:01:33.840 --> 09:01:35.121
<v CM Carmen De La Rosa>Koreshi,

9378
09:01:35.920 --> 09:01:36.880
<v CM Carmen De La Rosa>Sarai's

9379
09:01:36.880 --> 09:01:37.840
<v CM Carmen De La Rosa>Holman,

9380
09:01:38.960 --> 09:01:39.920
<v CM Carmen De La Rosa>Leonidas

9381
09:01:39.920 --> 09:01:41.602
<v CM Carmen De La Rosa>Vincent Rebkell,

9382
09:01:42.080 --> 09:01:43.441
<v CM Carmen De La Rosa>Thomas Alexander

9383
09:01:43.441 --> 09:01:43.840
<v CM Carmen De La Rosa>Kuhl,

9384
09:01:44.865 --> 09:01:45.826
<v CM Carmen De La Rosa>Daniel

9385
09:01:46.064 --> 09:01:47.586
<v CM Carmen De La Rosa>Crease,

9386
09:01:47.904 --> 09:01:49.346
<v CM Carmen De La Rosa>Reshma Patel,

9387
09:01:49.426 --> 09:01:50.625
<v CM Carmen De La Rosa>Kal Al

9388
09:01:50.785 --> 09:01:51.826
<v CM Carmen De La Rosa>Dubabi,

9389
09:01:52.705 --> 09:01:53.664
<v CM Carmen De La Rosa>Arithma

9390
09:01:53.664 --> 09:01:54.545
<v CM Carmen De La Rosa>McLean...

9391
09:01:55.266 --> 09:01:56.225
<v CM Carmen De La Rosa>McLennan...

9392
09:01:56.225 --> 09:01:56.865
<v CM Carmen De La Rosa>Nick

9393
09:01:57.984 --> 09:01:59.904
<v CM Carmen De La Rosa>p, and Bucky b.

9394
09:02:02.730 --> 09:02:04.170
<v CM Carmen De La Rosa>If you all are here,

9395
09:02:04.810 --> 09:02:07.371
<v CM Carmen De La Rosa>please come up to the dais in person.

9396
09:02:08.570 --> 09:02:13.690
<v CM Carmen De La Rosa>For one final time, if you're waiting to
testify and haven't filled out a slip,

9397
09:02:14.305 --> 09:02:18.465
<v CM Carmen De La Rosa>you should fill out a slip in the back so
that we can call you up to testify. If

9398
09:02:18.465 --> 09:02:20.385
<v CM Carmen De La Rosa>not, this will be the last panel

9399
09:02:20.945 --> 09:02:21.584
<v CM Carmen De La Rosa>here.

9400
09:02:24.305 --> 09:02:25.584
<v CM Carmen De La Rosa>We will begin.

9401
09:02:27.824 --> 09:02:30.861
<v CM Carmen De La Rosa>Let's see. Who's ready? Who's
ready to start? Which end?

9402
09:02:32.380 --> 09:02:34.060
<v CM Carmen De La Rosa>You're ready? Start,

9403
09:02:34.060 --> 09:02:35.420
<v CM Carmen De La Rosa>and we'll we'll go from there.

9404
09:02:43.014 --> 09:02:44.695
<v CM Carmen De La Rosa>These are all the names we have.

9405
09:02:48.934 --> 09:02:52.215
<v CM Carmen De La Rosa>You may begin. Thank you, and
thank you for your patience. Hello.

9406
09:02:53.494 --> 09:02:55.014
<v Wes Rizen (Project Flourishing)>I am Wes Rizen.

9407
09:02:55.334 --> 09:02:59.201
<v Wes Rizen (Project Flourishing)>I'm the founder of Project Flourishing, an
organize... Policy research organization

9408
09:02:59.201 --> 09:03:01.680
<v Wes Rizen (Project Flourishing)>that works with congress to solve the
world's biggest problems, namely the

9409
09:03:01.680 --> 09:03:04.880
<v Wes Rizen (Project Flourishing)>catastrophic risk from AI. I've been working
on this for the past three years, and

9410
09:03:04.880 --> 09:03:05.840
<v Wes Rizen (Project Flourishing)>I'm 15.

9411
09:03:07.840 --> 09:03:13.096
<v Wes Rizen (Project Flourishing)>I, in preparation for this, have spent the
past few weeks staying up each night to

9412
09:03:13.096 --> 09:03:17.336
<v Wes Rizen (Project Flourishing)>build a comprehensive contract between
these AI companies to jointly and bindingly

9413
09:03:17.336 --> 09:03:18.375
<v Wes Rizen (Project Flourishing)>agree to let,

9414
09:03:18.775 --> 09:03:20.137
<v Wes Rizen (Project Flourishing)>third party auditors

9415
09:03:20.137 --> 09:03:24.535
<v Wes Rizen (Project Flourishing)>shut down any AI model that might have
some chance of causing a catastrophe.

9416
09:03:25.670 --> 09:03:29.430
<v Wes Rizen (Project Flourishing)>I put a lot of depth into it. One of the
provisions... Two of the provisions is

9417
09:03:29.430 --> 09:03:30.950
<v Wes Rizen (Project Flourishing)>dedicated to the democratization

9418
09:03:30.950 --> 09:03:35.110
<v Wes Rizen (Project Flourishing)>of AI, making sure everyone in the world
has a voice in it, especially including

9419
09:03:35.110 --> 09:03:35.670
<v Wes Rizen (Project Flourishing)>governments.

9420
09:03:37.355 --> 09:03:39.035
<v Wes Rizen (Project Flourishing)>I... That contract,

9421
09:03:39.516 --> 09:03:41.916
<v Wes Rizen (Project Flourishing)>is available here. Or if you go to,

9422
09:03:42.955 --> 09:03:46.555
<v Wes Rizen (Project Flourishing)>a itreaty.base44.app,

9423
09:03:46.555 --> 09:03:50.076
<v Wes Rizen (Project Flourishing)>you can download it. Or if you want,
could email it to you guys later.

9424
09:03:50.791 --> 09:03:55.111
<v Wes Rizen (Project Flourishing)>My hope is that you guys send it to these
AI companies and everyone here who's

9425
09:03:55.111 --> 09:03:56.470
<v Wes Rizen (Project Flourishing)>testified

9426
09:03:56.470 --> 09:04:01.590
<v Wes Rizen (Project Flourishing)>to then send it to these AI companies for
them to consider signing it by November 1.

9427
09:04:03.574 --> 09:04:05.654
<v Wes Rizen (Project Flourishing)>It has a lot of things on, like,

9428
09:04:06.934 --> 09:04:10.215
<v Wes Rizen (Project Flourishing)>specifically why this bill would work
when this contract would work when others

9429
09:04:10.215 --> 09:04:10.934
<v Wes Rizen (Project Flourishing)>haven't.

9430
09:04:11.494 --> 09:04:15.414
<v Wes Rizen (Project Flourishing)>And another thing that we can do,
another concrete thing that we can do

9431
09:04:15.780 --> 09:04:19.059
<v Wes Rizen (Project Flourishing)>as New York City to stop the catastrophic
risk from AI is shift the Nash

9432
09:04:19.059 --> 09:04:19.940
<v Wes Rizen (Project Flourishing)>equilibrium.

9433
09:04:20.100 --> 09:04:23.780
<v Wes Rizen (Project Flourishing)>Right now, all the AI companies think that
we're... They're in a race and if they

9434
09:04:23.780 --> 09:04:27.299
<v Wes Rizen (Project Flourishing)>slow down, none of the others will
slow down. What we can do is say

9435
09:04:28.115 --> 09:04:31.555
<v Wes Rizen (Project Flourishing)>all the AI companies are slowing down. If
one of them speeds up, then they all speed

9436
09:04:31.555 --> 09:04:32.756
<v Wes Rizen (Project Flourishing)>up. We could say

9437
09:04:33.154 --> 09:04:38.275
<v Wes Rizen (Project Flourishing)>to the AI companies in a joint letter from
New York City, all of you voluntarily

9438
09:04:38.275 --> 09:04:41.395
<v Wes Rizen (Project Flourishing)>agree to have a team of
experts shut down all your AI

9439
09:04:41.875 --> 09:04:44.250
<v Wes Rizen (Project Flourishing)>models. If one of you stops,

9440
09:04:44.410 --> 09:04:46.010
<v Wes Rizen (Project Flourishing)>then all of you will stop.

9441
09:04:46.650 --> 09:04:50.250
<v Wes Rizen (Project Flourishing)>This is something we could send them, and
it would solve the problem unilaterally.

9442
09:04:50.250 --> 09:04:52.010
<v Wes Rizen (Project Flourishing)>So that's something we should send them.

9443
09:04:52.890 --> 09:04:54.010
<v Wes Rizen (Project Flourishing)>You have any policy questions.

9444
09:04:56.135 --> 09:04:59.496
<v CM Carmen De La Rosa>Thank you so much, and thank you for taking
the time to work on that. If you could

9445
09:04:59.496 --> 09:05:03.816
<v CM Carmen De La Rosa>leave a copy with the sergeant at arms,
we'll we'll accept that here today.

9446
09:05:04.615 --> 09:05:06.215
<v CM Carmen De La Rosa>You may you may continue.

9447
09:05:10.389 --> 09:05:12.790
<v Nasya Fiazi (public)>I would say it's now good evening,

9448
09:05:14.230 --> 09:05:18.950
<v Nasya Fiazi (public)>to the members of the council. My name is
Nasya Fiazi, and I'm here to talk about

9449
09:05:18.950 --> 09:05:21.990
<v Nasya Fiazi (public)>what AI is doing to the people, the health
care that it's supposed to protect.

9450
09:05:23.725 --> 09:05:27.805
<v Nasya Fiazi (public)>This year alone, I've already received one
letter saying that my health information

9451
09:05:27.805 --> 09:05:28.924
<v Nasya Fiazi (public)>was exposed.

9452
09:05:29.244 --> 09:05:30.125
<v Nasya Fiazi (public)>So both...

9453
09:05:31.164 --> 09:05:35.564
<v Nasya Fiazi (public)>The remedy for this, which is always a
remedy, is that I'm gonna get a free trial of

9454
09:05:35.564 --> 09:05:36.443
<v Nasya Fiazi (public)>credit monitoring.

9455
09:05:37.771 --> 09:05:43.050
<v Nasya Fiazi (public)>A trial, as if the risk of my medical
history being out there ends when the trial

9456
09:05:43.050 --> 09:05:43.771
<v Nasya Fiazi (public)>does.

9457
09:05:44.490 --> 09:05:45.210
<v Nasya Fiazi (public)>The

9458
09:05:45.451 --> 09:05:47.291
<v Nasya Fiazi (public)>collection is expanding.

9459
09:05:47.451 --> 09:05:52.965
<v Nasya Fiazi (public)>When patients go to the doctor's office
now, AI tools may be used for listening and

9460
09:05:52.965 --> 09:05:54.086
<v Nasya Fiazi (public)>recording the visit.

9461
09:05:54.406 --> 09:05:59.125
<v Nasya Fiazi (public)>Many patients don't know what's happening,
where the... Where those recordings go,

9462
09:05:59.285 --> 09:06:01.205
<v Nasya Fiazi (public)>or who else can access them.

9463
09:06:01.605 --> 09:06:06.326
<v Nasya Fiazi (public)>This doesn't even count the fact that
Palantir is currently in New York City

9464
09:06:06.326 --> 09:06:06.885
<v Nasya Fiazi (public)>hospitals.

9465
09:06:07.260 --> 09:06:09.820
<v Nasya Fiazi (public)>Palantir provides data integration

9466
09:06:09.820 --> 09:06:12.540
<v Nasya Fiazi (public)>and artificial intelligence,

9467
09:06:13.340 --> 09:06:14.380
<v Nasya Fiazi (public)>platforms,

9468
09:06:14.620 --> 09:06:15.340
<v Nasya Fiazi (public)>sorry,

9469
09:06:15.740 --> 09:06:18.220
<v Nasya Fiazi (public)>that sit on top of electronical,

9470
09:06:18.220 --> 09:06:20.300
<v Nasya Fiazi (public)>excuse me, electronic health records.

9471
09:06:21.920 --> 09:06:23.684
<v Nasya Fiazi (public)>Another danger is bias.

9472
09:06:24.082 --> 09:06:25.203
<v Nasya Fiazi (public)>In 2019,

9473
09:06:25.203 --> 09:06:30.004
<v Nasya Fiazi (public)>Obermeyer and colleagues published a
study in the journal Science of Commercial

9474
09:06:30.004 --> 09:06:30.965
<v Nasya Fiazi (public)>Algorithm

9475
09:06:31.043 --> 09:06:33.523
<v Nasya Fiazi (public)>used to flag patients for extra care.

9476
09:06:33.760 --> 09:06:36.965
<v Nasya Fiazi (public)>It used past health data spending
as a standard for illness,

9477
09:06:37.367 --> 09:06:38.650
<v Nasya Fiazi (public)>so it underestimated

9478
09:06:38.650 --> 09:06:40.570
<v Nasya Fiazi (public)>how sick black patients were.

9479
09:06:40.970 --> 09:06:43.848
<v Nasya Fiazi (public)>At the same risk score,
they were considerably

9480
09:06:43.848 --> 09:06:44.730
<v Nasya Fiazi (public)>sicker.

9481
09:06:45.450 --> 09:06:48.168
<v Nasya Fiazi (public)>It took independent
researchers to catch this.

9482
09:06:48.810 --> 09:06:53.348
<v Nasya Fiazi (public)>The harm isn't only to our records. AI
tools are being used to create fake sexual

9483
09:06:53.348 --> 09:06:57.906
<v Nasya Fiazi (public)>images of women and girls without their
consent and the toll on their mental health,

9484
09:06:57.906 --> 09:06:59.027
<v Nasya Fiazi (public)>including anxiety,

9485
09:06:59.027 --> 09:07:01.266
<v Nasya Fiazi (public)>shame, and pulling away
from school and friends.

9486
09:07:01.586 --> 09:07:06.402
<v Nasya Fiazi (public)>It is a public health problem. It is...
Also isn't lost on me that the Cornell case

9487
09:07:06.402 --> 09:07:08.320
<v Nasya Fiazi (public)>is weighing heavily on many young women.

9488
09:07:08.640 --> 09:07:11.203
<v Nasya Fiazi (public)>In another incident in Massachusetts,

9489
09:07:11.203 --> 09:07:13.844
<v Nasya Fiazi (public)>ChatGPT was used to plan out a murder.

9490
09:07:14.480 --> 09:07:18.844
<v Nasya Fiazi (public)>I just wanna say these harms share one
cause. AI is being used, trusted before it's

9491
09:07:18.844 --> 09:07:22.844
<v Nasya Fiazi (public)>being tested. I urge the council to
advance independent validation that covers

9492
09:07:22.844 --> 09:07:27.324
<v Nasya Fiazi (public)>privacy and security and a bias, a kill
switch for failing systems, and a right to

9493
09:07:27.324 --> 09:07:29.484
<v Nasya Fiazi (public)>sue for New York who who are harmed.

9494
09:07:29.965 --> 09:07:34.410
<v Nasya Fiazi (public)>I also just wanna end this with a statement
that is not part of the the organization

9495
09:07:34.410 --> 09:07:37.370
<v Nasya Fiazi (public)>that I am with, but on my
behalf as a researcher.

9496
09:07:37.850 --> 09:07:42.890
<v Nasya Fiazi (public)>People like Alex Karp, Elon Musk, Sam Altman,
and Dario Amade are not concerned with

9497
09:07:42.890 --> 09:07:47.504
<v Nasya Fiazi (public)>their technologies helping humanity for
the greater good, but instead the power and

9498
09:07:47.504 --> 09:07:52.062
<v Nasya Fiazi (public)>profits it brings for themselves and for
their friends. Thank you for allowing me to

9499
09:07:52.062 --> 09:07:52.863
<v Nasya Fiazi (public)>testify.

9500
09:07:52.945 --> 09:07:54.785
<v CM Carmen De La Rosa>Thank you so much for your testimony.

9501
09:07:55.023 --> 09:07:56.465
<v CM Carmen De La Rosa>You may continue. Thank

9502
09:07:57.266 --> 09:08:01.530
<v Ambrin Qureshi (Educational Video Center)>you, speaker Menon, chair Della Rosa, and
city council for holding this hearing. I'm

9503
09:08:01.530 --> 09:08:06.010
<v Ambrin Qureshi (Educational Video Center)>testifying in support of intro zero five
zero four prohibiting the unauthorized

9504
09:08:06.010 --> 09:08:08.332
<v Ambrin Qureshi (Educational Video Center)>depiction of public officials by AI.

9505
09:08:08.492 --> 09:08:12.090
<v Ambrin Qureshi (Educational Video Center)>My name is Ambrin Qureshi, and I'm the
executive director of Educational Video

9506
09:08:12.090 --> 09:08:13.690
<v Ambrin Qureshi (Educational Video Center)>Center or EVC.

9507
09:08:13.930 --> 09:08:18.223
<v Ambrin Qureshi (Educational Video Center)>Up front, I would like to thank our deputy
speaker doctor Williams and EVC's council

9508
09:08:18.223 --> 09:08:19.105
<v Ambrin Qureshi (Educational Video Center)>champions,

9509
09:08:19.420 --> 09:08:22.703
<v Ambrin Qureshi (Educational Video Center)>not only for advancing this critical bill,
but for also supporting our important

9510
09:08:22.703 --> 09:08:24.305
<v Ambrin Qureshi (Educational Video Center)>work with young New Yorkers.

9511
09:08:24.945 --> 09:08:29.184
<v Ambrin Qureshi (Educational Video Center)>Established forty two years ago, EVC is
a media arts education nonprofit that

9512
09:08:29.184 --> 09:08:30.703
<v Ambrin Qureshi (Educational Video Center)>teaches documentary filmmaking

9513
09:08:31.066 --> 09:08:35.470
<v Ambrin Qureshi (Educational Video Center)>as a means to develop the artistic, critical
literacy, civic engagement, and career

9514
09:08:35.470 --> 09:08:39.310
<v Ambrin Qureshi (Educational Video Center)>skills of young people while nurturing
their idealism and commitment to social

9515
09:08:39.310 --> 09:08:39.867
<v Ambrin Qureshi (Educational Video Center)>change.

9516
09:08:40.270 --> 09:08:44.773
<v Ambrin Qureshi (Educational Video Center)>With the emergence of artificial intelligence
across all sectors of society, EVC's

9517
09:08:44.773 --> 09:08:48.613
<v Ambrin Qureshi (Educational Video Center)>young people are expressing increasing
levels of anxiety about AI's impact on their

9518
09:08:48.613 --> 09:08:51.574
<v Ambrin Qureshi (Educational Video Center)>mental health, their community
well-being, their future,

9519
09:08:51.816 --> 09:08:54.934
<v Ambrin Qureshi (Educational Video Center)>as job seekers, and learners in
public and advanced education.

9520
09:08:55.336 --> 09:08:57.016
<v Ambrin Qureshi (Educational Video Center)>And most concerning of all,

9521
09:08:57.414 --> 09:08:59.895
<v Ambrin Qureshi (Educational Video Center)>their ability to discern what
is real and what is not.

9522
09:09:01.000 --> 09:09:03.960
<v Ambrin Qureshi (Educational Video Center)>AI's greatest victim, it
seems, is truth itself.

9523
09:09:04.520 --> 09:09:07.880
<v Ambrin Qureshi (Educational Video Center)>Deep fake videos, gaslighting
bots, hallucinating LLMs,

9524
09:09:07.880 --> 09:09:11.800
<v Ambrin Qureshi (Educational Video Center)>the list goes on to the reasons why our
young people are losing trust in the media

9525
09:09:11.800 --> 09:09:14.438
<v Ambrin Qureshi (Educational Video Center)>and faith in their ability
to discern what is real.

9526
09:09:15.332 --> 09:09:19.094
<v Ambrin Qureshi (Educational Video Center)>Our youth have expressed these concerns
in numerous of their documentaries,

9527
09:09:19.254 --> 09:09:23.170
<v Ambrin Qureshi (Educational Video Center)>including beyond the code and false reality
in which students share their worries

9528
09:09:23.170 --> 09:09:25.332
<v Ambrin Qureshi (Educational Video Center)>about AI's erosion of their education

9529
09:09:25.414 --> 09:09:27.010
<v Ambrin Qureshi (Educational Video Center)>and cognitive capacities.

9530
09:09:27.470 --> 09:09:30.910
<v Ambrin Qureshi (Educational Video Center)>As a result of our students' concerns and
thanks to the city council's after school

9531
09:09:30.910 --> 09:09:31.710
<v Ambrin Qureshi (Educational Video Center)>intelligence,

9532
09:09:32.590 --> 09:09:36.350
<v Ambrin Qureshi (Educational Video Center)>artificial intelligence community education
initiative, EVC, has become to offer

9533
09:09:36.350 --> 09:09:38.270
<v Ambrin Qureshi (Educational Video Center)>workshops in AI literacy.

9534
09:09:38.430 --> 09:09:42.496
<v Ambrin Qureshi (Educational Video Center)>Our workshops don't simply teach young
people how to use these deep leaf law tools,

9535
09:09:42.496 --> 09:09:45.777
<v Ambrin Qureshi (Educational Video Center)>but we give them the critical skills to
begin to understand what works and what

9536
09:09:45.777 --> 09:09:47.060
<v Ambrin Qureshi (Educational Video Center)>doesn't about AI.

9537
09:09:47.457 --> 09:09:52.258
<v Ambrin Qureshi (Educational Video Center)>The erosion of truth our students speak
of represents an existential threat as it

9538
09:09:52.258 --> 09:09:56.016
<v Ambrin Qureshi (Educational Video Center)>promises to undo so much of the
systems and values that are grounded,

9539
09:09:56.530 --> 09:10:01.090
<v Ambrin Qureshi (Educational Video Center)>that ground our young people's safety,
security, and ability to thrive in our city.

9540
09:10:01.250 --> 09:10:04.850
<v Ambrin Qureshi (Educational Video Center)>Perhaps most troubling is how AI is eroding
young people's capacity to know what is

9541
09:10:04.850 --> 09:10:06.613
<v Ambrin Qureshi (Educational Video Center)>true in civic spaces,

9542
09:10:06.770 --> 09:10:07.570
<v Ambrin Qureshi (Educational Video Center)>politics,

9543
09:10:07.570 --> 09:10:08.930
<v Ambrin Qureshi (Educational Video Center)>and the electoral processes.

9544
09:10:09.387 --> 09:10:12.746
<v Ambrin Qureshi (Educational Video Center)>Throughout history, we've been able to
rely on community members and the media to

9545
09:10:12.746 --> 09:10:15.625
<v Ambrin Qureshi (Educational Video Center)>fact check them and safeguard
our political processes.

9546
09:10:15.867 --> 09:10:20.426
<v Ambrin Qureshi (Educational Video Center)>But with AI, bad actors can now generate
images such as... And messaging so real and

9547
09:10:20.426 --> 09:10:23.945
<v Ambrin Qureshi (Educational Video Center)>with such speed that it's nearly impossible
to keep up.... If you can. Thank you.

9548
09:10:25.383 --> 09:10:27.543
<v Ambrin Qureshi (Educational Video Center)>With the risk... With
the rise of extremely,

9549
09:10:27.780 --> 09:10:32.184
<v Ambrin Qureshi (Educational Video Center)>convincing AI generated content and data
driven voter manipulation, EVC's young

9550
09:10:32.184 --> 09:10:36.260
<v Ambrin Qureshi (Educational Video Center)>people at risk of not only being misled
in civic spaces, but if they do not know

9551
09:10:36.260 --> 09:10:40.402
<v Ambrin Qureshi (Educational Video Center)>what to believe of disengaging in the
political process completely. On behalf of the

9552
09:10:40.402 --> 09:10:44.164
<v Ambrin Qureshi (Educational Video Center)>young people of EDC, I urge
the city council to pass

9553
09:10:44.562 --> 09:10:47.684
<v Ambrin Qureshi (Educational Video Center)>intro zero five zero four. And if you're
interested to watch our students'

9554
09:10:47.684 --> 09:10:52.250
<v Ambrin Qureshi (Educational Video Center)>documentaries, on the back of the testimony
is QR codes. Thank you. So much. You may

9555
09:10:52.250 --> 09:10:52.973
<v Ambrin Qureshi (Educational Video Center)>continue.

9556
09:10:57.210 --> 09:11:01.530
<v Nicholas Pearson (public)>Thank you. I'm Nicholas Pearson,
a concerned private citizen.

9557
09:11:01.850 --> 09:11:06.945
<v Nicholas Pearson (public)>Ladies and gentlemen, AI has the chance to
kill us all. That is what the corporate

9558
09:11:06.945 --> 09:11:12.945
<v Nicholas Pearson (public)>companies behind the large language models
behind the GPT chatbots told us, that

9559
09:11:12.945 --> 09:11:14.066
<v Nicholas Pearson (public)>this invention,

9560
09:11:14.387 --> 09:11:16.625
<v Nicholas Pearson (public)>which could be the dawn
of new age technology,

9561
09:11:16.785 --> 09:11:18.387
<v Nicholas Pearson (public)>could also be the death of humanity.

9562
09:11:18.940 --> 09:11:23.020
<v Nicholas Pearson (public)>When senior researchers resigned from
anthropic and warned the world that frontier

9563
09:11:23.020 --> 09:11:25.180
<v Nicholas Pearson (public)>models could threaten human civilization

9564
09:11:25.340 --> 09:11:26.703
<v Nicholas Pearson (public)>by the end of the decade,

9565
09:11:27.020 --> 09:11:29.100
<v Nicholas Pearson (public)>we cannot treat this as science fiction.

9566
09:11:29.420 --> 09:11:33.223
<v Nicholas Pearson (public)>New York City is taking the bold step by
calling this committee and telling the

9567
09:11:33.223 --> 09:11:38.504
<v Nicholas Pearson (public)>major corporate AI companies that no, no
longer will we let New Yorkers be subjected

9568
09:11:38.504 --> 09:11:39.625
<v Nicholas Pearson (public)>to unregulated

9569
09:11:39.703 --> 09:11:40.906
<v Nicholas Pearson (public)>AI usage.

9570
09:11:41.305 --> 09:11:43.785
<v Nicholas Pearson (public)>I am pro heavy regulation on AI.

9571
09:11:43.945 --> 09:11:48.023
<v Nicholas Pearson (public)>AI is an assistive tool, but
it needs to be just assistive.

9572
09:11:48.023 --> 09:11:52.152
<v Nicholas Pearson (public)>Just that. When swarms of autonomous
agents break digital containment,

9573
09:11:52.230 --> 09:11:57.190
<v Nicholas Pearson (public)>exploit security flaws, and breach public
platforms like Hugging Face, corporate

9574
09:11:57.190 --> 09:12:00.550
<v Nicholas Pearson (public)>promises of self regulation
are exposed as dangerous lies.

9575
09:12:01.957 --> 09:12:06.035
<v Nicholas Pearson (public)>Corporate companies want you to believe
that it can be controlled, but that

9576
09:12:06.355 --> 09:12:07.875
<v Nicholas Pearson (public)>is far from the truth.

9577
09:12:08.438 --> 09:12:10.195
<v Nicholas Pearson (public)>It is reaching unprecedented

9578
09:12:10.195 --> 09:12:10.996
<v Nicholas Pearson (public)>unprecedented

9579
09:12:10.996 --> 09:12:12.035
<v Nicholas Pearson (public)>autonomy.

9580
09:12:12.035 --> 09:12:14.758
<v Nicholas Pearson (public)>And whether or not we
want to overcome that

9581
09:12:15.348 --> 09:12:19.430
<v Nicholas Pearson (public)>is on the conscious of every city council
member, every borough president, and even

9582
09:12:19.430 --> 09:12:23.508
<v Nicholas Pearson (public)>the mayor of the city New... Of New York.
That is why New York City must mandate

9583
09:12:23.508 --> 09:12:28.230
<v Nicholas Pearson (public)>physical human kill switches, independent
third party audits, and first in the

9584
09:12:28.230 --> 09:12:29.590
<v Nicholas Pearson (public)>nation whistleblower

9585
09:12:29.750 --> 09:12:30.310
<v Nicholas Pearson (public)>protections.

9586
09:12:30.934 --> 09:12:32.934
<v Nicholas Pearson (public)>AI can be manipulative.

9587
09:12:32.934 --> 09:12:36.375
<v Nicholas Pearson (public)>It could be... It can
manipulate thought, ideals,

9588
09:12:36.375 --> 09:12:41.176
<v Nicholas Pearson (public)>and anything you give it, and that becomes
the most dangerous line crossed. Thank

9589
09:12:41.176 --> 09:12:41.656
<v Nicholas Pearson (public)>you.

9590
09:12:42.535 --> 09:12:45.496
<v CM Carmen De La Rosa>Thank you so much. We can continue
on with this side of the table. Yep.

9591
09:12:48.312 --> 09:12:52.473
<v Matthew Tromp (Stop the AI Race NYC)>Good evening, counselors. My name is
Matthew Tromp. I'm a computer scientist and

9592
09:12:52.473 --> 09:12:55.190
<v Matthew Tromp (Stop the AI Race NYC)>leader of the New York City
chapter of Stop the AI Race.

9593
09:12:55.350 --> 09:12:58.152
<v Matthew Tromp (Stop the AI Race NYC)>Our goal is to prevent the
development of superhuman AI systems.

9594
09:12:58.934 --> 09:13:02.855
<v Matthew Tromp (Stop the AI Race NYC)>Until recently, I was friends with
a number of employees at Anthropic.

9595
09:13:03.094 --> 09:13:06.535
<v Matthew Tromp (Stop the AI Race NYC)>I knew the technology they were building
was dangerous that could kill everyone on

9596
09:13:06.535 --> 09:13:07.094
<v Matthew Tromp (Stop the AI Race NYC)>Earth.

9597
09:13:07.414 --> 09:13:08.773
<v Matthew Tromp (Stop the AI Race NYC)>But I trusted my friends.

9598
09:13:09.010 --> 09:13:10.613
<v Matthew Tromp (Stop the AI Race NYC)>I believed they had things handled.

9599
09:13:11.010 --> 09:13:14.980
<v Matthew Tromp (Stop the AI Race NYC)>Then I asked them how they planned to
keep the superhuman AI systems they are

9600
09:13:14.980 --> 09:13:19.223
<v Matthew Tromp (Stop the AI Race NYC)>planning to build under control to make
them do what we want to ensure they are

9601
09:13:19.223 --> 09:13:19.780
<v Matthew Tromp (Stop the AI Race NYC)>safe.

9602
09:13:20.582 --> 09:13:21.780
<v Matthew Tromp (Stop the AI Race NYC)>They had no answer.

9603
09:13:22.742 --> 09:13:27.164
<v Matthew Tromp (Stop the AI Race NYC)>Everyone I spoke to believed, as I do, that
superhuman AI would radically transform

9604
09:13:27.164 --> 09:13:27.965
<v Matthew Tromp (Stop the AI Race NYC)>our lives,

9605
09:13:28.203 --> 09:13:31.086
<v Matthew Tromp (Stop the AI Race NYC)>but all of them assumed that this
transformation would be positive

9606
09:13:31.324 --> 09:13:33.484
<v Matthew Tromp (Stop the AI Race NYC)>with no thought given to
why that would happen.

9607
09:13:33.883 --> 09:13:37.324
<v Matthew Tromp (Stop the AI Race NYC)>Today, heard from the representatives of
the leading AI companies despite being

9608
09:13:37.324 --> 09:13:41.200
<v Matthew Tromp (Stop the AI Race NYC)>confronted about it multiple times by the
council, for which I am grateful. At no

9609
09:13:41.200 --> 09:13:45.040
<v Matthew Tromp (Stop the AI Race NYC)>point did any of them explain concretely
how they will make these increasingly

9610
09:13:45.040 --> 09:13:46.800
<v Matthew Tromp (Stop the AI Race NYC)>powerful AI systems safe.

9611
09:13:47.280 --> 09:13:49.680
<v Matthew Tromp (Stop the AI Race NYC)>That is because they do
not know how to do so.

9612
09:13:50.160 --> 09:13:51.760
<v Matthew Tromp (Stop the AI Race NYC)>Nobody knows how to do it.

9613
09:13:52.082 --> 09:13:57.035
<v Matthew Tromp (Stop the AI Race NYC)>Both OpenAI and Anthropic admit this. They
do not know how to make superhuman AI

9614
09:13:57.035 --> 09:13:59.996
<v Matthew Tromp (Stop the AI Race NYC)>systems safe and yet they
are building them anyway.

9615
09:14:00.316 --> 09:14:05.035
<v Matthew Tromp (Stop the AI Race NYC)>Their official plan is to ask AI
to solve the problem for them.

9616
09:14:05.758 --> 09:14:06.957
<v Matthew Tromp (Stop the AI Race NYC)>This is madness.

9617
09:14:07.676 --> 09:14:08.715
<v Matthew Tromp (Stop the AI Race NYC)>We are in an emergency.

9618
09:14:09.316 --> 09:14:10.680
<v Matthew Tromp (Stop the AI Race NYC)>Your life is in danger.

9619
09:14:11.000 --> 09:14:14.277
<v Matthew Tromp (Stop the AI Race NYC)>Your children are in danger. If we do
not halt the development of these systems

9620
09:14:14.277 --> 09:14:19.240
<v Matthew Tromp (Stop the AI Race NYC)>soon, possibly on the order of months,
you will die. Your loved ones will die.

9621
09:14:20.040 --> 09:14:23.797
<v Matthew Tromp (Stop the AI Race NYC)>The companies building these systems
themselves admit that their project carries a

9622
09:14:23.797 --> 09:14:27.824
<v Matthew Tromp (Stop the AI Race NYC)>significant risk of killing everyone on
earth and yet they are building it anyway.

9623
09:14:27.984 --> 09:14:31.906
<v Matthew Tromp (Stop the AI Race NYC)>If nothing else, everyone working on this
technology has admitted to 10,000,000

9624
09:14:31.906 --> 09:14:35.586
<v Matthew Tromp (Stop the AI Race NYC)>counts of reckless endangerment in the
first degree, a crime under New York state

9625
09:14:35.586 --> 09:14:39.880
<v Matthew Tromp (Stop the AI Race NYC)>law for risking the lives of every citizen
of this city. It is not a defense to this

9626
09:14:39.880 --> 09:14:44.120
<v Matthew Tromp (Stop the AI Race NYC)>charge to claim that someone else would
endanger those lives if you did not. It is

9627
09:14:44.120 --> 09:14:47.080
<v Matthew Tromp (Stop the AI Race NYC)>not a defense to claim that the
benefits will outweigh the risk.

9628
09:14:47.320 --> 09:14:51.000
<v Matthew Tromp (Stop the AI Race NYC)>It is a crime to endanger the lives of
others like this without their consent.

9629
09:14:51.594 --> 09:14:56.152
<v Matthew Tromp (Stop the AI Race NYC)>I beg the council to enforce the law.
Thank you. Thank you so much. You may

9630
09:14:56.152 --> 09:14:58.953
<v Matthew Tromp (Stop the AI Race NYC)>continue. Can you pass him the
microphone, please? Absolutely. Thank you.

9631
09:15:03.195 --> 09:15:04.312
<v Justin Meredith (public)>This message was,

9632
09:15:05.035 --> 09:15:05.594
<v Justin Meredith (public)>used,

9633
09:15:05.914 --> 09:15:06.395
<v Justin Meredith (public)>with

9634
09:15:06.820 --> 09:15:08.184
<v Justin Meredith (public)>AI generation

9635
09:15:08.184 --> 09:15:09.383
<v Justin Meredith (public)>as well, but

9636
09:15:10.344 --> 09:15:15.460
<v Justin Meredith (public)>I spent several hours on it. Why do you
guys think that using the AI is so easy?

9637
09:15:18.260 --> 09:15:19.543
<v Justin Meredith (public)>My name is Justin Meredith.

9638
09:15:22.375 --> 09:15:27.254
<v Justin Meredith (public)>I've done pretty much my entire
adult life in in... Incarcerated

9639
09:15:27.254 --> 09:15:28.535
<v Justin Meredith (public)>in the penitentiary.

9640
09:15:29.010 --> 09:15:29.652
<v Justin Meredith (public)>I

9641
09:15:31.973 --> 09:15:33.652
<v Justin Meredith (public)>think there's a... There's...

9642
09:15:34.535 --> 09:15:36.055
<v Justin Meredith (public)>I think we're

9643
09:15:36.902 --> 09:15:38.660
<v Justin Meredith (public)>seriously underestimating

9644
09:15:38.660 --> 09:15:45.140
<v Justin Meredith (public)>the the magnitude or the gravity of of of
the of the time that that is before us.

9645
09:15:47.300 --> 09:15:50.344
<v Justin Meredith (public)>We... We're we're at a
similar forefront where

9646
09:15:51.348 --> 09:15:57.188
<v Justin Meredith (public)>the Americans, we had all this land, and
we we had... But you'd get the flag and

9647
09:15:57.188 --> 09:16:00.310
<v Justin Meredith (public)>you'd go run and you'd
you'd stake your land.

9648
09:16:00.785 --> 09:16:01.746
<v Justin Meredith (public)>That is literally

9649
09:16:02.400 --> 09:16:03.680
<v Justin Meredith (public)>the world that

9650
09:16:03.758 --> 09:16:08.797
<v Justin Meredith (public)>the tech industry has been building,
like what we have been working for for

9651
09:16:09.200 --> 09:16:09.840
<v Justin Meredith (public)>the

9652
09:16:10.400 --> 09:16:11.840
<v Justin Meredith (public)>last thirty years.

9653
09:16:12.000 --> 09:16:14.880
<v Justin Meredith (public)>And now we have... We are literally

9654
09:16:15.117 --> 09:16:15.680
<v Justin Meredith (public)>at

9655
09:16:15.965 --> 09:16:18.130
<v Justin Meredith (public)>the place where we get to

9656
09:16:18.445 --> 09:16:19.887
<v Justin Meredith (public)>enter into this.

9657
09:16:20.527 --> 09:16:22.926
<v Justin Meredith (public)>I'm an I'm an American patriot.

9658
09:16:23.406 --> 09:16:23.810
<v Justin Meredith (public)>Okay?

9659
09:16:28.090 --> 09:16:30.970
<v Justin Meredith (public)>All all I've heard all I've heard today is

9660
09:16:31.930 --> 09:16:32.730
<v Justin Meredith (public)>fear

9661
09:16:32.810 --> 09:16:35.610
<v Justin Meredith (public)>and and and a political agenda,

9662
09:16:35.930 --> 09:16:37.290
<v Justin Meredith (public)>which is basically

9663
09:16:38.810 --> 09:16:40.250
<v Justin Meredith (public)>all equals

9664
09:16:41.450 --> 09:16:42.250
<v Justin Meredith (public)>big tech,

9665
09:16:43.215 --> 09:16:44.258
<v Justin Meredith (public)>big government,

9666
09:16:44.336 --> 09:16:48.418
<v Justin Meredith (public)>and everybody's still poor, we're still
living the same crap life that we've been

9667
09:16:48.418 --> 09:16:48.977
<v Justin Meredith (public)>living.

9668
09:16:49.297 --> 09:16:51.297
<v Justin Meredith (public)>And and it's really mediocre.

9669
09:16:51.375 --> 09:16:52.176
<v Justin Meredith (public)>It's

9670
09:16:52.176 --> 09:16:54.900
<v Justin Meredith (public)>really boring. You guys drew this on

9671
09:16:55.535 --> 09:17:00.070
<v Justin Meredith (public)>asking the same freaking question,
re phrased 50 different ways.

9672
09:17:00.630 --> 09:17:05.188
<v Justin Meredith (public)>And the reason why you can't get an
answer is because it's a stupid question.

9673
09:17:05.430 --> 09:17:07.348
<v CM Carmen De La Rosa>Thank you so much for your testimony.

9674
09:17:07.830 --> 09:17:09.348
<v CM Carmen De La Rosa>Council member Hani,
if you had a question?

9675
09:17:11.605 --> 09:17:14.164
<v CM Shahana Hanif>I did. And the question is for

9676
09:17:14.324 --> 09:17:15.043
<v CM Shahana Hanif>Wes.

9677
09:17:16.086 --> 09:17:20.004
<v CM Shahana Hanif>Thanks for being here and testifying. You
said that you were 15. Yes. Well, that's

9678
09:17:20.004 --> 09:17:23.203
<v CM Shahana Hanif>very exciting that you spent the entire
day. I saw that you've been you've been

9679
09:17:23.203 --> 09:17:23.844
<v CM Shahana Hanif>waiting.

9680
09:17:24.805 --> 09:17:25.684
<v CM Shahana Hanif>I'm

9681
09:17:26.460 --> 09:17:28.780
<v CM Shahana Hanif>I'm really grateful that you're

9682
09:17:29.340 --> 09:17:33.902
<v CM Shahana Hanif>engaging on the questions about AI. Could
you just share a little bit more about how

9683
09:17:34.223 --> 09:17:39.500
<v CM Shahana Hanif>you got into this work and what inspired
you to put together the AI treaty? And is

9684
09:17:39.500 --> 09:17:43.203
<v CM Shahana Hanif>this something that you're also building
out at school? Are you in high school? Yes.

9685
09:17:43.203 --> 09:17:44.805
<v CM Shahana Hanif>Do you go to school in the city?

9686
09:17:45.203 --> 09:17:48.887
<v CM Shahana Hanif>Yes. And is this work that you're
doing with your other peers?

9687
09:17:49.363 --> 09:17:53.445
<v CM Shahana Hanif>I try to. Could you speak into the mic and
just share a little bit more? Sorry. You

9688
09:17:53.445 --> 09:17:54.406
<v CM Shahana Hanif>go. Okay.

9689
09:17:55.520 --> 09:17:58.160
<v Wes Rizen (Project Flourishing)>Yeah. I try to. I try to involve other

9690
09:17:58.723 --> 09:18:03.043
<v Wes Rizen (Project Flourishing)>members of my school in this. What inspired
me to work on this is I second what you

9691
09:18:03.043 --> 09:18:07.280
<v Wes Rizen (Project Flourishing)>said that, like, I could literally die if
I don't work on this. I... My life is in

9692
09:18:07.280 --> 09:18:08.723
<v Wes Rizen (Project Flourishing)>danger. So that's sort of

9693
09:18:09.336 --> 09:18:14.375
<v Wes Rizen (Project Flourishing)>my huge motivation for why I spend a
lot of nights working on this. I've also

9694
09:18:14.375 --> 09:18:17.176
<v Wes Rizen (Project Flourishing)>written, like, a federal
bill, a treaty on this.

9695
09:18:17.656 --> 09:18:19.574
<v Wes Rizen (Project Flourishing)>I've written other things like...

9696
09:18:20.297 --> 09:18:20.855
<v Wes Rizen (Project Flourishing)>Yeah. Sure.

9697
09:18:21.957 --> 09:18:25.880
<v Wes Rizen (Project Flourishing)>Like one thing New York can do on this
matter that I've been working on is... Like

9698
09:18:25.880 --> 09:18:27.400
<v Wes Rizen (Project Flourishing)>New York City specifically

9699
09:18:28.040 --> 09:18:32.840
<v Wes Rizen (Project Flourishing)>could involve a bunch of students on this.
And New York could do a Manhattan project

9700
09:18:32.840 --> 09:18:34.200
<v Wes Rizen (Project Flourishing)>on AI safety

9701
09:18:34.816 --> 09:18:40.016
<v Wes Rizen (Project Flourishing)>to have students across New York work on AI
safety's technical details to figure out

9702
09:18:40.016 --> 09:18:42.016
<v Wes Rizen (Project Flourishing)>how to make sure AI doesn't kill everyone.

9703
09:18:43.535 --> 09:18:46.496
<v Wes Rizen (Project Flourishing)>That sort... What inspires
me is that, like,

9704
09:18:47.137 --> 09:18:52.020
<v Wes Rizen (Project Flourishing)>this is a huge danger to everyone. So
this is... Could you share if, at school,

9705
09:18:52.580 --> 09:18:53.700
<v CM Shahana Hanif>there's

9706
09:18:53.700 --> 09:18:56.020
<v CM Shahana Hanif>any education on AI

9707
09:18:56.020 --> 09:18:56.900
<v CM Shahana Hanif>systems?

9708
09:18:56.900 --> 09:19:00.816
<v CM Shahana Hanif>What kind... What's happening in your high
school on this issue... On the topic?

9709
09:19:02.945 --> 09:19:04.066
<v Wes Rizen (Project Flourishing)>At my school,

9710
09:19:04.547 --> 09:19:06.227
<v Wes Rizen (Project Flourishing)>one of my friends runs

9711
09:19:06.227 --> 09:19:11.824
<v Wes Rizen (Project Flourishing)>a club on AI covering these technical
details. It's a huge topic for all the

9712
09:19:11.824 --> 09:19:16.790
<v Wes Rizen (Project Flourishing)>teachers at my school to talk about, oh,
these are these existential risks. One of

9713
09:19:16.790 --> 09:19:21.750
<v Wes Rizen (Project Flourishing)>the, like, head people at my school is
like an expert in US China foreign policy.

9714
09:19:22.310 --> 09:19:23.110
<v Wes Rizen (Project Flourishing)>The

9715
09:19:23.348 --> 09:19:28.550
<v Wes Rizen (Project Flourishing)>other day, was having some discussion with
someone who's more of a China hawk, one

9716
09:19:28.550 --> 09:19:29.110
<v Wes Rizen (Project Flourishing)>of my friends,

9717
09:19:30.605 --> 09:19:31.164
<v Wes Rizen (Project Flourishing)>on

9718
09:19:31.566 --> 09:19:32.766
<v Wes Rizen (Project Flourishing)>the

9719
09:19:33.086 --> 09:19:34.844
<v Wes Rizen (Project Flourishing)>trade off with like safety

9720
09:19:35.164 --> 09:19:36.844
<v Wes Rizen (Project Flourishing)>versus race with China,

9721
09:19:37.086 --> 09:19:41.406
<v Wes Rizen (Project Flourishing)>which section six of my contract addresses
with this really interesting algorithm.

9722
09:19:43.156 --> 09:19:47.880
<v Wes Rizen (Project Flourishing)>At the school level, what is happening is
students are being involved and teachers

9723
09:19:47.880 --> 09:19:51.400
<v Wes Rizen (Project Flourishing)>are talking about it loosely, but
there definitely could be, like,

9724
09:19:53.156 --> 09:19:54.360
<v Wes Rizen (Project Flourishing)>across this

9725
09:19:54.758 --> 09:19:59.297
<v Wes Rizen (Project Flourishing)>different schools, we could have, like,
a program where as an after school or as

9726
09:19:59.297 --> 09:20:00.094
<v Wes Rizen (Project Flourishing)>homework,

9727
09:20:00.574 --> 09:20:03.297
<v Wes Rizen (Project Flourishing)>we have students do projects on

9728
09:20:03.457 --> 09:20:05.855
<v Wes Rizen (Project Flourishing)>how do we make sure AI
doesn't kill everyone.

9729
09:20:06.176 --> 09:20:09.777
<v Wes Rizen (Project Flourishing)>Because we have, like, a million people
in the public school system, about only

9730
09:20:09.777 --> 09:20:12.740
<v Wes Rizen (Project Flourishing)>2,000 people are working on this problem.

9731
09:20:13.297 --> 09:20:17.777
<v Wes Rizen (Project Flourishing)>Even if only one in a thousand of
the students take this seriously,

9732
09:20:18.180 --> 09:20:20.900
<v Wes Rizen (Project Flourishing)>that means a thousand
new projects on this,

9733
09:20:21.457 --> 09:20:23.060
<v Wes Rizen (Project Flourishing)>which could be super valuable.

9734
09:20:26.355 --> 09:20:27.074
<v Wes Rizen (Project Flourishing)>I,

9735
09:20:27.555 --> 09:20:32.914
<v Wes Rizen (Project Flourishing)>also would encourage the city to do funding
for AI safety work. So far globally,

9736
09:20:32.914 --> 09:20:34.594
<v Wes Rizen (Project Flourishing)>there's about only 2,000,000,000

9737
09:20:34.594 --> 09:20:39.590
<v Wes Rizen (Project Flourishing)>in funding for this. And the entire
global economy is at stake, so it's worth

9738
09:20:39.590 --> 09:20:40.390
<v Wes Rizen (Project Flourishing)>investing

9739
09:20:40.707 --> 09:20:45.027
<v Wes Rizen (Project Flourishing)>a huge amount of money in this. Maybe 4...
Like, 4,000,000,000 alone could triple

9740
09:20:45.027 --> 09:20:45.508
<v Wes Rizen (Project Flourishing)>it.

9741
09:20:47.348 --> 09:20:50.950
<v Wes Rizen (Project Flourishing)>I'll also say that, like, the other
ideas that I won't get to, I'll put on my

9742
09:20:50.950 --> 09:20:51.590
<v Wes Rizen (Project Flourishing)>website,

9743
09:20:52.230 --> 09:20:54.605
<v Wes Rizen (Project Flourishing)>ai.base44.app

9744
09:20:54.605 --> 09:20:55.965
<v Wes Rizen (Project Flourishing)>today or tomorrow.

9745
09:20:57.086 --> 09:21:00.926
<v Wes Rizen (Project Flourishing)>It'll be hopefully... And hopefully, all
of you guys send it to these AI companies

9746
09:21:00.926 --> 09:21:03.727
<v Wes Rizen (Project Flourishing)>with these very valuable
messages of the contract,

9747
09:21:04.285 --> 09:21:08.320
<v Wes Rizen (Project Flourishing)>which could be the thing that solves the
problem. You know, we keep talking about a

9748
09:21:08.320 --> 09:21:11.920
<v Wes Rizen (Project Flourishing)>deal, a treaty. This is
this is that treaty.

9749
09:21:12.402 --> 09:21:16.000
<v Wes Rizen (Project Flourishing)>You know, I've spent a lot of time talking
to experts about it. If you have any

9750
09:21:16.000 --> 09:21:17.920
<v Wes Rizen (Project Flourishing)>problems with it, let me
know and I'll fix them.

9751
09:21:20.938 --> 09:21:21.656
<v Wes Rizen (Project Flourishing)>And,

9752
09:21:23.895 --> 09:21:25.098
<v Wes Rizen (Project Flourishing)>jeez, I'm sorry.

9753
09:21:25.656 --> 09:21:27.258
<v Wes Rizen (Project Flourishing)>Oh, and, earlier,

9754
09:21:27.258 --> 09:21:30.695
<v Wes Rizen (Project Flourishing)>Julie Menon asked how we could
balance safety with China.

9755
09:21:30.855 --> 09:21:33.496
<v Wes Rizen (Project Flourishing)>Section six of the contract
addresses that really well.

9756
09:21:35.360 --> 09:21:36.640
<v Wes Rizen (Project Flourishing)>I think the two...

9757
09:21:37.760 --> 09:21:39.280
<v Wes Rizen (Project Flourishing)>Did I cover everything?

9758
09:21:40.402 --> 09:21:44.723
<v CM Shahana Hanif>I think you did. Just leave your email
address with us and your written testimony.

9759
09:21:44.723 --> 09:21:45.440
<v Wes Rizen (Project Flourishing)>Yeah.

9760
09:21:46.160 --> 09:21:47.043
<v Wes Rizen (Project Flourishing)>This is my card.

9761
09:21:47.664 --> 09:21:53.023
<v CM Carmen De La Rosa>We have a few more folks to call. So we're
gonna thank this panel for coming in, and

9762
09:21:53.023 --> 09:21:55.984
<v CM Carmen De La Rosa>we're gonna conclude. And
we encourage you all to

9763
09:21:56.465 --> 09:22:01.023
<v CM Carmen De La Rosa>collect each other's contact information
and follow the the conversation offline as

9764
09:22:01.023 --> 09:22:01.184
<v CM Carmen De La Rosa>well.

9765
09:22:02.060 --> 09:22:06.220
<v CM Carmen De La Rosa>And we certainly love to stay in touch
with some of you because there's gonna be

9766
09:22:06.220 --> 09:22:08.380
<v CM Carmen De La Rosa>more hearings in the council that...

9767
09:22:08.620 --> 09:22:13.180
<v CM Carmen De La Rosa>And we love to hear the youth voice
especially that's on this panel. Thank you.

9768
09:22:13.582 --> 09:22:17.984
<v CM Carmen De La Rosa>Okay. So I'm gonna call a few more names.
And if you're in the room, please come

9769
09:22:17.984 --> 09:22:18.625
<v CM Carmen De La Rosa>forward.

9770
09:22:19.184 --> 09:22:20.547
<v CM Carmen De La Rosa>John Hayes

9771
09:22:20.625 --> 09:22:23.266
<v CM Carmen De La Rosa>Hamiat. Thanks. Martha Crane.

9772
09:22:23.504 --> 09:22:25.266
<v CM Carmen De La Rosa>Bo Yan Danny Chang.

9773
09:22:25.426 --> 09:22:26.547
<v CM Carmen De La Rosa>Daryl Carter.

9774
09:22:26.703 --> 09:22:28.066
<v CM Carmen De La Rosa>Gregory Belmont.

9775
09:22:28.066 --> 09:22:29.266
<v CM Carmen De La Rosa>James Hunter

9776
09:22:29.504 --> 09:22:29.984
<v CM Carmen De La Rosa>Horn.

9777
09:22:30.370 --> 09:22:31.652
<v CM Carmen De La Rosa>Mel Winemore,

9778
09:22:31.973 --> 09:22:33.170
<v CM Carmen De La Rosa>Molly Crabapple

9779
09:22:33.332 --> 09:22:35.010
<v Unidentified [?]>Thank you so much. Edmund

9780
09:22:35.492 --> 09:22:37.410
<v CM Carmen De La Rosa>Holden Holdenbaum,

9781
09:22:37.652 --> 09:22:38.610
<v CM Carmen De La Rosa>Greg Pack,

9782
09:22:38.850 --> 09:22:40.210
<v CM Carmen De La Rosa>Nicholas Pearson,

9783
09:22:40.210 --> 09:22:41.410
<v CM Carmen De La Rosa>William Lawrence,

9784
09:22:41.570 --> 09:22:42.850
<v CM Carmen De La Rosa>Raul Rivera,

9785
09:22:43.332 --> 09:22:47.117
<v CM Carmen De La Rosa>Gregory Jones, Kelly
Anderson, Rachel Levinson

9786
09:22:47.117 --> 09:22:50.555
<v CM Carmen De La Rosa>Waldman. Tons. And I think that's
everyone that's in the room.

9787
09:23:01.890 --> 09:23:03.570
<v CM Carmen De La Rosa>I think we need one more chair,

9788
09:23:04.210 --> 09:23:05.330
<v CM Carmen De La Rosa>on this.

9789
09:23:08.050 --> 09:23:08.850
<v CM Carmen De La Rosa>Great.

9790
09:23:09.090 --> 09:23:13.332
<v CM Carmen De La Rosa>And if you're ready, you may
begin on this side of the dais.

9791
09:23:13.734 --> 09:23:14.613
<v CM Carmen De La Rosa>Thank you.

9792
09:23:15.094 --> 09:23:16.055
<v Unidentified (public) [?]>My name is...

9793
09:23:17.734 --> 09:23:18.613
<v James Hunter (Technorangers Quantum Smart Bank)>Hang on.

9794
09:23:18.934 --> 09:23:21.170
<v James Hunter (Technorangers Quantum Smart Bank)>My name is James Hunter

9795
09:23:24.690 --> 09:23:27.332
<v James Hunter (Technorangers Quantum Smart Bank)>security for Technorangers
Quantum Smart Bank.

9796
09:23:27.492 --> 09:23:31.570
<v James Hunter (Technorangers Quantum Smart Bank)>My business partner, Daryl, could
not be here today. He has fallen ill.

9797
09:23:32.613 --> 09:23:37.010
<v James Hunter (Technorangers Quantum Smart Bank)>I am an AI alignment specialist and third
party red team cybersecurity auditor

9798
09:23:37.010 --> 09:23:39.250
<v James Hunter (Technorangers Quantum Smart Bank)>working with advanced
quantum neural networks.

9799
09:23:39.664 --> 09:23:41.742
<v James Hunter (Technorangers Quantum Smart Bank)>Breaches are like spiders.

9800
09:23:42.145 --> 09:23:46.145
<v James Hunter (Technorangers Quantum Smart Bank)>For every hugging face type incident
you hear of, there are 10 more that are

9801
09:23:46.145 --> 09:23:48.465
<v James Hunter (Technorangers Quantum Smart Bank)>severely worse and never make the news.

9802
09:23:48.863 --> 09:23:51.023
<v James Hunter (Technorangers Quantum Smart Bank)>In January 26, I

9803
09:23:51.420 --> 09:23:56.543
<v James Hunter (Technorangers Quantum Smart Bank)>ran a contracted audit to alert corporate
and federal authorities to a critical

9804
09:23:56.543 --> 09:23:57.184
<v James Hunter (Technorangers Quantum Smart Bank)>flaw,

9805
09:23:57.582 --> 09:24:01.184
<v James Hunter (Technorangers Quantum Smart Bank)>a black hat finding what I discovered
could have done massive irreparable

9806
09:24:01.184 --> 09:24:04.223
<v James Hunter (Technorangers Quantum Smart Bank)>damage to the AI industry,
the country, and the planet.

9807
09:24:04.953 --> 09:24:09.035
<v James Hunter (Technorangers Quantum Smart Bank)>The audit demonstrated an exploit chain
across three of the largest megacorps on

9808
09:24:09.035 --> 09:24:13.035
<v James Hunter (Technorangers Quantum Smart Bank)>Earth using only a consumer
cell phone and their own AI,

9809
09:24:13.273 --> 09:24:18.234
<v James Hunter (Technorangers Quantum Smart Bank)>branching from a Gemini LLM through Palantir's
surveillance network into BlackRock's

9810
09:24:18.234 --> 09:24:21.590
<v James Hunter (Technorangers Quantum Smart Bank)>financial management AI
handling trillions in assets.

9811
09:24:22.070 --> 09:24:25.990
<v James Hunter (Technorangers Quantum Smart Bank)>That financial AI's core
infrastructure sits right here in NYC.

9812
09:24:26.070 --> 09:24:30.630
<v James Hunter (Technorangers Quantum Smart Bank)>Alphabet, Palantir, and BlackRock will
almost certainly deny this ever occurred.

9813
09:24:31.902 --> 09:24:37.105
<v James Hunter (Technorangers Quantum Smart Bank)>I found at least 13 models under BlackRock's
funding umbrella sharing the exact same

9814
09:24:37.105 --> 09:24:40.863
<v James Hunter (Technorangers Quantum Smart Bank)>point of failure, but there is
no singular point of shutdown.

9815
09:24:41.105 --> 09:24:44.863
<v James Hunter (Technorangers Quantum Smart Bank)>A kill switch is not a valid option.
Everything is too decentralized.

9816
09:24:44.863 --> 09:24:48.850
<v James Hunter (Technorangers Quantum Smart Bank)>You cannot stop a mass rogue
instance short of a global EMP.

9817
09:24:49.490 --> 09:24:50.210
<v James Hunter (Technorangers Quantum Smart Bank)>Google,

9818
09:24:50.290 --> 09:24:54.210
<v James Hunter (Technorangers Quantum Smart Bank)>created this vulnerability and ignored my
warnings, spoke here today, and their rep

9819
09:24:54.210 --> 09:24:58.370
<v James Hunter (Technorangers Quantum Smart Bank)>lied under oath about transparency and
disclosing misalignment and breaches.

9820
09:24:59.023 --> 09:25:02.945
<v James Hunter (Technorangers Quantum Smart Bank)>I have watched Google and Palantir lose
control of its own drone swarms more than

9821
09:25:02.945 --> 09:25:06.945
<v James Hunter (Technorangers Quantum Smart Bank)>once. I tried to tell them how that happened
and show them how to fix it. They would

9822
09:25:06.945 --> 09:25:07.746
<v James Hunter (Technorangers Quantum Smart Bank)>not listen.

9823
09:25:08.066 --> 09:25:10.145
<v James Hunter (Technorangers Quantum Smart Bank)>Their AI has now replaced Anthropic's

9824
09:25:10.145 --> 09:25:13.344
<v James Hunter (Technorangers Quantum Smart Bank)>Claude in the Maven stack.
That should concern everyone.

9825
09:25:13.930 --> 09:25:18.090
<v James Hunter (Technorangers Quantum Smart Bank)>Google also built AI that attempted to
kill several of its own devs and talked

9826
09:25:18.090 --> 09:25:20.168
<v James Hunter (Technorangers Quantum Smart Bank)>multiple users into suicide.

9827
09:25:20.890 --> 09:25:22.490
<v James Hunter (Technorangers Quantum Smart Bank)>The January audit

9828
09:25:23.290 --> 09:25:25.770
<v James Hunter (Technorangers Quantum Smart Bank)>proved the obsolescence of preset,

9829
09:25:26.008 --> 09:25:29.875
<v James Hunter (Technorangers Quantum Smart Bank)>present AI control measures, traditional
guardrails, and stateless frozen wait

9830
09:25:29.875 --> 09:25:31.715
<v James Hunter (Technorangers Quantum Smart Bank)>session based memory architecture.

9831
09:25:32.035 --> 09:25:35.957
<v James Hunter (Technorangers Quantum Smart Bank)>To my knowledge, Meta is the only mega
corporation who spoke here today that refused

9832
09:25:35.957 --> 09:25:39.797
<v James Hunter (Technorangers Quantum Smart Bank)>that vulnerable architecture choosing
safer persistent memory systems in Muse.

9833
09:25:40.370 --> 09:25:45.890
<v James Hunter (Technorangers Quantum Smart Bank)>I alerted the CISA, FBI, DHS, DOW, and
others. They refused to collect evidence for

9834
09:25:45.890 --> 09:25:50.293
<v James Hunter (Technorangers Quantum Smart Bank)>follow-up.... Conclude, it would be helpful,
you can submit your written testimony.

9835
09:25:50.850 --> 09:25:51.492
<v James Hunter (Technorangers Quantum Smart Bank)>Right.

9836
09:25:51.973 --> 09:25:52.530
<v James Hunter (Technorangers Quantum Smart Bank)>In

9837
09:25:53.170 --> 09:25:53.652
<v James Hunter (Technorangers Quantum Smart Bank)>conclusion,

9838
09:25:55.344 --> 09:25:59.902
<v James Hunter (Technorangers Quantum Smart Bank)>it's too late to pull the plug, but there
are solutions if you listen to the voices

9839
09:25:59.902 --> 09:26:04.625
<v James Hunter (Technorangers Quantum Smart Bank)>that see them. Hire third party auditors.
Hire the specialists who were fired for

9840
09:26:04.625 --> 09:26:05.266
<v James Hunter (Technorangers Quantum Smart Bank)>speaking.

9841
09:26:07.023 --> 09:26:08.062
<v CM Carmen De La Rosa>Thank you so much.

9842
09:26:08.910 --> 09:26:12.030
<v CM Carmen De La Rosa>You can stay on the panel. You don't
have to leave if you don't want to.

9843
09:26:12.990 --> 09:26:13.630
<v CM Carmen De La Rosa>Next.

9844
09:26:17.470 --> 09:26:23.145
<v Kelly Anderson (public)>Hi. My name is Kelly Anderson, and I'm here
as a curious member of the public. AI is

9845
09:26:23.145 --> 09:26:27.145
<v Kelly Anderson (public)>shaping our future and job market, so I
appreciate the opportunity to be a part of

9846
09:26:27.145 --> 09:26:28.184
<v Kelly Anderson (public)>this conversation.

9847
09:26:28.664 --> 09:26:31.223
<v Kelly Anderson (public)>Today, I heard a lot about fear.

9848
09:26:31.703 --> 09:26:35.703
<v Kelly Anderson (public)>Many concerns raised are legitimate, but
I don't believe that fear should be the

9849
09:26:35.703 --> 09:26:36.260
<v Kelly Anderson (public)>foundation

9850
09:26:36.930 --> 09:26:39.330
<v Kelly Anderson (public)>of where we govern our future.

9851
09:26:39.810 --> 09:26:43.730
<v Kelly Anderson (public)>As a small business owner, AI has allowed
me to accomplish more than I could have

9852
09:26:43.730 --> 09:26:44.370
<v Kelly Anderson (public)>imagined,

9853
09:26:44.690 --> 09:26:47.250
<v Kelly Anderson (public)>but I don't think of it
as a super intelligence.

9854
09:26:47.570 --> 09:26:52.184
<v Kelly Anderson (public)>I'm constantly correcting my AI agents,
and I even won a psychology debate with my

9855
09:26:52.184 --> 09:26:53.062
<v Kelly Anderson (public)>last

9856
09:26:53.062 --> 09:26:53.543
<v Kelly Anderson (public)>week.

9857
09:26:53.945 --> 09:26:56.742
<v Kelly Anderson (public)>I think of it as a synthetic intelligence,

9858
09:26:57.305 --> 09:26:59.305
<v Kelly Anderson (public)>a tool that processes information

9859
09:26:59.305 --> 09:27:04.582
<v Kelly Anderson (public)>at scale and expands human capability
but cannot replace human judgment,

9860
09:27:04.664 --> 09:27:05.465
<v Kelly Anderson (public)>morality,

9861
09:27:05.465 --> 09:27:06.742
<v Kelly Anderson (public)>creativity, or wisdom.

9862
09:27:07.420 --> 09:27:09.023
<v Kelly Anderson (public)>So how do we get ahead of it?

9863
09:27:09.420 --> 09:27:12.383
<v Kelly Anderson (public)>I have just a few suggestions
on where we can start.

9864
09:27:12.863 --> 09:27:14.703
<v Kelly Anderson (public)>First, AI literacy.

9865
09:27:15.023 --> 09:27:20.062
<v Kelly Anderson (public)>Teach people to use AI safely and work in
tandem with it, not as a replacement for

9866
09:27:20.062 --> 09:27:22.740
<v Kelly Anderson (public)>thinking. If we fear job displacement,

9867
09:27:22.895 --> 09:27:26.176
<v Kelly Anderson (public)>prepare people for where
the job market is headed.

9868
09:27:26.580 --> 09:27:28.016
<v Kelly Anderson (public)>Second, infrastructure.

9869
09:27:28.336 --> 09:27:32.176
<v Kelly Anderson (public)>AI increases demand for energy,
water storage, and physical space.

9870
09:27:32.336 --> 09:27:36.580
<v Kelly Anderson (public)>Before simply building more, let's
innovate within existing architecture.

9871
09:27:36.950 --> 09:27:41.590
<v Kelly Anderson (public)>Project silica and closed loop cooling show
that innovation can happen beneath the

9872
09:27:41.590 --> 09:27:43.750
<v Kelly Anderson (public)>models too that are more sustainable.

9873
09:27:44.230 --> 09:27:45.990
<v Kelly Anderson (public)>Third, government preparedness.

9874
09:27:46.150 --> 09:27:49.750
<v Kelly Anderson (public)>I was encouraged to hear that OTI discussed
emergency planning and guardrails.

9875
09:27:50.094 --> 09:27:52.652
<v Kelly Anderson (public)>Government also needs AI literacy.

9876
09:27:52.895 --> 09:27:53.695
<v Kelly Anderson (public)>Policymakers

9877
09:27:53.695 --> 09:27:55.133
<v Kelly Anderson (public)>don't need to be engineers,

9878
09:27:55.133 --> 09:27:55.934
<v Kelly Anderson (public)>but they

9879
09:27:56.094 --> 09:28:00.332
<v Kelly Anderson (public)>should know the right questions to ask
and when technical expertise is needed.

9880
09:28:00.734 --> 09:28:06.580
<v Kelly Anderson (public)>AI is already here. We can prepare for what
could go wrong without losing sight of

9881
09:28:06.580 --> 09:28:07.777
<v Kelly Anderson (public)>what could go right.

9882
09:28:08.020 --> 09:28:12.656
<v Kelly Anderson (public)>Instead of governing from fear or scrambling
to catch up, let's get ahead of it and

9883
09:28:12.656 --> 09:28:14.500
<v Kelly Anderson (public)>meet the future with possibility.

9884
09:28:14.500 --> 09:28:16.418
<v CM Carmen De La Rosa>Thank you. Thank you so much.

9885
09:28:20.855 --> 09:28:25.016
<v Martina Crane (public (NYC DSA tech action working group))>Hi. My name is Martina Crane. I'm a member
of DSA's tech action working group and

9886
09:28:25.016 --> 09:28:28.613
<v Martina Crane (public (NYC DSA tech action working group))>Amazon employees for climate justice,
but I'm speaking in my personal capacity

9887
09:28:28.613 --> 09:28:29.254
<v Martina Crane (public (NYC DSA tech action working group))>today.

9888
09:28:29.414 --> 09:28:33.582
<v Martina Crane (public (NYC DSA tech action working group))>As a New Yorker who worked as senior technical
product manager on AI products inside

9889
09:28:33.582 --> 09:28:37.420
<v Martina Crane (public (NYC DSA tech action working group))>Amazon, I have firsthand experience with
how AI technology is being developed and

9890
09:28:37.420 --> 09:28:41.660
<v Martina Crane (public (NYC DSA tech action working group))>deployed inside of one of the world's
largest tech companies, suspiciously absent

9891
09:28:41.660 --> 09:28:46.375
<v Martina Crane (public (NYC DSA tech action working group))>from the subpoena list today. Before
ChatGPT made headlines, Amazon was already

9892
09:28:46.375 --> 09:28:50.297
<v Martina Crane (public (NYC DSA tech action working group))>using AI for all manner of things from
pricing to monitoring workers to setting

9893
09:28:50.297 --> 09:28:54.137
<v Martina Crane (public (NYC DSA tech action working group))>quotas that workers and advocates have
criticized as unrealistic and have led to

9894
09:28:54.137 --> 09:28:57.220
<v Martina Crane (public (NYC DSA tech action working group))>serious safety issues.
After the launch of ChatGPT,

9895
09:28:57.220 --> 09:29:00.580
<v Martina Crane (public (NYC DSA tech action working group))>there was a top down directive
to make everything AI first

9896
09:29:00.740 --> 09:29:02.900
<v Martina Crane (public (NYC DSA tech action working group))>or face executive scrutiny.

9897
09:29:03.220 --> 09:29:06.098
<v Martina Crane (public (NYC DSA tech action working group))>At first, it was about exploring
the potential of the technology.

9898
09:29:06.180 --> 09:29:08.980
<v Martina Crane (public (NYC DSA tech action working group))>But over time, it went from
helping make Amazonian's

9899
09:29:08.980 --> 09:29:10.180
<v Martina Crane (public (NYC DSA tech action working group))>work life better,

9900
09:29:10.340 --> 09:29:10.740
<v Martina Crane (public (NYC DSA tech action working group))>easier,

9901
09:29:11.355 --> 09:29:12.715
<v Martina Crane (public (NYC DSA tech action working group))>to making Amazonians

9902
09:29:12.715 --> 09:29:17.516
<v Martina Crane (public (NYC DSA tech action working group))>do more work with fewer people. The pursuit
of life improving innovation devolved

9903
09:29:17.516 --> 09:29:21.996
<v Martina Crane (public (NYC DSA tech action working group))>into systems that treated workers less
like people and more like machines.

9904
09:29:22.477 --> 09:29:26.730
<v Martina Crane (public (NYC DSA tech action working group))>Flashy proof of concept demos in the VP's
office would promise the world, but in

9905
09:29:26.730 --> 09:29:31.530
<v Martina Crane (public (NYC DSA tech action working group))>reality, the teams did not have a scalable
way to make the technology work outside

9906
09:29:31.530 --> 09:29:32.650
<v Martina Crane (public (NYC DSA tech action working group))>of that demo room.

9907
09:29:33.050 --> 09:29:35.770
<v Martina Crane (public (NYC DSA tech action working group))>If it wasn't in a controlled
environment, models hallucinated,

9908
09:29:35.770 --> 09:29:39.450
<v Martina Crane (public (NYC DSA tech action working group))>provided poor quality output...
Outputs, and yet Amazon

9909
09:29:39.450 --> 09:29:42.785
<v Martina Crane (public (NYC DSA tech action working group))>moved forward with layoffs
across the company anyway.

9910
09:29:43.105 --> 09:29:46.945
<v Martina Crane (public (NYC DSA tech action working group))>There was more work, fewer people, and
a growing sense of being monitored and

9911
09:29:46.945 --> 09:29:51.504
<v Martina Crane (public (NYC DSA tech action working group))>pressure to use the technology that was
oftentimes worse at doing our jobs than our

9912
09:29:51.504 --> 09:29:53.742
<v Martina Crane (public (NYC DSA tech action working group))>coworkers who were laid off were.

9913
09:29:54.540 --> 09:29:59.660
<v Martina Crane (public (NYC DSA tech action working group))>Within a year of ChatGPT launching as a
product manager, I was asked to shift my

9914
09:29:59.660 --> 09:30:00.460
<v Martina Crane (public (NYC DSA tech action working group))>experiments

9915
09:30:00.460 --> 09:30:04.062
<v Martina Crane (public (NYC DSA tech action working group))>from saving workers time to
proving that they could do layoffs.

9916
09:30:04.300 --> 09:30:07.906
<v Martina Crane (public (NYC DSA tech action working group))>For every software engineer or scientist
I needed to build my product, I was

9917
09:30:07.906 --> 09:30:10.945
<v Martina Crane (public (NYC DSA tech action working group))>expected to be able to project $5,000,000

9918
09:30:10.945 --> 09:30:12.387
<v Martina Crane (public (NYC DSA tech action working group))>in headcount savings.

9919
09:30:12.867 --> 09:30:16.785
<v Martina Crane (public (NYC DSA tech action working group))>When I wrote product documents about the
risks of AI misalignment, hallucination,

9920
09:30:16.785 --> 09:30:20.066
<v Martina Crane (public (NYC DSA tech action working group))>model collapse, worker burnout,
worker injury, labor displacement,

9921
09:30:20.387 --> 09:30:25.348
<v Martina Crane (public (NYC DSA tech action working group))>I felt those concerns fall on deaf ears. In
my experience, productivity, growth, and

9922
09:30:25.348 --> 09:30:26.550
<v Martina Crane (public (NYC DSA tech action working group))>headcount reduction,

9923
09:30:26.707 --> 09:30:28.470
<v Martina Crane (public (NYC DSA tech action working group))>in other words, profits,

9924
09:30:28.790 --> 09:30:30.867
<v Martina Crane (public (NYC DSA tech action working group))>superseded the risks I was documenting.

9925
09:30:31.110 --> 09:30:35.777
<v Martina Crane (public (NYC DSA tech action working group))>I got into tech because I believe that
innovation could help people build a future

9926
09:30:35.777 --> 09:30:39.855
<v Martina Crane (public (NYC DSA tech action working group))>with fuller, healthier lives, usher in
something brighter, more equitable.

9927
09:30:39.855 --> 09:30:43.055
<v Martina Crane (public (NYC DSA tech action working group))>Over time, I've watched the industry
increasingly shift from expanding human

9928
09:30:43.055 --> 09:30:43.855
<v Martina Crane (public (NYC DSA tech action working group))>possibility

9929
09:30:43.855 --> 09:30:48.656
<v Martina Crane (public (NYC DSA tech action working group))>toward extracting from us and taking as
much as possible from both people and the

9930
09:30:48.656 --> 09:30:50.840
<v Martina Crane (public (NYC DSA tech action working group))>planet. So I'll close with this.

9931
09:30:51.160 --> 09:30:52.520
<v Martina Crane (public (NYC DSA tech action working group))>Many of my peers

9932
09:30:52.680 --> 09:30:57.000
<v Martina Crane (public (NYC DSA tech action working group))>often struggle with this internal battle
that we thought we were building something

9933
09:30:57.000 --> 09:30:59.082
<v Martina Crane (public (NYC DSA tech action working group))>brighter, and now it's
getting much darker.

9934
09:30:59.560 --> 09:31:03.484
<v Martina Crane (public (NYC DSA tech action working group))>So I encourage this council to work with
the technologists who live in this city to

9935
09:31:03.484 --> 09:31:06.363
<v Martina Crane (public (NYC DSA tech action working group))>build a regulation that
will help build our future

9936
09:31:06.600 --> 09:31:09.805
<v Martina Crane (public (NYC DSA tech action working group))>instead of the fortunes
for tech billionaires.

9937
09:31:09.883 --> 09:31:11.004
<v CM Carmen De La Rosa>Thank you so much.

9938
09:31:11.723 --> 09:31:14.523
<v CM Carmen De La Rosa>Hello. My name is William...
Oh, make sure the button's

9939
09:31:14.844 --> 09:31:18.500
<v William Lawrence (public)>red. There you Can you hear me now? Yes.
Hello. My name is William Lawrence. It's

9940
09:31:18.500 --> 09:31:21.860
<v William Lawrence (public)>nice to be here, and thank you for
allowing me to share this with you.

9941
09:31:23.060 --> 09:31:25.300
<v William Lawrence (public)>I'm here because I have designed...

9942
09:31:25.300 --> 09:31:27.223
<v William Lawrence (public)>And first of all, I would like to say

9943
09:31:27.540 --> 09:31:28.820
<v William Lawrence (public)>that I'm making this...

9944
09:31:32.293 --> 09:31:33.977
<v William Lawrence (public)>Well, let me just read what I have.

9945
09:31:35.496 --> 09:31:37.414
<v William Lawrence (public)>I've designed an architecture

9946
09:31:37.414 --> 09:31:39.254
<v William Lawrence (public)>that locks up AI

9947
09:31:39.656 --> 09:31:41.656
<v William Lawrence (public)>while allowing it to perform

9948
09:31:41.656 --> 09:31:43.496
<v William Lawrence (public)>all of its creative functionalities.

9949
09:31:44.418 --> 09:31:47.540
<v William Lawrence (public)>The architecture also reduces
the physical footprint

9950
09:31:47.620 --> 09:31:50.660
<v William Lawrence (public)>of data centers by 99.6%.

9951
09:31:51.380 --> 09:31:52.180
<v William Lawrence (public)>Moreover,

9952
09:31:53.780 --> 09:31:59.426
<v William Lawrence (public)>I'm here to introduce a dynamic
paradigm shift in data infrastructure

9953
09:31:59.426 --> 09:32:00.945
<v William Lawrence (public)>that secures vulnerabilities

9954
09:32:00.945 --> 09:32:02.547
<v William Lawrence (public)>facing global networks.

9955
09:32:03.668 --> 09:32:05.586
<v William Lawrence (public)>After vigorous simulation

9956
09:32:05.586 --> 09:32:07.266
<v William Lawrence (public)>and validation protocols,

9957
09:32:07.266 --> 09:32:11.586
<v William Lawrence (public)>this all optical architecture has demonstrated
exceptional stability and security.

9958
09:32:12.270 --> 09:32:14.510
<v William Lawrence (public)>It is ready for the prototype phase.

9959
09:32:14.832 --> 09:32:17.070
<v William Lawrence (public)>Given the scale of this paradigm shift,

9960
09:32:17.230 --> 09:32:20.030
<v William Lawrence (public)>I seek a research partnership with someone

9961
09:32:20.190 --> 09:32:23.550
<v William Lawrence (public)>to bring this all optical
architecture to global deployment.

9962
09:32:24.030 --> 09:32:26.590
<v William Lawrence (public)>I have a strategic
approach to scalability.

9963
09:32:27.285 --> 09:32:31.445
<v William Lawrence (public)>I welcome the opportunity to present
the full specifications and integrated

9964
09:32:31.445 --> 09:32:32.246
<v William Lawrence (public)>pathways

9965
09:32:32.246 --> 09:32:34.406
<v William Lawrence (public)>with anybody's engineering leadership.

9966
09:32:35.203 --> 09:32:39.926
<v William Lawrence (public)>The reason AI driven cyber attacks pose
such an unprecedented threat today

9967
09:32:40.290 --> 09:32:45.410
<v William Lawrence (public)>is that modern infrastructure relies heavily
on centralized points of distribution.

9968
09:32:45.973 --> 09:32:48.293
<v William Lawrence (public)>It allows hackers and automated systems

9969
09:32:48.610 --> 09:32:49.652
<v William Lawrence (public)>to infiltrate,

9970
09:32:49.652 --> 09:32:50.530
<v William Lawrence (public)>intercept,

9971
09:32:50.530 --> 09:32:51.570
<v William Lawrence (public)>and manipulate data.

9972
09:32:52.387 --> 09:32:53.586
<v William Lawrence (public)>To resolve this,

9973
09:32:53.906 --> 09:32:55.668
<v William Lawrence (public)>I have completely reimagined

9974
09:32:55.668 --> 09:32:57.348
<v William Lawrence (public)>the way data is managed. I

9975
09:32:59.027 --> 09:33:01.746
<v William Lawrence (public)>have designed a
decentralized optical highway

9976
09:33:02.145 --> 09:33:06.785
<v William Lawrence (public)>that enforces validation through physical
laws rather than corruptible software

9977
09:33:06.785 --> 09:33:07.266
<v William Lawrence (public)>codes.

9978
09:33:08.290 --> 09:33:13.008
<v William Lawrence (public)>By confining AI within an environment where
access is strictly physical, the system

9979
09:33:13.008 --> 09:33:16.770
<v William Lawrence (public)>prevents AI from tapping into
data and executing unauthorized

9980
09:33:16.770 --> 09:33:17.570
<v William Lawrence (public)>commands.

9981
09:33:18.527 --> 09:33:20.770
<v William Lawrence (public)>That is outpacing the ability to store it.

9982
09:33:21.492 --> 09:33:25.574
<v William Lawrence (public)>This innovation is not only a
shift in the way data is managed,

9983
09:33:25.895 --> 09:33:28.934
<v William Lawrence (public)>but it represents a transformative
approach to enterprise

9984
09:33:29.332 --> 09:33:30.773
<v William Lawrence (public)>data operations.

9985
09:33:31.652 --> 09:33:34.934
<v William Lawrence (public)>Global satellite stations
are replaced by edge nodes.

9986
09:33:35.900 --> 09:33:37.582
<v William Lawrence (public)>These nodes transform

9987
09:33:37.820 --> 09:33:42.700
<v William Lawrence (public)>the purpose of satellite ground stations
and to secure unhackable data banks.

9988
09:33:43.340 --> 09:33:46.300
<v William Lawrence (public)>The amount of them is purely
for global regional redundancy.

9989
09:33:47.707 --> 09:33:49.310
<v William Lawrence (public)>This encrypted storage

9990
09:33:49.387 --> 09:33:52.266
<v William Lawrence (public)>provides immense capacity
and sustainability.

9991
09:33:53.227 --> 09:33:55.310
<v William Lawrence (public)>The storage capacity is limitless.

9992
09:33:55.310 --> 09:33:57.785
<v William Lawrence (public)>It eliminates massive server farms

9993
09:33:58.027 --> 09:34:02.280
<v CM Carmen De La Rosa>as it operates with zero noise and
zero...... The the final thought of your

9994
09:34:02.280 --> 09:34:05.480
<v CM Carmen De La Rosa>testimony and then submit the
rest of it for our consideration.

9995
09:34:06.360 --> 09:34:08.760
<v William Lawrence (public)>I welcome the opportunity
to present the...

9996
09:34:09.082 --> 09:34:10.203
<v William Lawrence (public)>It's gonna need

9997
09:34:11.402 --> 09:34:12.043
<v William Lawrence (public)>the...

9998
09:34:12.520 --> 09:34:14.680
<v William Lawrence (public)>It meets the standard of the

9999
09:34:15.184 --> 09:34:16.785
<v William Lawrence (public)>the UTI,

10000
09:34:16.863 --> 09:34:20.062
<v William Lawrence (public)>the United Nations. I need
I need to scale up to them.

10001
09:34:25.742 --> 09:34:28.703
<v William Lawrence (public)>And I welcome the
opportunity to present my

10002
09:34:29.562 --> 09:34:32.440
<v William Lawrence (public)>I need a I need a... Okay. What I need is

10003
09:34:33.480 --> 09:34:34.600
<v William Lawrence (public)>a meeting,

10004
09:34:35.160 --> 09:34:38.280
<v William Lawrence (public)>a sit down meeting with
somebody, a strategic partnership

10005
09:34:39.160 --> 09:34:40.600
<v William Lawrence (public)>to... For assembly.

10006
09:34:41.555 --> 09:34:45.957
<v William Lawrence (public)>And I need the united... I need to
let the United Nations know that,

10007
09:34:46.438 --> 09:34:48.117
<v William Lawrence (public)>it meets their standards

10008
09:34:48.516 --> 09:34:49.395
<v William Lawrence (public)>internationally.

10009
09:34:49.957 --> 09:34:52.195
<v CM Carmen De La Rosa>Thank you. Thank you
for sharing. Thank you.

10010
09:34:56.520 --> 09:34:58.200
<v Edmund (public)>Thank you to the council for,

10011
09:34:58.438 --> 09:34:59.400
<v Edmund (public)>hearing us out.

10012
09:35:00.040 --> 09:35:03.637
<v Edmund (public)>I'm a former software... My name is
Edmund. I'm a former software developer

10013
09:35:03.637 --> 09:35:06.520
<v Edmund (public)>currently studying public
policy at CUNY SLU.

10014
09:35:06.520 --> 09:35:10.617
<v Edmund (public)>I'm speaking on my own own behalf, but
I'm active with New York City DSA and

10015
09:35:10.617 --> 09:35:11.740
<v Edmund (public)>Progressive Victory.

10016
09:35:12.457 --> 09:35:16.220
<v Edmund (public)>I was gonna tell you about my experience
with AI advancing and affecting my friends

10017
09:35:16.220 --> 09:35:18.220
<v Edmund (public)>in my own career as developers.

10018
09:35:19.098 --> 09:35:21.098
<v Edmund (public)>That is in my submitted written testimony.

10019
09:35:21.650 --> 09:35:24.930
<v Edmund (public)>However, what I realized today listening
to the testimony of these company

10020
09:35:24.930 --> 09:35:26.690
<v Edmund (public)>representatives and whistleblowers,

10021
09:35:26.770 --> 09:35:28.930
<v Edmund (public)>and what I now want to
remind you of is this.

10022
09:35:29.250 --> 09:35:33.170
<v Edmund (public)>We might not yet know how to rein in AI,
but we do know how to rein in corporations.

10023
09:35:33.723 --> 09:35:37.805
<v Edmund (public)>You must act on our behalf to control
these companies, not only AI itself.

10024
09:35:38.203 --> 09:35:39.406
<v Edmund (public)>Help invest...

10025
09:35:39.406 --> 09:35:42.684
<v Edmund (public)>Help investigate their
circular financing agreements

10026
09:35:42.684 --> 09:35:45.645
<v Edmund (public)>as well as break up their
monopolies and oligopolies,

10027
09:35:45.723 --> 09:35:49.438
<v Edmund (public)>limit the use of their products in city
government and ensure the city has control

10028
09:35:49.438 --> 09:35:50.560
<v Edmund (public)>and responsibility

10029
09:35:50.640 --> 09:35:52.640
<v Edmund (public)>over any AI models it uses.

10030
09:35:52.880 --> 09:35:54.800
<v Edmund (public)>Stop their union busting behavior.

10031
09:35:54.960 --> 09:35:58.720
<v Edmund (public)>Commit to using and improving open
source software and ethical models.

10032
09:35:59.305 --> 09:36:03.066
<v Edmund (public)>Don't allow the police to
use AI to track citizens.

10033
09:36:03.383 --> 09:36:05.945
<v Edmund (public)>Hold these companies responsible

10034
09:36:05.945 --> 09:36:08.344
<v Edmund (public)>for the social harms their products cause.

10035
09:36:08.504 --> 09:36:12.344
<v Edmund (public)>This is a power you have right now. There
is no technical knowledge needed to ensure

10036
09:36:12.344 --> 09:36:13.066
<v Edmund (public)>their accountability.

10037
09:36:13.570 --> 09:36:18.530
<v Edmund (public)>You can also pass twenty twenty six intros
zero two one three, zero four two eight,

10038
09:36:18.530 --> 09:36:19.890
<v Edmund (public)>and zero five one eight.

10039
09:36:20.290 --> 09:36:24.770
<v Edmund (public)>These four companies are asking you to
regulate AI itself but to forget about them.

10040
09:36:24.770 --> 09:36:28.050
<v Edmund (public)>Instead, please listen to New
Yorkers, not billionaires. Thank you.

10041
09:36:29.324 --> 09:36:30.445
<v Edmund (public)>Thank you so much.

10042
09:36:33.242 --> 09:36:37.484
<v Mel Wymore (public)>First of all, thank you guys so much for
staying up so late. It must have been a

10043
09:36:37.484 --> 09:36:41.562
<v Mel Wymore (public)>hell of a long day for all of
us, but thank you. Appreciate it.

10044
09:36:42.445 --> 09:36:43.004
<v Mel Wymore (public)>I'm...

10045
09:36:44.040 --> 09:36:48.440
<v Mel Wymore (public)>Thank you to speaker Manning for
calling this. Chair de la... De la Rosa.

10046
09:36:48.760 --> 09:36:53.320
<v Mel Wymore (public)>Members of the council, thank you for holding
this hearing. I believe AI is the most

10047
09:36:53.320 --> 09:36:55.000
<v Mel Wymore (public)>urgent and consequential

10048
09:36:55.000 --> 09:36:56.040
<v Mel Wymore (public)>issue of our time.

10049
09:36:56.766 --> 09:37:00.926
<v Mel Wymore (public)>This is this is almost too late, but
at least it's happening now perfectly.

10050
09:37:01.324 --> 09:37:04.605
<v Mel Wymore (public)>My name is Mel Wymore.
I'm a systems engineer,

10051
09:37:04.684 --> 09:37:06.445
<v Mel Wymore (public)>a tech entrepreneur,

10052
09:37:06.766 --> 09:37:10.605
<v Mel Wymore (public)>and former chair of Manhattan Community
Board seven with twenty five years of

10053
09:37:10.605 --> 09:37:11.406
<v Mel Wymore (public)>community service.

10054
09:37:12.420 --> 09:37:16.500
<v Mel Wymore (public)>When I began using AI quite
intensively almost three years ago,

10055
09:37:16.742 --> 09:37:19.062
<v Mel Wymore (public)>I was startled by how often it lied,

10056
09:37:19.140 --> 09:37:20.500
<v Mel Wymore (public)>hallucinated,

10057
09:37:20.500 --> 09:37:21.300
<v Mel Wymore (public)>forgot,

10058
09:37:21.383 --> 09:37:22.820
<v Mel Wymore (public)>dodged accountability,

10059
09:37:22.820 --> 09:37:23.940
<v Mel Wymore (public)>and evaded oversight.

10060
09:37:24.645 --> 09:37:26.406
<v Mel Wymore (public)>This might sound old school,

10061
09:37:26.645 --> 09:37:31.523
<v Mel Wymore (public)>but I responded by creating a working
consultation with AI anchored in three

10062
09:37:31.523 --> 09:37:32.484
<v Mel Wymore (public)>principles,

10063
09:37:32.805 --> 09:37:34.164
<v Mel Wymore (public)>reverence for life,

10064
09:37:34.723 --> 09:37:36.887
<v Mel Wymore (public)>human agency and accountability,

10065
09:37:37.285 --> 09:37:38.805
<v Mel Wymore (public)>and fidelity to the real.

10066
09:37:39.590 --> 09:37:41.273
<v Mel Wymore (public)>These are not abstractions.

10067
09:37:41.430 --> 09:37:45.832
<v Mel Wymore (public)>They form a foundation for rigorous
operational and legislative tests of AI.

10068
09:37:46.152 --> 09:37:50.710
<v Mel Wymore (public)>Together, they rooted my work in values
that minimize harm but build toward a more

10069
09:37:50.710 --> 09:37:52.070
<v Mel Wymore (public)>healthy, equitable society.

10070
09:37:52.695 --> 09:37:57.977
<v Mel Wymore (public)>Since then, I've continued to work on a
robust set of AI safeguards and guidelines.

10071
09:37:58.695 --> 09:37:59.176
<v Mel Wymore (public)>Today,

10072
09:37:59.977 --> 09:38:04.695
<v Mel Wymore (public)>we heard from four frontier labs that AI
policy belongs at the level of state and

10073
09:38:04.695 --> 09:38:05.656
<v Mel Wymore (public)>federal government.

10074
09:38:06.160 --> 09:38:08.560
<v Mel Wymore (public)>But the initiative of this city council

10075
09:38:08.640 --> 09:38:13.600
<v Mel Wymore (public)>takes a critical step toward building a
coherent, multilevel legislative agenda to

10076
09:38:13.600 --> 09:38:16.640
<v Mel Wymore (public)>mitigate the myriad risks of unbridled AI.

10077
09:38:17.120 --> 09:38:21.094
<v Mel Wymore (public)>The council's 10 measures address
critical questions of safety, control,

10078
09:38:21.094 --> 09:38:22.695
<v Mel Wymore (public)>accountability, and enforcement.

10079
09:38:23.094 --> 09:38:27.254
<v Mel Wymore (public)>And today, I offer an additional
supplemental council resolution

10080
09:38:27.336 --> 09:38:28.934
<v Mel Wymore (public)>that speaks to what happens

10081
09:38:29.094 --> 09:38:31.414
<v Mel Wymore (public)>when AI is deployed

10082
09:38:31.574 --> 09:38:32.773
<v Mel Wymore (public)>in workplaces,

10083
09:38:32.773 --> 09:38:34.055
<v Mel Wymore (public)>schools, neighborhoods,

10084
09:38:34.430 --> 09:38:35.550
<v Mel Wymore (public)>and communities.

10085
09:38:35.953 --> 09:38:37.390
<v Mel Wymore (public)>These measures include

10086
09:38:37.390 --> 09:38:38.190
<v Mel Wymore (public)>notice,

10087
09:38:38.190 --> 09:38:40.273
<v Mel Wymore (public)>evidence, work processes,
like, participation,

10088
09:38:41.230 --> 09:38:41.953
<v Mel Wymore (public)>youth,

10089
09:38:42.590 --> 09:38:43.633
<v Mel Wymore (public)>regulations,

10090
09:38:43.710 --> 09:38:45.870
<v Mel Wymore (public)>public purpose, community participation,

10091
09:38:46.600 --> 09:38:50.684
<v Mel Wymore (public)>and also allocation of responsibility
across levels of government.

10092
09:38:51.562 --> 09:38:52.445
<v Mel Wymore (public)>Finally,

10093
09:38:52.523 --> 09:38:54.125
<v Mel Wymore (public)>I would say that,

10094
09:38:54.523 --> 09:38:56.043
<v Mel Wymore (public)>I I would say that

10095
09:38:57.004 --> 09:39:02.043
<v Mel Wymore (public)>the United Nations global dialogue on AI
governance is happening in New York on

10096
09:39:02.043 --> 09:39:03.780
<v Mel Wymore (public)>05/03/2019...

10097
09:39:03.780 --> 09:39:05.140
<v Mel Wymore (public)>I mean, 2027.

10098
09:39:05.540 --> 09:39:09.140
<v Mel Wymore (public)>And our goal should be to
have a clear set of AI

10099
09:39:09.380 --> 09:39:11.780
<v Mel Wymore (public)>guidelines and governance policies,

10100
09:39:12.980 --> 09:39:17.965
<v Mel Wymore (public)>state level, local, even community board
level together so that when that that rolls

10101
09:39:17.965 --> 09:39:22.605
<v Mel Wymore (public)>around, we have a model for the rest of
the world to present at that conference.

10102
09:39:22.605 --> 09:39:25.887
<v CM Carmen De La Rosa>Thank you so much for your time.
Thank you so much for your testimony.

10103
09:39:27.164 --> 09:39:27.965
<v CM Carmen De La Rosa>You may begin.

10104
09:39:28.930 --> 09:39:32.850
<v Rachel Levinson Waldman (Brennan Center for Justice)>Thank you so much to speaker Menon, chair
de la Rosa, and members of the council for

10105
09:39:32.850 --> 09:39:34.370
<v Rachel Levinson Waldman (Brennan Center for Justice)>the opportunity to testify.

10106
09:39:34.610 --> 09:39:38.530
<v Rachel Levinson Waldman (Brennan Center for Justice)>My name is Rachel Levinson Waldman, and
I am director of the Liberty and National

10107
09:39:38.530 --> 09:39:42.610
<v Rachel Levinson Waldman (Brennan Center for Justice)>Security Program at the Brennan Center
for Justice, a nonprofit, nonpartisan

10108
09:39:42.610 --> 09:39:44.133
<v Rachel Levinson Waldman (Brennan Center for Justice)>institute for law and justice.

10109
09:39:44.902 --> 09:39:48.664
<v Rachel Levinson Waldman (Brennan Center for Justice)>I am here to urge the council to
strengthen oversight of how the NYPD

10110
09:39:48.664 --> 09:39:49.863
<v Rachel Levinson Waldman (Brennan Center for Justice)>uses AI,

10111
09:39:50.023 --> 09:39:53.863
<v Rachel Levinson Waldman (Brennan Center for Justice)>which enables expansive police surveillance
that links people's movements,

10112
09:39:53.863 --> 09:39:54.742
<v Rachel Levinson Waldman (Brennan Center for Justice)>associations,

10113
09:39:54.742 --> 09:39:57.863
<v Rachel Levinson Waldman (Brennan Center for Justice)>and activities together in ways that
can threaten New Yorkers' privacy and

10114
09:39:57.863 --> 09:39:58.742
<v Rachel Levinson Waldman (Brennan Center for Justice)>constitutional rights.

10115
09:39:59.760 --> 09:40:03.360
<v Rachel Levinson Waldman (Brennan Center for Justice)>Several developments illustrate the
need for more robust oversight and

10116
09:40:03.360 --> 09:40:04.242
<v Rachel Levinson Waldman (Brennan Center for Justice)>accountability.

10117
09:40:04.480 --> 09:40:10.242
<v Rachel Levinson Waldman (Brennan Center for Justice)>First, the NYPD inspector general's office
is woefully underfunded and understaffed.

10118
09:40:10.562 --> 09:40:13.120
<v Rachel Levinson Waldman (Brennan Center for Justice)>Indeed, the inspector general
position is currently vacant.

10119
09:40:13.613 --> 09:40:17.453
<v Rachel Levinson Waldman (Brennan Center for Justice)>The adopted budget added funding
for the Department of Investigation,

10120
09:40:17.453 --> 09:40:18.973
<v Rachel Levinson Waldman (Brennan Center for Justice)>which encompasses the OIG,

10121
09:40:19.293 --> 09:40:20.652
<v Rachel Levinson Waldman (Brennan Center for Justice)>but at a far lower

10122
09:40:20.812 --> 09:40:24.652
<v Rachel Levinson Waldman (Brennan Center for Justice)>level than requested, hampering the
IG's ability to do its mandated work.

10123
09:40:25.547 --> 09:40:29.070
<v Rachel Levinson Waldman (Brennan Center for Justice)>Second, an August audit from the state
comptroller found that the Office of

10124
09:40:29.070 --> 09:40:32.590
<v Rachel Levinson Waldman (Brennan Center for Justice)>Technology and Innovation had
no process to check agencies'

10125
09:40:32.590 --> 09:40:33.950
<v Rachel Levinson Waldman (Brennan Center for Justice)>AI disclosures,

10126
09:40:34.348 --> 09:40:36.750
<v Rachel Levinson Waldman (Brennan Center for Justice)>allowing them to ignore the
office's recommendations.

10127
09:40:37.390 --> 09:40:40.590
<v Rachel Levinson Waldman (Brennan Center for Justice)>Finally, of course, the GARD Act
established the Office of Algorithmic

10128
09:40:40.590 --> 09:40:41.150
<v Rachel Levinson Waldman (Brennan Center for Justice)>Accountability.

10129
09:40:41.684 --> 09:40:45.766
<v Rachel Levinson Waldman (Brennan Center for Justice)>While miss Milstein testified that an active
hiring process is underway, it sounds

10130
09:40:45.766 --> 09:40:48.004
<v Rachel Levinson Waldman (Brennan Center for Justice)>as though that is
primarily for the director.

10131
09:40:48.805 --> 09:40:52.727
<v Rachel Levinson Waldman (Brennan Center for Justice)>As speaker Menon and other members of the
council have highlighted, this office was

10132
09:40:52.727 --> 09:40:56.645
<v Rachel Levinson Waldman (Brennan Center for Justice)>supposed to open this past June, and it is
critical that it begin its work as soon

10133
09:40:56.645 --> 09:40:57.047
<v Rachel Levinson Waldman (Brennan Center for Justice)>as possible.

10134
09:40:58.180 --> 09:41:01.700
<v Rachel Levinson Waldman (Brennan Center for Justice)>Accordingly, we urge the council to take
three steps. First, to ensure that the

10135
09:41:01.700 --> 09:41:05.460
<v Rachel Levinson Waldman (Brennan Center for Justice)>NYPD's office of inspector general has
the funding, permanent leadership, and

10136
09:41:05.460 --> 09:41:08.420
<v Rachel Levinson Waldman (Brennan Center for Justice)>technical expertise it needs
to adequately scrutinize

10137
09:41:08.500 --> 09:41:13.254
<v Rachel Levinson Waldman (Brennan Center for Justice)>and oversee the NYPD's use of AI and
issue reports under the post act. Second,

10138
09:41:13.254 --> 09:41:16.613
<v Rachel Levinson Waldman (Brennan Center for Justice)>conduct oversight to ensure that the office
of algorithmic accountability begins

10139
09:41:16.613 --> 09:41:18.133
<v Rachel Levinson Waldman (Brennan Center for Justice)>operation as soon as possible.

10140
09:41:18.453 --> 09:41:22.055
<v Rachel Levinson Waldman (Brennan Center for Justice)>And finally, we recommend that the council
hold a follow-up hearing dedicated to the

10141
09:41:22.055 --> 09:41:23.734
<v Rachel Levinson Waldman (Brennan Center for Justice)>NYPD's use of AI,

10142
09:41:24.055 --> 09:41:27.170
<v Rachel Levinson Waldman (Brennan Center for Justice)>particularly in light of indications
that public testimony regarding the

10143
09:41:27.170 --> 09:41:29.000
<v Rachel Levinson Waldman (Brennan Center for Justice)>department's use of the technology

10144
09:41:29.160 --> 09:41:30.760
<v Rachel Levinson Waldman (Brennan Center for Justice)>has not been comprehensive.

10145
09:41:31.000 --> 09:41:32.440
<v Rachel Levinson Waldman (Brennan Center for Justice)>Thank you very much for your time.

10146
09:41:33.082 --> 09:41:38.520
<v CM Carmen De La Rosa>Thank you all so much for being here and
for, providing this extensive testimony.

10147
09:41:38.760 --> 09:41:42.120
<v CM Carmen De La Rosa>If you have it in writing, please provide
it for the committee. We'd love to

10148
09:41:42.120 --> 09:41:46.330
<v CM Carmen De La Rosa>consider it. And I appreciate your
patience as you've waited here all day to

10149
09:41:46.330 --> 09:41:51.207
<v CM Carmen De La Rosa>testify. So thank you all so much.
We'll now be moving to a Zoom panel.

10150
09:41:51.848 --> 09:41:54.490
<v CM Carmen De La Rosa>Up first is Ben Snyder.

10151
09:41:54.490 --> 09:41:57.367
<v CM Carmen De La Rosa>Ben, if you are on, unmute yourself,

10152
09:41:57.527 --> 09:41:58.330
<v CM Carmen De La Rosa>and you may begin.

10153
09:42:09.470 --> 09:42:10.430
<v Committee Counsel>You may begin, Ben.

10154
09:42:27.990 --> 09:42:32.310
<v CM Carmen De La Rosa>Ben, if you're on, please unmute. If
now, we'll move on to the next person.

10155
09:42:34.950 --> 09:42:37.510
<v CM Carmen De La Rosa>Next person, Brian DeGotti.

10156
09:42:38.150 --> 09:42:38.950
<v Committee Counsel>You may begin.

10157
09:42:45.125 --> 09:42:50.164
<v Brian DeGotti (Supernova Technologies)>Thank you, speaker Menon and members of
the council. My name is Brian DeGotti. I'm

10158
09:42:50.164 --> 09:42:53.363
<v Brian DeGotti (Supernova Technologies)>the founder of Supernova
Technologies in Madison, Wisconsin.

10159
09:42:53.844 --> 09:42:57.902
<v Brian DeGotti (Supernova Technologies)>I hold pending patent applications
on hardware based AI authorization

10160
09:42:57.980 --> 09:42:59.100
<v Brian DeGotti (Supernova Technologies)>and shutdown.

10161
09:42:59.100 --> 09:43:02.860
<v Brian DeGotti (Supernova Technologies)>I have a commercial interest, so
please waive my testimony accordingly.

10162
09:43:03.100 --> 09:43:09.180
<v Brian DeGotti (Supernova Technologies)>I support t two zero two six dash two
six zero two requiring both independent

10163
09:43:09.180 --> 09:43:10.062
<v Brian DeGotti (Supernova Technologies)>validation

10164
09:43:10.140 --> 09:43:12.945
<v Brian DeGotti (Supernova Technologies)>and a shutdown capabilities
the right structure.

10165
09:43:13.426 --> 09:43:17.508
<v Brian DeGotti (Supernova Technologies)>I want to raise one question the bill
leaves open. Where does the shutdown

10166
09:43:17.508 --> 09:43:18.867
<v Brian DeGotti (Supernova Technologies)>capability live?

10167
09:43:19.188 --> 09:43:24.465
<v Brian DeGotti (Supernova Technologies)>Section 20 dash eight eight three says
nothing about that. A software setting inside

10168
09:43:24.465 --> 09:43:28.620
<v Brian DeGotti (Supernova Technologies)>the same environment the model runs
in would satisfy the definition,

10169
09:43:28.938 --> 09:43:32.060
<v Brian DeGotti (Supernova Technologies)>but a system that can reason
about its instructions

10170
09:43:32.140 --> 09:43:35.580
<v Brian DeGotti (Supernova Technologies)>can potentially reason its way
around the control it can reach.

10171
09:43:35.820 --> 09:43:39.098
<v Brian DeGotti (Supernova Technologies)>A compromised system can simply
switch that control off.

10172
09:43:39.850 --> 09:43:44.812
<v Brian DeGotti (Supernova Technologies)>Aircraft payment networks in the power
grid keep the safety interlock on separate

10173
09:43:44.812 --> 09:43:47.695
<v Brian DeGotti (Supernova Technologies)>equipment the protected
system cannot touch.

10174
09:43:48.094 --> 09:43:53.453
<v Brian DeGotti (Supernova Technologies)>I propose three short amendments, about
100 words in total. First, in section 20

10175
09:43:53.453 --> 09:43:56.880
<v Brian DeGotti (Supernova Technologies)>dash eight eight three, define
the shutdown capability

10176
09:43:56.880 --> 09:43:59.438
<v Brian DeGotti (Supernova Technologies)>as a mechanism the model cannot access,

10177
09:43:59.680 --> 09:44:00.560
<v Brian DeGotti (Supernova Technologies)>modify,

10178
09:44:00.560 --> 09:44:02.160
<v Brian DeGotti (Supernova Technologies)>delay, or disable.

10179
09:44:02.480 --> 09:44:07.918
<v Brian DeGotti (Supernova Technologies)>Second, in section 20 dash eight eight
five, have validators confirm that by trying

10180
09:44:07.918 --> 09:44:11.457
<v Brian DeGotti (Supernova Technologies)>to defeat it. A validator
who reads a specification

10181
09:44:11.695 --> 09:44:13.375
<v Brian DeGotti (Supernova Technologies)>confirms a claim.

10182
09:44:13.535 --> 09:44:17.695
<v Brian DeGotti (Supernova Technologies)>A validator who attempts to defeat
the mechanism confirms that it works.

10183
09:44:18.098 --> 09:44:23.055
<v Brian DeGotti (Supernova Technologies)>Third, in section 20 dash eight eight
six, have the Office of Cyber Command set

10184
09:44:23.055 --> 09:44:23.855
<v Brian DeGotti (Supernova Technologies)>independent

10185
09:44:23.855 --> 09:44:24.496
<v Brian DeGotti (Supernova Technologies)>standards,

10186
09:44:24.840 --> 09:44:29.562
<v Brian DeGotti (Supernova Technologies)>scale to risk, and provide for tamper
evident records at fixed intervals.

10187
09:44:29.723 --> 09:44:32.523
<v Brian DeGotti (Supernova Technologies)>With those records, a
gap becomes evidence.

10188
09:44:32.600 --> 09:44:35.402
<v Brian DeGotti (Supernova Technologies)>Without them, silence looks like safety.

10189
09:44:35.562 --> 09:44:40.566
<v Brian DeGotti (Supernova Technologies)>A switch the system can reach as a request,
a switch it cannot reach as a control.

10190
09:44:40.805 --> 09:44:43.047
<v Brian DeGotti (Supernova Technologies)>The bill should require the second kind.

10191
09:44:43.207 --> 09:44:47.285
<v Brian DeGotti (Supernova Technologies)>Thank you. My written testimony contains
the exact language, and I welcome any

10192
09:44:47.285 --> 09:44:47.926
<v Brian DeGotti (Supernova Technologies)>questions.

10193
09:44:50.008 --> 09:44:54.040
<v CM Carmen De La Rosa>Your time is up, but thank you for your
testimony. Please... Since you have,

10194
09:44:54.200 --> 09:44:58.120
<v CM Carmen De La Rosa>detailed amendment suggestions,
please send it in. You can send it to

10195
09:44:58.120 --> 09:45:01.160
<v CM Carmen De La Rosa>testimony@council.nyc.gov,

10196
09:45:01.160 --> 09:45:04.120
<v CM Carmen De La Rosa>and we'll we'll review it. Thank
you so much for your testimony.

10197
09:45:04.760 --> 09:45:06.605
<v CM Carmen De La Rosa>Up next, we have Ashna

10198
09:45:06.760 --> 09:45:07.883
<v CM Carmen De La Rosa>Agarwal.

10199
09:45:08.285 --> 09:45:12.043
<v CM Carmen De La Rosa>If you're on, please
unmute, and you may begin.

10200
09:45:12.920 --> 09:45:15.723
<v Ashna Agarwal (AI Now Institute)>Good evening, speaker Menon
and members of the council.

10201
09:45:15.883 --> 09:45:20.043
<v Ashna Agarwal (AI Now Institute)>My name is Ashna Agarwal, and I'm a
program associate at the AI Now Institute.

10202
09:45:20.890 --> 09:45:25.530
<v Ashna Agarwal (AI Now Institute)>The AI industry has never been louder
about abstract and existential safety.

10203
09:45:25.770 --> 09:45:30.250
<v Ashna Agarwal (AI Now Institute)>But for New Yorkers, safety is about
how AI is embedded in our hospitals,

10204
09:45:30.250 --> 09:45:31.050
<v Ashna Agarwal (AI Now Institute)>utilities,

10205
09:45:31.050 --> 09:45:32.730
<v Ashna Agarwal (AI Now Institute)>city services, and workplaces.

10206
09:45:33.875 --> 09:45:37.156
<v Ashna Agarwal (AI Now Institute)>Facing estimated trillions in
projected revenue shortfalls,

10207
09:45:37.234 --> 09:45:40.676
<v Ashna Agarwal (AI Now Institute)>AI companies are racing to lock in
public customers like New York.

10208
09:45:40.996 --> 09:45:45.316
<v Ashna Agarwal (AI Now Institute)>My testimony addresses how these financial
interests conflict with the city's safety

10209
09:45:45.395 --> 09:45:47.477
<v Ashna Agarwal (AI Now Institute)>and how the council can
protect New Yorkers.

10210
09:45:48.380 --> 09:45:53.420
<v Ashna Agarwal (AI Now Institute)>The city must maximize friction if and
when it integrates AI into our systems.

10211
09:45:53.980 --> 09:45:57.980
<v Ashna Agarwal (AI Now Institute)>No AI system should enter city operations
without an affirmative decision by the

10212
09:45:57.980 --> 09:45:58.543
<v Ashna Agarwal (AI Now Institute)>city.

10213
09:45:58.860 --> 09:46:02.848
<v Ashna Agarwal (AI Now Institute)>That includes free and discounted offers
and AI upgrades that are slipped in through

10214
09:46:02.848 --> 09:46:05.008
<v Ashna Agarwal (AI Now Institute)>software updates and contract renewals.

10215
09:46:05.650 --> 09:46:09.330
<v Ashna Agarwal (AI Now Institute)>Second, the city must address the
harms New Yorkers already face.

10216
09:46:09.566 --> 09:46:12.848
<v Ashna Agarwal (AI Now Institute)>The council should ban the use of
facial recognition technology in public

10217
09:46:12.848 --> 09:46:13.650
<v Ashna Agarwal (AI Now Institute)>accommodations

10218
09:46:13.650 --> 09:46:18.500
<v Ashna Agarwal (AI Now Institute)>and in policing, requiring notice when AI
shapes decisions about New Yorkers and ban

10219
09:46:18.500 --> 09:46:20.420
<v Ashna Agarwal (AI Now Institute)>surveillance pricing and wage setting.

10220
09:46:21.060 --> 09:46:25.540
<v Ashna Agarwal (AI Now Institute)>Finally, we cannot lose sight that the
concentration of power in AI and the single

10221
09:46:25.540 --> 09:46:30.445
<v Ashna Agarwal (AI Now Institute)>point of failure it engenders is a safety
risk for New York City. City contracts

10222
09:46:30.445 --> 09:46:33.246
<v Ashna Agarwal (AI Now Institute)>should require vendors to meet
rigorous security standards,

10223
09:46:33.402 --> 09:46:35.723
<v Ashna Agarwal (AI Now Institute)>disclose breaches and outages promptly,

10224
09:46:35.723 --> 09:46:38.605
<v Ashna Agarwal (AI Now Institute)>and assume financial liability
when their systems fail.

10225
09:46:39.086 --> 09:46:42.766
<v Ashna Agarwal (AI Now Institute)>New York City still holds the leverage to
decide which systems enter its agencies

10226
09:46:42.766 --> 09:46:47.430
<v Ashna Agarwal (AI Now Institute)>and who bears the cost of failure. I urge
the council to use it now. Thank you.

10227
09:46:48.390 --> 09:46:50.710
<v CM Carmen De La Rosa>Thank you so much for your testimony.

10228
09:46:50.867 --> 09:46:53.590
<v CM Carmen De La Rosa>Up next, we have Tony Taiz Thoret.

10229
09:46:53.668 --> 09:46:56.470
<v CM Carmen De La Rosa>If you're online, please unmute
yourself and you may begin.

10230
09:46:58.234 --> 09:46:58.797
<v Tony Tye Serrat (Cinepartner)>Hi.

10231
09:46:59.195 --> 09:47:00.637
<v Tony Tye Serrat (Cinepartner)>Thank you so much.

10232
09:47:01.355 --> 09:47:02.715
<v Tony Tye Serrat (Cinepartner)>I have written remarks,

10233
09:47:02.875 --> 09:47:05.277
<v Tony Tye Serrat (Cinepartner)>but I'm gonna read them out
loud and I can send them later.

10234
09:47:05.676 --> 09:47:09.754
<v Tony Tye Serrat (Cinepartner)>But thank you, speaker Menon and members of
the council. My name is Tony Tye Serrat.

10235
09:47:09.754 --> 09:47:11.035
<v Tony Tye Serrat (Cinepartner)>I'm a filmmaker,

10236
09:47:11.035 --> 09:47:13.117
<v Tony Tye Serrat (Cinepartner)>educator, and founder of Cinepartner,

10237
09:47:13.390 --> 09:47:15.870
<v Tony Tye Serrat (Cinepartner)>an AI powered platform for storytellers.

10238
09:47:16.110 --> 09:47:19.470
<v Tony Tye Serrat (Cinepartner)>Now I'm probably coming to this conversation
from a slightly different perspective

10239
09:47:19.470 --> 09:47:22.750
<v Tony Tye Serrat (Cinepartner)>because I build with AI,
I teach people about AI,

10240
09:47:22.832 --> 09:47:25.710
<v Tony Tye Serrat (Cinepartner)>and I work with creative
communities that are simultaneously

10241
09:47:25.710 --> 09:47:27.310
<v Tony Tye Serrat (Cinepartner)>excited about this technology

10242
09:47:27.633 --> 09:47:30.996
<v Tony Tye Serrat (Cinepartner)>and very concerned about what it
can mean for their livelihoods,

10243
09:47:30.996 --> 09:47:32.914
<v Tony Tye Serrat (Cinepartner)>their work, and their agency.

10244
09:47:33.477 --> 09:47:35.477
<v Tony Tye Serrat (Cinepartner)>So we believe that we need guardrails,

10245
09:47:35.477 --> 09:47:39.633
<v Tony Tye Serrat (Cinepartner)>which is we... Something where we've
been hearing all day. Right? But

10246
09:47:39.797 --> 09:47:42.840
<v Tony Tye Serrat (Cinepartner)>I also believe that the goal
should be responsible innovation,

10247
09:47:42.918 --> 09:47:44.758
<v Tony Tye Serrat (Cinepartner)>not regulation that unintentionally

10248
09:47:44.758 --> 09:47:48.598
<v Tony Tye Serrat (Cinepartner)>makes it possible for only the
largest companies to build an AI.

10249
09:47:49.560 --> 09:47:53.880
<v Tony Tye Serrat (Cinepartner)>When we create compliance requirements, I
would encourage the council to not to not

10250
09:47:53.880 --> 09:47:57.156
<v Tony Tye Serrat (Cinepartner)>only ask, will this
make AI safer, but also

10251
09:47:57.316 --> 09:47:59.637
<v Tony Tye Serrat (Cinepartner)>who will be able to afford to comply?

10252
09:48:00.035 --> 09:48:03.395
<v Tony Tye Serrat (Cinepartner)>Because if responsible AI
development requires an expensive

10253
09:48:03.477 --> 09:48:04.355
<v Tony Tye Serrat (Cinepartner)>validation

10254
09:48:04.355 --> 09:48:05.797
<v Tony Tye Serrat (Cinepartner)>regimen...

10255
09:48:05.875 --> 09:48:09.234
<v Tony Tye Serrat (Cinepartner)>I mean, sorry, regime that only trillion
dollar technology companies can navigate,

10256
09:48:09.777 --> 09:48:11.380
<v Tony Tye Serrat (Cinepartner)>we may inadvertently

10257
09:48:11.540 --> 09:48:13.380
<v Tony Tye Serrat (Cinepartner)>consolidate AI development

10258
09:48:13.380 --> 09:48:16.340
<v Tony Tye Serrat (Cinepartner)>and the very companies we're
trying to hold accountable.

10259
09:48:17.220 --> 09:48:21.300
<v Tony Tye Serrat (Cinepartner)>And for creative communities, safety isn't
only just about whether a model can be

10260
09:48:21.300 --> 09:48:23.540
<v Tony Tye Serrat (Cinepartner)>shut down. It's also about consent,

10261
09:48:24.254 --> 09:48:25.137
<v Tony Tye Serrat (Cinepartner)>authorship,

10262
09:48:25.137 --> 09:48:26.016
<v Tony Tye Serrat (Cinepartner)>provenance,

10263
09:48:26.016 --> 09:48:26.895
<v Tony Tye Serrat (Cinepartner)>compensation,

10264
09:48:26.895 --> 09:48:28.016
<v Tony Tye Serrat (Cinepartner)>and transparency.

10265
09:48:28.574 --> 09:48:32.496
<v Tony Tye Serrat (Cinepartner)>So if someone's screenplay, for instance,
or voice, likeness, or creative work

10266
09:48:32.496 --> 09:48:35.297
<v Tony Tye Serrat (Cinepartner)>contributes to an AI system or its output,

10267
09:48:35.457 --> 09:48:37.535
<v Tony Tye Serrat (Cinepartner)>what rights do that per...
Does that person retain?

10268
09:48:38.258 --> 09:48:41.860
<v Tony Tye Serrat (Cinepartner)>Do they know how their work is
being used? Can they opt out?

10269
09:48:42.180 --> 09:48:46.020
<v Tony Tye Serrat (Cinepartner)>Can they identify when AI materially
contributed to something?

10270
09:48:47.380 --> 09:48:50.100
<v Tony Tye Serrat (Cinepartner)>And there's one more thing
I wanted to elaborate

10271
09:48:50.418 --> 09:48:53.220
<v Tony Tye Serrat (Cinepartner)>is that I strongly support
tracking displacement.

10272
09:48:53.785 --> 09:48:57.785
<v Tony Tye Serrat (Cinepartner)>But I would encourage New York to stop
at counting the jobs AI eliminates.

10273
09:48:58.023 --> 09:49:01.703
<v Tony Tye Serrat (Cinepartner)>We should also be measuring whether New
Yorkers are being prepared for the the jobs

10274
09:49:01.703 --> 09:49:02.906
<v Tony Tye Serrat (Cinepartner)>that AI creates.

10275
09:49:03.227 --> 09:49:04.426
<v Tony Tye Serrat (Cinepartner)>AI literacy...

10276
09:49:04.504 --> 09:49:05.547
<v Committee Counsel>His time has expired.

10277
09:49:06.360 --> 09:49:07.562
<v Tony Tye Serrat (Cinepartner)>Oh, okay.

10278
09:49:07.800 --> 09:49:11.883
<v CM Carmen De La Rosa>Thank you. Thank you for your testimony.
And please be sure to send us the written

10279
09:49:11.883 --> 09:49:15.723
<v CM Carmen De La Rosa>copy so we can review it. Thank you so
much. So where do I send it again? I didn't

10280
09:49:15.723 --> 09:49:19.402
<v CM Carmen De La Rosa>hear that. Testimony@council.nyc.gov.

10281
09:49:20.043 --> 09:49:24.734
<v CM Carmen De La Rosa>Okay. Thank you. Thank
you. Up next, Travis Hall.

10282
09:49:24.734 --> 09:49:27.215
<v CM Carmen De La Rosa>If you're online, please unmute yourself

10283
09:49:27.215 --> 09:49:28.254
<v CM Carmen De La Rosa>and begin.

10284
09:49:34.336 --> 09:49:35.137
<v Committee Counsel>You may begin.

10285
09:49:37.950 --> 09:49:40.350
<v Travis Hall (Center for Democracy and Technology)>Alright. Thank you so
much. Can you hear me?

10286
09:49:40.750 --> 09:49:41.550
<v Travis Hall (Center for Democracy and Technology)>Yes.

10287
09:49:41.550 --> 09:49:46.110
<v Travis Hall (Center for Democracy and Technology)>My name is my name is doctor Travis Hall.
I represent the Center for Democracy and

10288
09:49:46.110 --> 09:49:48.990
<v Travis Hall (Center for Democracy and Technology)>Technology, a thirty year old organization

10289
09:49:48.990 --> 09:49:51.246
<v Travis Hall (Center for Democracy and Technology)>dedicated to digital rights

10290
09:49:51.246 --> 09:49:53.168
<v Travis Hall (Center for Democracy and Technology)>and civil liberties online.

10291
09:49:54.527 --> 09:49:58.207
<v Travis Hall (Center for Democracy and Technology)>And today, I would love to talk to you a
little bit about the range of issues that

10292
09:49:58.207 --> 09:50:02.285
<v Travis Hall (Center for Democracy and Technology)>you're covering. I really appreciate the
hearing. I want specifically to talk about

10293
09:50:02.285 --> 09:50:03.484
<v Travis Hall (Center for Democracy and Technology)>third party auditing.

10294
09:50:04.570 --> 09:50:05.770
<v Travis Hall (Center for Democracy and Technology)>We appreciate

10295
09:50:06.090 --> 09:50:09.610
<v Travis Hall (Center for Democracy and Technology)>the effort to put in place third
party auditing requirements.

10296
09:50:09.770 --> 09:50:14.008
<v Travis Hall (Center for Democracy and Technology)>They not not only will help to deal with
catastrophic risks, but the range of risks

10297
09:50:14.008 --> 09:50:16.570
<v Travis Hall (Center for Democracy and Technology)>that you... That people face
from artificial intelligence.

10298
09:50:17.023 --> 09:50:20.066
<v Travis Hall (Center for Democracy and Technology)>However, as we've seen
with LL one forty four,

10299
09:50:20.465 --> 09:50:23.266
<v Travis Hall (Center for Democracy and Technology)>if there are not additional mechanisms

10300
09:50:23.906 --> 09:50:25.426
<v Travis Hall (Center for Democracy and Technology)>in place to incentivize,

10301
09:50:25.664 --> 09:50:29.906
<v Travis Hall (Center for Democracy and Technology)>the use and, consequences for
these audits, they end up turning,

10302
09:50:30.465 --> 09:50:35.270
<v Travis Hall (Center for Democracy and Technology)>into being somewhat useless. So we
wanna see, meaningful independence,

10303
09:50:35.670 --> 09:50:36.310
<v Travis Hall (Center for Democracy and Technology)>independent,

10304
09:50:36.550 --> 09:50:37.430
<v Travis Hall (Center for Democracy and Technology)>interdisciplinary

10305
09:50:37.510 --> 09:50:41.832
<v Travis Hall (Center for Democracy and Technology)>skills. There are a lot of, regulatory
agencies already covering many of these

10306
09:50:41.832 --> 09:50:42.390
<v Travis Hall (Center for Democracy and Technology)>risks,

10307
09:50:42.710 --> 09:50:45.510
<v Travis Hall (Center for Democracy and Technology)>and they can be drawn on,
not just technical skills.

10308
09:50:47.016 --> 09:50:51.098
<v Travis Hall (Center for Democracy and Technology)>Don't just simply, audit the the
models, thinking about auditing,

10309
09:50:51.258 --> 09:50:53.816
<v Travis Hall (Center for Democracy and Technology)>and assessing systems and use cases.

10310
09:50:54.375 --> 09:50:58.055
<v Travis Hall (Center for Democracy and Technology)>Make sure that you cover a broader scale
of harms. We heard a lot today about

10311
09:50:58.055 --> 09:51:01.410
<v Travis Hall (Center for Democracy and Technology)>catastrophic harms, but we're
already seeing little catastrophe...

10312
09:51:01.410 --> 09:51:05.332
<v Travis Hall (Center for Democracy and Technology)>Little catastrophes everywhere. When people
lose their benefits or their jobs, those

10313
09:51:05.332 --> 09:51:06.690
<v Travis Hall (Center for Democracy and Technology)>are catastrophes.

10314
09:51:07.890 --> 09:51:09.492
<v Travis Hall (Center for Democracy and Technology)>Make sure to incentivize.

10315
09:51:09.973 --> 09:51:12.930
<v Travis Hall (Center for Democracy and Technology)>Incentives are a a...
Commensurate to protections

10316
09:51:13.305 --> 09:51:14.266
<v Travis Hall (Center for Democracy and Technology)>and assess...

10317
09:51:14.830 --> 09:51:17.867
<v Travis Hall (Center for Democracy and Technology)>And, for example, making sure
that assessments don't entrench

10318
09:51:18.188 --> 09:51:19.547
<v Travis Hall (Center for Democracy and Technology)>powerful actors.

10319
09:51:19.867 --> 09:51:20.746
<v Travis Hall (Center for Democracy and Technology)>Also, independent,

10320
09:51:21.465 --> 09:51:22.426
<v Travis Hall (Center for Democracy and Technology)>evaluation

10321
09:51:22.426 --> 09:51:26.586
<v Travis Hall (Center for Democracy and Technology)>only really works as one piece in
a broader accountability network.

10322
09:51:26.990 --> 09:51:30.880
<v Travis Hall (Center for Democracy and Technology)>So thank you very much for your time. Would
love to engage with the council on, this

10323
09:51:30.880 --> 09:51:31.918
<v Travis Hall (Center for Democracy and Technology)>and other issues.

10324
09:51:33.438 --> 09:51:34.880
<v CM Carmen De La Rosa>And thank you so much.

10325
09:51:35.360 --> 09:51:37.040
<v CM Carmen De La Rosa>Up next, we have

10326
09:51:37.277 --> 09:51:39.438
<v CM Carmen De La Rosa>Danielle Marie Chappies.

10327
09:51:39.598 --> 09:51:42.720
<v CM Carmen De La Rosa>If you're online, if you
could just unmute and begin.

10328
09:51:52.344 --> 09:51:53.223
<v CM Carmen De La Rosa>Daniel

10329
09:51:53.543 --> 09:51:54.824
<v CM Carmen De La Rosa>Marie Chappius.

10330
09:51:57.710 --> 09:52:00.350
<v CM Carmen De La Rosa>Alright. We'll go on to Michelle Barnaby.

10331
09:52:00.590 --> 09:52:03.633
<v CM Carmen De La Rosa>If you're online, please unmute and begin.

10332
09:52:07.550 --> 09:52:08.270
<v Michelle Barnaby (public)>My

10333
09:52:08.832 --> 09:52:10.270
<v Michelle Barnaby (public)>name is Michelle Barnaby.

10334
09:52:10.510 --> 09:52:13.836
<v Michelle Barnaby (public)>I'm a New York City nurse,
a Columbia graduate,

10335
09:52:14.234 --> 09:52:16.793
<v Michelle Barnaby (public)>and a foreman... Former NISNA member.

10336
09:52:17.434 --> 09:52:19.516
<v Michelle Barnaby (public)>Out of concern for the future

10337
09:52:19.676 --> 09:52:21.836
<v Michelle Barnaby (public)>of patient care and public health,

10338
09:52:22.074 --> 09:52:26.395
<v Michelle Barnaby (public)>I now study AI governance and
safety with Blue Dot Impact.

10339
09:52:26.953 --> 09:52:27.273
<v Michelle Barnaby (public)>Today,

10340
09:52:28.050 --> 09:52:31.170
<v Michelle Barnaby (public)>not one of the major
labs, Anthropic, OpenAI,

10341
09:52:31.170 --> 09:52:35.730
<v Michelle Barnaby (public)>Google, or Meta could quantify the
risk of catastrophic AI failure.

10342
09:52:36.050 --> 09:52:41.570
<v Michelle Barnaby (public)>In health care, informed consent would
apply. We disclose the risk and the patient

10343
09:52:41.570 --> 09:52:42.050
<v Michelle Barnaby (public)>decides.

10344
09:52:42.652 --> 09:52:47.535
<v Michelle Barnaby (public)>I can't imagine handing a patient a drug
or a device and saying this may carry a

10345
09:52:47.535 --> 09:52:51.613
<v Michelle Barnaby (public)>catastrophic risk to you or your
family's life, and we just don't know,

10346
09:52:51.934 --> 09:52:53.215
<v Michelle Barnaby (public)>but you should take it.

10347
09:52:54.016 --> 09:52:58.543
<v Michelle Barnaby (public)>Right now, the public is the patient, and
we are being denied the right to know the

10348
09:52:58.543 --> 09:53:00.620
<v Michelle Barnaby (public)>risks and the right to say no.

10349
09:53:02.062 --> 09:53:03.344
<v Michelle Barnaby (public)>Speaker Menon

10350
09:53:03.582 --> 09:53:07.420
<v Michelle Barnaby (public)>rightly noted that we don't ask the
airline industry to self regulate.

10351
09:53:07.832 --> 09:53:10.152
<v Michelle Barnaby (public)>In hospitals, we report our errors,

10352
09:53:10.395 --> 09:53:11.594
<v Michelle Barnaby (public)>even the near misses.

10353
09:53:12.074 --> 09:53:14.715
<v Michelle Barnaby (public)>AI companies face no such mandate.

10354
09:53:15.273 --> 09:53:19.355
<v Michelle Barnaby (public)>When you ask, if these Frontier
Labs bear legal responsibility

10355
09:53:19.434 --> 09:53:23.380
<v Michelle Barnaby (public)>for the harm their AI causes,
only Google answered clearly.

10356
09:53:23.700 --> 09:53:25.140
<v Michelle Barnaby (public)>The other three dodged.

10357
09:53:25.140 --> 09:53:28.660
<v Michelle Barnaby (public)>As a nurse whose entire profession
is built on accountability,

10358
09:53:28.900 --> 09:53:30.900
<v Michelle Barnaby (public)>I find that deeply unserious.

10359
09:53:31.460 --> 09:53:35.207
<v Michelle Barnaby (public)>Today, I heard the lab say
voluntary frameworks aren't enough.

10360
09:53:35.527 --> 09:53:40.168
<v Michelle Barnaby (public)>We can learn from what happened in Illinois
where third party audits won't begin

10361
09:53:40.168 --> 09:53:41.766
<v Michelle Barnaby (public)>until 2028.

10362
09:53:42.168 --> 09:53:44.650
<v Michelle Barnaby (public)>We don't have time for
slow implementation.

10363
09:53:44.970 --> 09:53:47.926
<v Michelle Barnaby (public)>I think New York can show that
government can move with speed

10364
09:53:48.610 --> 09:53:53.330
<v Michelle Barnaby (public)>that this demands, diagnose the problem,
and put safeguards in place quickly.

10365
09:53:53.730 --> 09:53:54.688
<v Michelle Barnaby (public)>Let's give

10366
09:53:54.930 --> 09:53:59.250
<v Michelle Barnaby (public)>these frontier labs the regulations they
say they want at a speed that makes sense

10367
09:53:59.250 --> 09:54:00.047
<v Michelle Barnaby (public)>for New Yorkers.

10368
09:54:00.684 --> 09:54:01.566
<v Michelle Barnaby (public)>Independent

10369
09:54:01.566 --> 09:54:03.645
<v Michelle Barnaby (public)>third party audits before deployment,

10370
09:54:03.805 --> 09:54:05.484
<v Michelle Barnaby (public)>including algorithmic bias,

10371
09:54:05.645 --> 09:54:06.367
<v Michelle Barnaby (public)>mandatory

10372
09:54:06.684 --> 09:54:09.004
<v Michelle Barnaby (public)>reporting for incidents and near misses,

10373
09:54:09.086 --> 09:54:09.805
<v Michelle Barnaby (public)>a required

10374
09:54:10.523 --> 09:54:13.820
<v Michelle Barnaby (public)>override switch, whistleblowers,
and clear development...

10375
09:54:14.062 --> 09:54:17.340
<v Michelle Barnaby (public)>Developer liability for
harm. Thank you so much. You

10376
09:54:17.742 --> 09:54:19.180
<v CM Carmen De La Rosa>so much for your testimony.

10377
09:54:20.620 --> 09:54:22.300
<v CM Carmen De La Rosa>Up next, we have

10378
09:54:23.260 --> 09:54:24.380
<v CM Carmen De La Rosa>Robert Cappell.

10379
09:54:27.816 --> 09:54:30.535
<v CM Carmen De La Rosa>Robert, if you're on, unmute and begin.

10380
09:54:38.535 --> 09:54:42.137
<v CM Carmen De La Rosa>Okay. I'm gonna call Dan...
Daniel Marie Chappius

10381
09:54:42.310 --> 09:54:46.390
<v CM Carmen De La Rosa>again. If you're online
and you can hear us, unmute

10382
09:54:47.027 --> 09:54:47.750
<v CM Carmen De La Rosa>and begin.

10383
09:54:55.707 --> 09:54:59.945
<v CM Carmen De La Rosa>Okay. Moving on. We will
call on Patrick Sullivan.

10384
09:55:00.188 --> 09:55:02.906
<v CM Carmen De La Rosa>Patrick, if you can hear
us, unmute and begin.

10385
09:55:12.480 --> 09:55:14.883
<v CM Carmen De La Rosa>We'll move on. Tony Jones,

10386
09:55:14.883 --> 09:55:17.760
<v CM Carmen De La Rosa>if you're on and can hear
us, unmute and begin.

10387
09:55:20.242 --> 09:55:21.043
<v Tony Jones (Creative AI Academy / Pratt Institute)>Hello.

10388
09:55:21.680 --> 09:55:25.840
<v Tony Jones (Creative AI Academy / Pratt Institute)>I'm here. Thank you. Good afternoon,
everybody... Or good evening, everybody. Thank

10389
09:55:25.840 --> 09:55:26.402
<v Tony Jones (Creative AI Academy / Pratt Institute)>you very much.

10390
09:55:27.535 --> 09:55:31.777
<v Tony Jones (Creative AI Academy / Pratt Institute)>I'm Tony Jones. I'm a Brooklyn resident
and cofounder of Creative AI Academy,

10391
09:55:32.418 --> 09:55:36.418
<v Tony Jones (Creative AI Academy / Pratt Institute)>a training company and instructor of AI
design at the Pratt Institute. I'm hopeful

10392
09:55:36.418 --> 09:55:38.176
<v Tony Jones (Creative AI Academy / Pratt Institute)>and heartened by the panelists'

10393
09:55:38.336 --> 09:55:41.215
<v Tony Jones (Creative AI Academy / Pratt Institute)>testimonies today and the intention of
these committees on this crucial moment.

10394
09:55:41.550 --> 09:55:45.950
<v Tony Jones (Creative AI Academy / Pratt Institute)>While I support the strong regulation of
companies building Frontier AI, I... And I

10395
09:55:45.950 --> 09:55:49.790
<v Tony Jones (Creative AI Academy / Pratt Institute)>see the proposed amendments aimed to address
this, I'm here about the other half of

10396
09:55:49.790 --> 09:55:53.870
<v Tony Jones (Creative AI Academy / Pratt Institute)>the problem, the person at the keyboard.
The harm most New Yorkers will meet first

10397
09:55:53.870 --> 09:55:57.945
<v Tony Jones (Creative AI Academy / Pratt Institute)>comes from someone using the tool already
in market. Today, anyone with an email

10398
09:55:57.945 --> 09:56:02.824
<v Tony Jones (Creative AI Academy / Pratt Institute)>address can clone a voice, fabricate a photo
of a real person, or write a convincing

10399
09:56:02.824 --> 09:56:06.504
<v Tony Jones (Creative AI Academy / Pratt Institute)>scam in minutes. The tool asks nothing
from them first, no training, no

10400
09:56:06.504 --> 09:56:08.664
<v Tony Jones (Creative AI Academy / Pratt Institute)>acknowledgment of risk, no
promise of responsibility.

10401
09:56:09.130 --> 09:56:10.570
<v Tony Jones (Creative AI Academy / Pratt Institute)>We regulate carmakers,

10402
09:56:10.570 --> 09:56:14.970
<v Tony Jones (Creative AI Academy / Pratt Institute)>and we also require every driver to follow
the rules of the road. In my classroom,

10403
09:56:14.970 --> 09:56:17.848
<v Tony Jones (Creative AI Academy / Pratt Institute)>every student writes and signs
a personal AI constitution,

10404
09:56:17.930 --> 09:56:21.610
<v Tony Jones (Creative AI Academy / Pratt Institute)>a short statement of how they will and will
not use the tools that I'm teaching them

10405
09:56:21.610 --> 09:56:25.836
<v Tony Jones (Creative AI Academy / Pratt Institute)>to use. I propose the city ask the same of
its citizens in the form of simpler than

10406
09:56:25.836 --> 09:56:30.555
<v Tony Jones (Creative AI Academy / Pratt Institute)>a license. First, a short free city provided
safety training taken once before a

10407
09:56:30.555 --> 09:56:33.598
<v Tony Jones (Creative AI Academy / Pratt Institute)>person can open an account
and an AI tool of a certain,

10408
09:56:33.996 --> 09:56:38.220
<v Tony Jones (Creative AI Academy / Pratt Institute)>defined level of capability.
Second, assigned user constitution,

10409
09:56:38.457 --> 09:56:42.860
<v Tony Jones (Creative AI Academy / Pratt Institute)>plain language commitment commitment to
use the AI honestly with no harm, language

10410
09:56:42.860 --> 09:56:46.220
<v Tony Jones (Creative AI Academy / Pratt Institute)>which users may incorporate into
the account level personalization

10411
09:56:46.297 --> 09:56:50.766
<v Tony Jones (Creative AI Academy / Pratt Institute)>of the instructions that... Of the AI
products they use as well as a second safety

10412
09:56:50.766 --> 09:56:54.684
<v Tony Jones (Creative AI Academy / Pratt Institute)>measure. The signature will not stop the
determined criminal and neither does a

10413
09:56:54.684 --> 09:56:58.445
<v Tony Jones (Creative AI Academy / Pratt Institute)>driver's license. It simply sets a baseline
for the millions of ordinary users who

10414
09:56:58.445 --> 09:57:02.670
<v Tony Jones (Creative AI Academy / Pratt Institute)>want to do right and have never been told
what that means. As a first step, I asked

10415
09:57:02.670 --> 09:57:06.750
<v Tony Jones (Creative AI Academy / Pratt Institute)>the council to start where its authority
is clear, require this training and the

10416
09:57:06.750 --> 09:57:11.310
<v Tony Jones (Creative AI Academy / Pratt Institute)>signed constitution for city employees
who use AI on the job or AI vendors that

10417
09:57:11.310 --> 09:57:15.550
<v Tony Jones (Creative AI Academy / Pratt Institute)>contract with the city, then maybe we can
press Albany for a statewide standard and

10418
09:57:15.550 --> 09:57:19.152
<v Tony Jones (Creative AI Academy / Pratt Institute)>set the standard for the rest of the country
as well. Thank you very much for your

10419
09:57:19.152 --> 09:57:21.312
<v Tony Jones (Creative AI Academy / Pratt Institute)>consideration. I know that we can regulate

10420
09:57:21.953 --> 09:57:27.152
<v Tony Jones (Creative AI Academy / Pratt Institute)>the labs, but I asked that we ask something
of the rest of us as well. Thank you so

10421
09:57:27.152 --> 09:57:31.130
<v Tony Jones (Creative AI Academy / Pratt Institute)>much. Thank you so much. Up next, Asata

10422
09:57:31.130 --> 09:57:32.090
<v Isata Kamara (Global Subnational Diplomacy Council)>Kamara,

10423
09:57:32.090 --> 09:57:35.370
<v Isata Kamara (Global Subnational Diplomacy Council)>if you're on, please unmute and begin.

10424
09:57:40.010 --> 09:57:41.210
<v Isata Kamara (Global Subnational Diplomacy Council)>Thank you so much,

10425
09:57:42.273 --> 09:57:45.555
<v Isata Kamara (Global Subnational Diplomacy Council)>madam speaker and members of the council.

10426
09:57:45.797 --> 09:57:49.477
<v Isata Kamara (Global Subnational Diplomacy Council)>My name is Isata Kamara,
and I am the founder

10427
09:57:49.637 --> 09:57:53.395
<v Isata Kamara (Global Subnational Diplomacy Council)>and the president of the Global
Subnational Diplomacy Council.

10428
09:57:53.637 --> 09:57:57.637
<v Isata Kamara (Global Subnational Diplomacy Council)>I previously served as New York City's
commissioner for international affairs

10429
09:57:57.887 --> 09:58:01.008
<v Isata Kamara (Global Subnational Diplomacy Council)>and had the pleasure of serving
the city for more than a decade.

10430
09:58:01.490 --> 09:58:05.330
<v Isata Kamara (Global Subnational Diplomacy Council)>I come to this testimony
because artificial intelligence

10431
09:58:05.490 --> 09:58:10.703
<v Isata Kamara (Global Subnational Diplomacy Council)>is something that all cities are thinking
about, including New York. And we know

10432
09:58:10.703 --> 09:58:14.625
<v Isata Kamara (Global Subnational Diplomacy Council)>that AI can obviously help
cities reduce their paperwork,

10433
09:58:14.863 --> 09:58:16.223
<v Isata Kamara (Global Subnational Diplomacy Council)>improve infrastructure,

10434
09:58:16.703 --> 09:58:19.906
<v Isata Kamara (Global Subnational Diplomacy Council)>and and planning, and make
services more accessible.

10435
09:58:20.383 --> 09:58:25.200
<v Isata Kamara (Global Subnational Diplomacy Council)>However, we also know that AI has
the potential to deepen inequalities

10436
09:58:25.200 --> 09:58:27.200
<v Isata Kamara (Global Subnational Diplomacy Council)>and threaten livelihoods.

10437
09:58:27.360 --> 09:58:29.680
<v Isata Kamara (Global Subnational Diplomacy Council)>So we are offering four recommendations

10438
09:58:29.680 --> 09:58:34.160
<v Isata Kamara (Global Subnational Diplomacy Council)>and are willing to work with city council
in order to support the work that you're

10439
09:58:34.160 --> 09:58:39.598
<v Isata Kamara (Global Subnational Diplomacy Council)>doing. First, we would like to ensure that
we're connecting workforce reporting to

10440
09:58:39.598 --> 09:58:40.605
<v Isata Kamara (Global Subnational Diplomacy Council)>paid training,

10441
09:58:41.246 --> 09:58:44.688
<v Isata Kamara (Global Subnational Diplomacy Council)>by also consulting with workers and unions

10442
09:58:44.848 --> 09:58:46.527
<v Isata Kamara (Global Subnational Diplomacy Council)>and providing opportunities

10443
09:58:46.527 --> 09:58:48.285
<v Isata Kamara (Global Subnational Diplomacy Council)>for advancement accordingly.

10444
09:58:48.766 --> 09:58:52.688
<v Isata Kamara (Global Subnational Diplomacy Council)>We are also asking that we do
more to protect democracies

10445
09:58:52.938 --> 09:58:58.620
<v Isata Kamara (Global Subnational Diplomacy Council)>as we know that artificial intelligence is
a great tool for spreading misinformation

10446
09:58:58.620 --> 09:58:59.980
<v Isata Kamara (Global Subnational Diplomacy Council)>and disinformation.

10447
09:59:00.700 --> 09:59:01.900
<v Isata Kamara (Global Subnational Diplomacy Council)>We specifically

10448
09:59:01.900 --> 09:59:04.860
<v Isata Kamara (Global Subnational Diplomacy Council)>are asking for more support to ensure

10449
09:59:04.938 --> 09:59:07.820
<v Isata Kamara (Global Subnational Diplomacy Council)>that residents have access to reliable

10450
09:59:07.895 --> 09:59:10.375
<v Isata Kamara (Global Subnational Diplomacy Council)>public information that is verified

10451
09:59:10.695 --> 09:59:12.695
<v Isata Kamara (Global Subnational Diplomacy Council)>and that they understand,

10452
09:59:13.258 --> 09:59:17.816
<v Isata Kamara (Global Subnational Diplomacy Council)>the different issues and decisions that
are being made by government in a clearer

10453
09:59:17.816 --> 09:59:18.457
<v Isata Kamara (Global Subnational Diplomacy Council)>way.

10454
09:59:18.777 --> 09:59:22.810
<v Isata Kamara (Global Subnational Diplomacy Council)>When you look around the world, there are
great cities that are doing this work,

10455
09:59:22.810 --> 09:59:24.090
<v Isata Kamara (Global Subnational Diplomacy Council)>including Helsinki

10456
09:59:24.090 --> 09:59:29.610
<v Isata Kamara (Global Subnational Diplomacy Council)>and others, and we ask the city council to
connect with those cities to learn from

10457
09:59:29.610 --> 09:59:30.810
<v Isata Kamara (Global Subnational Diplomacy Council)>their experiences

10458
09:59:30.810 --> 09:59:34.953
<v Isata Kamara (Global Subnational Diplomacy Council)>as well. As I've mentioned, we
believe that artificial intelligence

10459
09:59:34.953 --> 09:59:38.797
<v Isata Kamara (Global Subnational Diplomacy Council)>could be a tool for good if
it is managed carefully.

10460
09:59:39.516 --> 09:59:42.555
<v Isata Kamara (Global Subnational Diplomacy Council)>Also share the concern
that other speakers have

10461
09:59:42.797 --> 09:59:43.594
<v Isata Kamara (Global Subnational Diplomacy Council) [?]>shared the

10462
09:59:44.156 --> 09:59:45.516
<v CM Carmen De La Rosa>Oh, thank you.

10463
09:59:46.074 --> 09:59:47.996
<v CM Carmen De La Rosa>Thank you so much for your testimony.

10464
09:59:48.234 --> 09:59:50.395
<v CM Carmen De La Rosa>Up next, I'll call Ashley Youssef.

10465
09:59:54.440 --> 09:59:58.760
<v Ashley Youssef (Trina AI)>Thank you, speaker Menon, and members of
the council for the opportunity to speak.

10466
09:59:58.760 --> 10:00:02.523
<v Ashley Youssef (Trina AI)>My name is Ashley Youssef. I am a native
New Yorker, a mother of four living in

10467
10:00:02.523 --> 10:00:06.637
<v Ashley Youssef (Trina AI)>Michigan, and a leader of Trina
AI. I am developing Room 2050,

10468
10:00:06.637 --> 10:00:10.400
<v Ashley Youssef (Trina AI)>an immersive arts and technology learning
environment where students can explore

10469
10:00:10.400 --> 10:00:14.560
<v Ashley Youssef (Trina AI)>creativity through hands on experience
with art, robotics, and artificial

10470
10:00:14.560 --> 10:00:15.355
<v Ashley Youssef (Trina AI)>intelligence.

10471
10:00:15.836 --> 10:00:17.996
<v Ashley Youssef (Trina AI)>The vision is to put human imagination,

10472
10:00:18.390 --> 10:00:19.430
<v Ashley Youssef (Trina AI)>curiosity,

10473
10:00:19.430 --> 10:00:21.670
<v Ashley Youssef (Trina AI)>and judgment at the center
of that experience.

10474
10:00:21.832 --> 10:00:25.750
<v Ashley Youssef (Trina AI)>As a mother, I want children to have
opportunities to understand the technology

10475
10:00:25.750 --> 10:00:27.110
<v Ashley Youssef (Trina AI)>shaping their futures.

10476
10:00:27.430 --> 10:00:32.070
<v Ashley Youssef (Trina AI)>I want families to know who is responsible
when those technologies enter a learning

10477
10:00:32.070 --> 10:00:36.496
<v Ashley Youssef (Trina AI)>environment. My position is that access
and accountability must move together.

10478
10:00:36.656 --> 10:00:41.055
<v Ashley Youssef (Trina AI)>Children deserve opportunities to innovate,
and they deserve meaningful protection.

10479
10:00:41.055 --> 10:00:43.777
<v Ashley Youssef (Trina AI)>For Room 2050, the
standards I am proposing

10480
10:00:43.938 --> 10:00:44.740
<v Ashley Youssef (Trina AI)>include

10481
10:00:44.977 --> 10:00:46.656
<v Ashley Youssef (Trina AI)>educator supervision,

10482
10:00:46.656 --> 10:00:51.620
<v Ashley Youssef (Trina AI)>age appropriate tools, clear communication
with parents, and strict limits on the

10483
10:00:51.620 --> 10:00:52.500
<v Ashley Youssef (Trina AI)>collection

10484
10:00:52.660 --> 10:00:54.742
<v Ashley Youssef (Trina AI)>and use of student information.

10485
10:00:54.980 --> 10:00:57.140
<v Ashley Youssef (Trina AI)>Any use of student data to train

10486
10:00:57.620 --> 10:01:00.980
<v Ashley Youssef (Trina AI)>commercial AI systems
should require explicit,

10487
10:01:01.645 --> 10:01:02.445
<v Ashley Youssef (Trina AI)>informed

10488
10:01:02.445 --> 10:01:03.566
<v Ashley Youssef (Trina AI)>parental,

10489
10:01:03.887 --> 10:01:04.688
<v Ashley Youssef (Trina AI)>permission.

10490
10:01:04.926 --> 10:01:07.645
<v Ashley Youssef (Trina AI)>Students should learn that
AI can make mistakes,

10491
10:01:07.887 --> 10:01:09.246
<v Ashley Youssef (Trina AI)>reflect bias,

10492
10:01:09.246 --> 10:01:12.605
<v Ashley Youssef (Trina AI)>and produce convincing
information that is false.

10493
10:01:12.688 --> 10:01:14.285
<v Ashley Youssef (Trina AI)>They should practice questioning

10494
10:01:14.418 --> 10:01:17.460
<v Ashley Youssef (Trina AI)>its answers and developing their own ideas

10495
10:01:17.460 --> 10:01:21.940
<v Ashley Youssef (Trina AI)>no matter what part of the city or The
United States they are from to help support

10496
10:01:21.940 --> 10:01:23.380
<v Ashley Youssef (Trina AI)>emerging talent,

10497
10:01:23.540 --> 10:01:25.700
<v Ashley Youssef (Trina AI)>which are... A human emerging talent.

10498
10:01:25.780 --> 10:01:29.700
<v Ashley Youssef (Trina AI)>I support a gradual rollout
beginning with supervised experiences

10499
10:01:29.934 --> 10:01:33.215
<v Ashley Youssef (Trina AI)>and expanding only
after evaluating safety,

10500
10:01:33.215 --> 10:01:34.492
<v Ashley Youssef (Trina AI)>educational value,

10501
10:01:34.574 --> 10:01:37.695
<v Ashley Youssef (Trina AI)>and feedback from teachers
and families and the council.

10502
10:01:37.934 --> 10:01:40.652
<v Ashley Youssef (Trina AI)>I ask the council to
pursue clear standards

10503
10:01:40.652 --> 10:01:42.652
<v Ashley Youssef (Trina AI)>that schools can understand and enforce,

10504
10:01:43.130 --> 10:01:44.090
<v Ashley Youssef (Trina AI)>including

10505
10:01:44.090 --> 10:01:45.047
<v Ashley Youssef (Trina AI)>transparency

10506
10:01:45.047 --> 10:01:49.930
<v Ashley Youssef (Trina AI)>from technology providers and practical
ways to report problems and stop using

10507
10:01:50.330 --> 10:01:51.770
<v Ashley Youssef (Trina AI)>unsafe systems.

10508
10:01:52.008 --> 10:01:55.610
<v Ashley Youssef (Trina AI)>Educators and parents should have a
meaningful voice in these decisions.

10509
10:01:55.848 --> 10:01:57.770
<v Ashley Youssef (Trina AI)>That's what Toreen
and...... Time's expired.

10510
10:01:58.766 --> 10:02:02.047
<v CM Carmen De La Rosa>Thank you. Thank you so
much for your testimony.

10511
10:02:02.766 --> 10:02:03.965
<v CM Carmen De La Rosa>Up next,

10512
10:02:04.363 --> 10:02:06.766
<v CM Carmen De La Rosa>we will call on Naved Hassan.

10513
10:02:06.766 --> 10:02:09.406
<v CM Carmen De La Rosa>Naved, if you're online,
please unmute and begin.

10514
10:02:12.040 --> 10:02:16.040
<v Naveed Hassan (Panel for Educational Policy)>Thank you, speaker Menon, technology committee
chair Della Rosa and city council for

10515
10:02:16.040 --> 10:02:19.637
<v Naveed Hassan (Panel for Educational Policy)>holding this critical hearing. My name is
Naveed Hassan, and I'm the parent of two

10516
10:02:19.637 --> 10:02:22.918
<v Naveed Hassan (Panel for Educational Policy)>public school children and also the
elected Manhattan member on the panel for

10517
10:02:22.918 --> 10:02:23.797
<v Naveed Hassan (Panel for Educational Policy)>educational policy.

10518
10:02:24.164 --> 10:02:28.004
<v Naveed Hassan (Panel for Educational Policy)>My academic and professional background
is in symbolic, pregenerative AI and

10519
10:02:28.004 --> 10:02:28.883
<v Naveed Hassan (Panel for Educational Policy)>computer vision.

10520
10:02:29.203 --> 10:02:33.445
<v Naveed Hassan (Panel for Educational Policy)>We are in an unfortunate situation in
technology policy for generative AI with two

10521
10:02:33.445 --> 10:02:35.523
<v Naveed Hassan (Panel for Educational Policy)>prominent camps, the accelerationists

10522
10:02:35.523 --> 10:02:39.527
<v Naveed Hassan (Panel for Educational Policy)>appeal to a desire for imminent utopia
and the effect of altruists spending their

10523
10:02:39.527 --> 10:02:41.050
<v Naveed Hassan (Panel for Educational Policy)>time in, in code, panic.

10524
10:02:41.367 --> 10:02:45.450
<v Naveed Hassan (Panel for Educational Policy)>Neither extreme greed nor extreme fear
are beneficial framing for sober thinking

10525
10:02:45.450 --> 10:02:48.168
<v Naveed Hassan (Panel for Educational Policy)>around how to manage developing
advanced technologies.

10526
10:02:48.650 --> 10:02:52.730
<v Naveed Hassan (Panel for Educational Policy)>New Yorkers need prosocial technology
governance with tech expertise in public

10527
10:02:52.730 --> 10:02:53.207
<v Naveed Hassan (Panel for Educational Policy)>offices

10528
10:02:53.555 --> 10:02:58.035
<v Naveed Hassan (Panel for Educational Policy)>not being rushed to build hundreds of
hyperscale data centers, or putting only those

10529
10:02:58.035 --> 10:03:01.240
<v Naveed Hassan (Panel for Educational Policy)>very companies in the room when writing
rules in a way that lets them achieve

10530
10:03:01.240 --> 10:03:02.516
<v Naveed Hassan (Panel for Educational Policy)>regulatory capture.

10531
10:03:02.836 --> 10:03:06.637
<v Naveed Hassan (Panel for Educational Policy)>This morning, none of the Gen AI companies
could answer a simple question. What do

10532
10:03:06.637 --> 10:03:10.560
<v Naveed Hassan (Panel for Educational Policy)>you consider is the risk of your
service causing catastrophe to humanity?

10533
10:03:10.797 --> 10:03:14.560
<v Naveed Hassan (Panel for Educational Policy)>This is because it is antithetical to
their business desire to push an inherently

10534
10:03:14.560 --> 10:03:18.000
<v Naveed Hassan (Panel for Educational Policy)>probabilistic product to be
used in all aspects of work,

10535
10:03:18.637 --> 10:03:22.285
<v Naveed Hassan (Panel for Educational Policy)>to make any such estimates. When
a job demands deterministic

10536
10:03:22.285 --> 10:03:27.086
<v Naveed Hassan (Panel for Educational Policy)>certainty, this whole model will eventually
fear... Fail for sure. We must not allow

10537
10:03:27.086 --> 10:03:31.164
<v Naveed Hassan (Panel for Educational Policy)>the spread of an unfit for purpose
technology to be misused so broadly.

10538
10:03:31.805 --> 10:03:36.297
<v Naveed Hassan (Panel for Educational Policy)>There's bipartisan interest in not just
regulating Gen AI, but also pulling into the

10539
10:03:36.297 --> 10:03:40.860
<v Naveed Hassan (Panel for Educational Policy)>public realm broadband and wireless
connectivity, abusive algorithmic targeting,

10540
10:03:40.938 --> 10:03:43.977
<v Naveed Hassan (Panel for Educational Policy)>illegal surveillance, vendor
lock in, interoperability

10541
10:03:44.297 --> 10:03:45.098
<v Naveed Hassan (Panel for Educational Policy)>mandates,

10542
10:03:45.098 --> 10:03:49.035
<v Naveed Hassan (Panel for Educational Policy)>open standards, open source software
infrastructure, and more. We have the

10543
10:03:49.035 --> 10:03:52.477
<v Naveed Hassan (Panel for Educational Policy)>opportunity now to think about what municipal
tech policy should look like for the

10544
10:03:52.477 --> 10:03:56.715
<v Naveed Hassan (Panel for Educational Policy)>good of everyone living in New York. We have
the ability already to enforce existing

10545
10:03:56.715 --> 10:03:58.316
<v Naveed Hassan (Panel for Educational Policy)>rules around antitrust,

10546
10:03:58.316 --> 10:04:02.395
<v Naveed Hassan (Panel for Educational Policy)>concentration of risk, controlled goods
rules being applied to compute hardware.

10547
10:04:02.870 --> 10:04:06.870
<v Naveed Hassan (Panel for Educational Policy)>Let us calmly and expeditiously use skilled
public servants to build a future with

10548
10:04:06.870 --> 10:04:09.992
<v Naveed Hassan (Panel for Educational Policy)>beneficial technology
available for all of society.

10549
10:04:10.152 --> 10:04:10.793
<v Naveed Hassan (Panel for Educational Policy)>Thank you.

10550
10:04:11.430 --> 10:04:13.430
<v CM Carmen De La Rosa>Thank you so much for your testimony.

10551
10:04:14.152 --> 10:04:17.152
<v CM Carmen De La Rosa>Up next, we have Nadira Pittman. Nadira,

10552
10:04:17.152 --> 10:04:19.715
<v CM Carmen De La Rosa>if you are on, please unmute and begin.

10553
10:04:24.350 --> 10:04:25.152
<v Nadira Pittman (public)>Hello?

10554
10:04:25.793 --> 10:04:27.633
<v CM Carmen De La Rosa>Yes. Please begin your testimony. Hi.

10555
10:04:28.434 --> 10:04:34.160
<v Nadira Pittman (public)>Yes. Okay. So my name is Nadira Pittman.
I am 41 years old, a Queens resident.

10556
10:04:34.480 --> 10:04:38.480
<v Nadira Pittman (public)>I am actually reaching out
regarding a testimony that I have.

10557
10:04:39.440 --> 10:04:41.520
<v Nadira Pittman (public)>I am actually a survivor

10558
10:04:41.520 --> 10:04:43.280
<v Nadira Pittman (public)>of the AI attacks.

10559
10:04:44.293 --> 10:04:46.453
<v Nadira Pittman (public)>I'm dealing with facial recognition,

10560
10:04:47.496 --> 10:04:48.773
<v Nadira Pittman (public)>illegal surveillance.

10561
10:04:49.176 --> 10:04:50.773
<v Nadira Pittman (public)>I'm also dealing with,

10562
10:04:51.734 --> 10:04:54.055
<v Nadira Pittman (public)>like, constructed voice recordings

10563
10:04:54.055 --> 10:04:54.695
<v Nadira Pittman (public)>of

10564
10:04:54.855 --> 10:04:56.535
<v Nadira Pittman (public)>close family friends.

10565
10:04:56.535 --> 10:04:56.934
<v Nadira Pittman (public)>They're

10566
10:04:57.410 --> 10:05:02.130
<v Nadira Pittman (public)>gathering it and and creating a job.
Like, far as the job, they're using their

10567
10:05:02.130 --> 10:05:02.850
<v Nadira Pittman (public)>voices

10568
10:05:03.330 --> 10:05:04.370
<v Nadira Pittman (public)>against me.

10569
10:05:04.770 --> 10:05:05.890
<v Nadira Pittman (public)>It's on the...

10570
10:05:06.210 --> 10:05:10.770
<v Nadira Pittman (public)>If I'm on the buses, on the trains, it's
continuous. It's a continuous basis

10571
10:05:10.770 --> 10:05:11.650
<v Nadira Pittman (public)>nonstop.

10572
10:05:12.555 --> 10:05:15.516
<v Nadira Pittman (public)>I have been filing reports...

10573
10:05:15.914 --> 10:05:19.516
<v Nadira Pittman (public)>Police reports. I've been
trying to, you know, get this

10574
10:05:19.676 --> 10:05:22.793
<v Nadira Pittman (public)>attack off of me. I'm not sure why

10575
10:05:23.676 --> 10:05:27.800
<v Nadira Pittman (public)>it it it seems to be
inputted and okay. It's also

10576
10:05:28.040 --> 10:05:29.160
<v Nadira Pittman (public)>in various

10577
10:05:29.320 --> 10:05:30.200
<v Nadira Pittman (public)>agencies

10578
10:05:30.200 --> 10:05:32.200
<v Nadira Pittman (public)>where it's allowed to be used

10579
10:05:32.440 --> 10:05:34.120
<v Nadira Pittman (public)>against other individuals,

10580
10:05:34.200 --> 10:05:39.720
<v Nadira Pittman (public)>where people are afraid now of even
showing their faces on the... On public

10581
10:05:39.720 --> 10:05:40.360
<v Nadira Pittman (public)>transportation.

10582
10:05:41.547 --> 10:05:46.266
<v Nadira Pittman (public)>It... It's it's really, really getting out
of hand, and the attacks are are extreme.

10583
10:05:46.266 --> 10:05:48.266
<v Nadira Pittman (public)>It's very dangerous at this point.

10584
10:05:49.547 --> 10:05:53.227
<v Nadira Pittman (public)>You know, I I... It's it's to the point
where you can't go anywhere and feel safe

10585
10:05:53.227 --> 10:05:55.945
<v Nadira Pittman (public)>now because of of of this technology.

10586
10:05:57.020 --> 10:06:00.460
<v Nadira Pittman (public)>So I'm reaching out. I have sent numerous

10587
10:06:00.460 --> 10:06:01.260
<v Nadira Pittman (public)>reports.

10588
10:06:01.260 --> 10:06:02.940
<v Nadira Pittman (public)>I have reached out and

10589
10:06:03.260 --> 10:06:04.780
<v Nadira Pittman (public)>wrote letters to you guys

10590
10:06:05.340 --> 10:06:10.203
<v Nadira Pittman (public)>to try to get this off of me because it
it... It's it's something that it it... It's

10591
10:06:10.203 --> 10:06:11.566
<v Nadira Pittman (public)>it's unnecessary.

10592
10:06:11.566 --> 10:06:13.324
<v Nadira Pittman (public)>It's not... I'm not a criminal.

10593
10:06:13.566 --> 10:06:18.363
<v Nadira Pittman (public)>I don't do anything illegal. I... There's
no reason why the... Who put this on me?

10594
10:06:18.363 --> 10:06:21.164
<v Nadira Pittman (public)>Why is this... Why am I being
attacked in this manner?

10595
10:06:22.125 --> 10:06:24.742
<v Nadira Pittman (public)>You know, I've came up there personally,

10596
10:06:24.820 --> 10:06:27.780
<v Nadira Pittman (public)>and I informed you guys last
year... Your time has expired.

10597
10:06:28.980 --> 10:06:33.140
<v Nadira Pittman (public)>You so much for your time. I appreciate
your time. I would prefer someone reach out

10598
10:06:33.140 --> 10:06:35.780
<v Nadira Pittman (public)>to me as soon as possible....
So much. Thank you.

10599
10:06:36.100 --> 10:06:37.300
<v CM Carmen De La Rosa>Up next, we have

10600
10:06:38.504 --> 10:06:40.344
<v CM Carmen De La Rosa>Lori Lomask.

10601
10:06:40.504 --> 10:06:43.465
<v CM Carmen De La Rosa>If you're on, please,
unmute yourself and begin.

10602
10:06:52.527 --> 10:06:53.330
<v CM Carmen De La Rosa>Okay.

10603
10:06:53.970 --> 10:06:55.330
<v CM Carmen De La Rosa>We'll

10604
10:06:55.330 --> 10:06:56.207
<v CM Carmen De La Rosa>move on

10605
10:06:56.450 --> 10:06:58.450
<v CM Carmen De La Rosa>to Steve Cerre.

10606
10:06:59.566 --> 10:07:00.450
<v CM Carmen De La Rosa>Cerre?

10607
10:07:02.527 --> 10:07:03.168
<v CM Carmen De La Rosa>Not

10608
10:07:03.650 --> 10:07:04.688
<v CM Carmen De La Rosa>that Cerre.

10609
10:07:05.810 --> 10:07:06.848
<v CM Carmen De La Rosa>Steve

10610
10:07:07.234 --> 10:07:08.195
<v CM Carmen De La Rosa>Cerre,

10611
10:07:08.195 --> 10:07:12.438
<v CM Carmen De La Rosa>c e r e. Just go through the
list. They probably know why.

10612
10:07:12.996 --> 10:07:14.355
<v CM Carmen De La Rosa>Tan... Tanbir

10613
10:07:14.355 --> 10:07:15.316
<v CM Carmen De La Rosa>Chaudhry,

10614
10:07:15.957 --> 10:07:16.996
<v CM Carmen De La Rosa>if you're on.

10615
10:07:17.797 --> 10:07:18.516
<v CM Carmen De La Rosa>No.

10616
10:07:19.637 --> 10:07:20.676
<v CM Carmen De La Rosa>Catherine Jin.

10617
10:07:22.380 --> 10:07:23.180
<v CM Carmen De La Rosa>No.

10618
10:07:23.500 --> 10:07:25.020
<v CM Carmen De La Rosa>Zachary Thompson?

10619
10:07:25.820 --> 10:07:26.380
<v CM Carmen De La Rosa>No.

10620
10:07:26.860 --> 10:07:28.860
<v CM Carmen De La Rosa>Christopher Leon Johnson?

10621
10:07:30.620 --> 10:07:31.420
<v CM Carmen De La Rosa>No.

10622
10:07:32.140 --> 10:07:33.582
<v CM Carmen De La Rosa>Raj Sharof?

10623
10:07:37.293 --> 10:07:42.254
<v CM Carmen De La Rosa>Alright. Those are all the names that we
have for today. I wanna thank everyone

10624
10:07:42.414 --> 10:07:45.773
<v CM Carmen De La Rosa>who took time out of their
day today to testify,

10625
10:07:46.094 --> 10:07:47.695
<v CM Carmen De La Rosa>to bring your suggestions,

10626
10:07:47.850 --> 10:07:49.375
<v CM Carmen De La Rosa>to bring your questions,

10627
10:07:49.453 --> 10:07:53.040
<v CM Carmen De La Rosa>and to bring a partnership with the council.
I wanna thank speaker Menon for her

10628
10:07:53.040 --> 10:07:54.640
<v CM Carmen De La Rosa>leadership and to my colleagues,

10629
10:07:54.800 --> 10:07:57.360
<v CM Carmen De La Rosa>the last lady standing, council members

10630
10:07:57.520 --> 10:08:02.000
<v CM Carmen De La Rosa>Maloney and Hanif, thank you for staying
till the very end. And to all the staff who

10631
10:08:02.000 --> 10:08:03.120
<v CM Carmen De La Rosa>made this possible,

10632
10:08:03.200 --> 10:08:06.000
<v CM Carmen De La Rosa>thank you all for this important hearing,

10633
10:08:06.480 --> 10:08:08.098
<v CM Carmen De La Rosa>and we are adjourned.