1
00:26:16,160 --> 00:26:21,517
I thank Mr. Benj for his briefing.
I now give the floor to Mr. Sam Alman.

2
00:26:21,600 --> 00:26:25,437
>> Thank you, Mr. President and your
excellencies. I'm honored to be here and

3
00:26:25,520 --> 00:26:29,197
to be able to speak with you all at this
important time. We've been talking about

4
00:26:29,280 --> 00:26:33,277
artificial intelligence for years,
but it feels different in recent weeks and

5
00:26:33,360 --> 00:26:37,597
months. Rapid model progress has made
the timeline feel more compressed, the

6
00:26:37,680 --> 00:26:39,277
upside more tangible, but also the

7
00:26:39,360 --> 00:26:42,317
stakes and the risks more immediate. We

8
00:26:42,400 --> 00:26:45,197
have a choice in front of us. AI can

9
00:26:45,280 --> 00:26:49,437
either be more like a new renaissance
of creativity and discovery or more like a

10
00:26:49,520 --> 00:26:51,437
new industrial revolution of upheaval

11
00:26:51,520 --> 00:26:54,157
and disarray. The renaissance was a time

12
00:26:54,240 --> 00:26:56,877
when new tools, new ideas, and new

13
00:26:56,960 --> 00:26:58,557
institutions expanded what people

14
00:26:58,640 --> 00:27:01,517
believed they could do. AI can play that

15
00:27:01,600 --> 00:27:05,917
role for our time. I believe that people
are much more capable than they imagine

16
00:27:06,000 --> 00:27:09,837
themselves to be, but have been held
back by limited technology that was

17
00:27:09,920 --> 00:27:14,477
meant to empower but instead
distracts and often steals attention.

18
00:27:14,560 --> 00:27:18,797
The best version of AI is not about
making people cogs in a giant machine or

19
00:27:18,880 --> 00:27:23,037
optimizing every part of life until
the human parts disappear. It is about

20
00:27:23,120 --> 00:27:26,717
giving people more agency in an ever
more complex world. More ability to

21
00:27:26,800 --> 00:27:29,357
learn, to create, discover, build,

22
00:27:29,440 --> 00:27:31,677
participate. There are many things that

23
00:27:31,760 --> 00:27:36,397
AI cannot and should not automate.
AI should be built for people, not simply

24
00:27:36,480 --> 00:27:41,197
to turn the crank crank of the
machines faster for their own sake.

25
00:27:41,280 --> 00:27:44,317
Some of the things people working to build
AI have said are so dystopian that

26
00:27:44,400 --> 00:27:45,917
they sound like the plot of bad science

27
00:27:46,000 --> 00:27:48,717
fiction movies. Anything this power is

28
00:27:48,800 --> 00:27:53,517
also powerful is also intimidating.
This moment is complicated as we feel both

29
00:27:53,600 --> 00:27:58,797
tremendous potential and very
understandable anxiety at the same time.

30
00:27:58,880 --> 00:28:02,077
These models can help people discover
new knowledge, new medicines, better

31
00:28:02,160 --> 00:28:05,757
healthare, stronger education, and build
more productive economies. They can be

32
00:28:05,840 --> 00:28:09,597
new tools for people to pursue dreams
and ambition that today feel out of

33
00:28:09,680 --> 00:28:13,517
reach. The complexity of life has
increased so dramatically for the

34
00:28:13,600 --> 00:28:18,157
average person. AI can be a defensive
tool against this complexity as much as

35
00:28:18,240 --> 00:28:20,237
it can be this tool of unprecedented

36
00:28:20,320 --> 00:28:23,037
discovery and innovation.

37
00:28:23,120 --> 00:28:25,597
On the other hand, as AI systems become

38
00:28:25,680 --> 00:28:28,157
more capable and more autonomous, they

39
00:28:28,240 --> 00:28:30,237
could move faster than our institutions,

40
00:28:30,320 --> 00:28:33,037
concentrate power in two too few hands,

41
00:28:33,120 --> 00:28:38,077
or make decisions that people
no longer understand or control.

42
00:28:38,160 --> 00:28:41,597
That concern becomes especially
important as we approach systems that

43
00:28:41,680 --> 00:28:45,437
can improve themselves and future
versions of themselves, often called

44
00:28:45,520 --> 00:28:49,677
recursive self-improvement.
As the process of building AI becomes

45
00:28:49,760 --> 00:28:51,997
more automated, the pace of AI progress

46
00:28:52,080 --> 00:28:54,397
could accelerate rapidly. This moment

47
00:28:54,480 --> 00:28:59,917
calls for extreme care.
I largely agree with Professor Benjio uh

48
00:29:00,000 --> 00:29:01,277
although I'll say two instead of three.

49
00:29:01,360 --> 00:29:04,957
I believe there are two ways that AI
progress could go very badly that we

50
00:29:05,040 --> 00:29:08,077
have to avoid. First, we could lose

51
00:29:08,160 --> 00:29:12,797
control of the future to AI.
The risk is that it moves so fast that

52
00:29:12,880 --> 00:29:16,557
people can no longer follow what's
happening or intervene when needed. This

53
00:29:16,640 --> 00:29:20,077
would obviously be terrible.
The industry must not accept too much

54
00:29:20,160 --> 00:29:23,357
technological risk just because
the benefits are too great and that they

55
00:29:23,440 --> 00:29:27,437
feel too important to slow down. Beating
companies in competitive race is not a

56
00:29:27,520 --> 00:29:30,557
reason to make rash decisions. Nor do we
believe we are locked in a race where we

57
00:29:30,640 --> 00:29:33,837
are unable to do that. We have
unilaterally slowed down in the past. We

58
00:29:33,920 --> 00:29:36,717
will do so in the future. We do not
want to build systems. We think it's

59
00:29:36,800 --> 00:29:40,237
important that everyone avoids building
systems where we cannot make the

60
00:29:40,320 --> 00:29:44,397
alignment, monitor, and safety
guarantees that we must. We need to

61
00:29:44,480 --> 00:29:47,517
understand what these systems are doing
and have strong evidence that they will

62
00:29:47,600 --> 00:29:51,917
do what people intend even as they get
very very smart. Actually, especially as

63
00:29:52,000 --> 00:29:54,397
they get very very smart. It doesn't

64
00:29:54,480 --> 00:29:59,357
matter whether people put the risk
of catastrophe at 10% or 1% or 12% or.1%.

65
00:29:59,440 --> 00:30:02,637
None of these levels are remotely
acceptable and we should not train

66
00:30:02,720 --> 00:30:06,077
models that we cannot make an extremely
strong case that we'll be able to keep

67
00:30:06,160 --> 00:30:11,197
under human control. So that's one
way things can go wrong. In the other

68
00:30:11,280 --> 00:30:14,157
direction, these systems could
concentrate too much power in too few

69
00:30:14,240 --> 00:30:17,117
hands. No one person or company or

70
00:30:17,200 --> 00:30:19,037
country should be able to use the most

71
00:30:19,120 --> 00:30:24,397
powerful AI models to impose their
worldview on everyone else. A company or

72
00:30:24,480 --> 00:30:27,597
country that believes only it can be
trusted with this technology can use

73
00:30:27,680 --> 00:30:29,277
that belief to justify almost anything

74
00:30:29,360 --> 00:30:31,837
else. We have to reject that logic even

75
00:30:31,920 --> 00:30:34,637
when it's convenient for any one actor.

76
00:30:34,720 --> 00:30:39,517
I believe there is a pragmatic path
through both of these. At OpenAI, we

77
00:30:39,600 --> 00:30:43,677
approach these questions with a clear
goal. AI should give people more power

78
00:30:43,760 --> 00:30:47,197
over their own lives, more freedom
to live lives the way they want, more

79
00:30:47,280 --> 00:30:51,677
ability to solve problems, big or small,
and to improve their communities. We

80
00:30:51,760 --> 00:30:54,797
don't want to fall into the trap
of blind optimism. We also don't want to

81
00:30:54,880 --> 00:30:57,037
fall into the trap of dumerism either.

82
00:30:57,120 --> 00:31:00,877
Getting pulled into either extreme will
not help us successfully figure out the

83
00:31:00,960 --> 00:31:04,797
challenges in front of us. Especially
at a time when the world and its people

84
00:31:04,880 --> 00:31:09,357
face such enormous challenges that this
technology can help solve. So we try to

85
00:31:09,440 --> 00:31:10,877
walk the middle path between these two

86
00:31:10,960 --> 00:31:13,357
poles. For us, that means three

87
00:31:13,440 --> 00:31:17,117
principles. First, as AI becomes more

88
00:31:17,200 --> 00:31:19,197
capable, people must remain at the

89
00:31:19,280 --> 00:31:24,317
center of AI decision-m alignment is not
an abstract research question. It is the

90
00:31:24,400 --> 00:31:28,237
work of ensuring that these systems
reliably remain under human control,

91
00:31:28,320 --> 00:31:32,797
reflect human values, and help people
guide the next stages of development. We

92
00:31:32,880 --> 00:31:37,277
are not trying to and must not
automate human judgment or human values.

93
00:31:37,360 --> 00:31:41,517
Second, the benefits of scientific
progress and economic growth must be by

94
00:31:41,600 --> 00:31:45,277
people and for people. AI should help
researchers make discoveries, doctors

95
00:31:45,360 --> 00:31:48,637
treat patients, teachers teach students,
entrepreneurs start companies,

96
00:31:48,720 --> 00:31:52,557
communities solve problems that felt out
of reach, and much more. The continuous

97
00:31:52,640 --> 00:31:57,197
story of human progress is not the tools
replace human agency. It is that the

98
00:31:57,280 --> 00:31:58,957
right tools make people more capable.

99
00:31:59,040 --> 00:32:03,517
And I believe this is going to go much
further than we believe possible. Third,

100
00:32:03,600 --> 00:32:07,677
this technology must empower people
individually. A great future will come

101
00:32:07,760 --> 00:32:11,437
from people realizing their potential
and building value for themselves, their

102
00:32:11,520 --> 00:32:15,437
communities, their countries, and the
broader world. We might be the builders,

103
00:32:15,520 --> 00:32:18,717
one of the builders of this technology,
but we are not the heroes of the story.

104
00:32:18,800 --> 00:32:20,717
Our role is to enable people all over

105
00:32:20,800 --> 00:32:22,877
the world and to trust in the magic of

106
00:32:22,960 --> 00:32:26,477
humanity's skill and faculties.

107
00:32:26,560 --> 00:32:29,357
We now have models capable of discovering
knowledge that we have never

108
00:32:29,440 --> 00:32:31,837
had before. Consider the progress we've

109
00:32:31,920 --> 00:32:37,037
seen in math. Three summers ago,
our models were okay at grade school math.

110
00:32:37,120 --> 00:32:41,277
Two summers ago, they were pretty good
at high school math competitions. Last

111
00:32:41,360 --> 00:32:44,557
summer, we reached gold a gold
level medal in the most prestigious

112
00:32:44,640 --> 00:32:49,037
international math competition. And just
a few weeks ago, this summer, one of our

113
00:32:49,120 --> 00:32:53,517
models solved one of the Millennium Prize
problems, the Navia Stokes equations.

114
00:32:53,600 --> 00:32:56,957
These equations are used for aircraft
design, weather forecasting, and the

115
00:32:57,040 --> 00:33:00,637
study of blood flow. We are not solving
math problems for their own sake, but to

116
00:33:00,720 --> 00:33:04,637
empower people to discover new knowledge
and to use it to improve healthcare,

117
00:33:04,720 --> 00:33:08,637
raise standards of living of living,
and expand everyone's potential around the

118
00:33:08,720 --> 00:33:12,077
world. There remains so
much to be discovered,

119
00:33:12,160 --> 00:33:16,317
but it's very important that companies
not substitute for the democratic process.

120
00:33:16,400 --> 00:33:20,077
If AI is to be democratic, the most
important decisions cannot be made by

121
00:33:20,160 --> 00:33:24,557
labs in San Francisco alone. They must
be shaped through democratic processes

122
00:33:24,640 --> 00:33:28,157
and by governments accountable
to the people that they serve.

123
00:33:28,240 --> 00:33:32,077
At the international level,
this will require cooperation.

124
00:33:32,160 --> 00:33:35,357
In our history, there have been times
where countries who compete and don't

125
00:33:35,440 --> 00:33:38,957
always like each other very much still
come together for shared interests and

126
00:33:39,040 --> 00:33:43,117
the collective good in the face of a
powerful new technology. We believe this

127
00:33:43,200 --> 00:33:47,597
must be one of those times.
We need a mechanism for complimentary

128
00:33:47,680 --> 00:33:51,677
national and international frontier
AI standards. Standards for measuring

129
00:33:51,760 --> 00:33:55,517
capabilities, assessing risks,
determining whether safeguards are

130
00:33:55,600 --> 00:33:59,597
sufficient, and preserving meaningful
human oversight as systems become more

131
00:33:59,680 --> 00:34:02,797
autonomous. We need common
standards so countries

132
00:34:02,880 --> 00:34:07,037
can compare evidence, verify compliance,
and have a shared language and

133
00:34:07,120 --> 00:34:09,437
understanding about what is happening.

134
00:34:09,520 --> 00:34:13,357
We need accurate and speedy incident
reporting, classification and reporting

135
00:34:13,440 --> 00:34:18,477
protocols so the world can learn from
failures before they become catastrophes.

136
00:34:18,560 --> 00:34:21,757
And we need secure channels among
governments, critical infrastructure

137
00:34:21,840 --> 00:34:25,437
operators and technical experts
to share emerging vulnerabilities and new

138
00:34:25,520 --> 00:34:28,797
threats. Obviously,
these standards should not

139
00:34:28,880 --> 00:34:33,197
lock in incumbents or favor one business
model over another. They must support

140
00:34:33,280 --> 00:34:37,277
open and closed model developers, new
entrance and established labs, national

141
00:34:37,360 --> 00:34:40,877
authority, and international cooperation.
Each government should

142
00:34:40,960 --> 00:34:44,557
decide how to incorporate standards into
its own legal system, but we will all be

143
00:34:44,640 --> 00:34:48,397
better off if we can agree on what
good evidence, good safeguards, and good

144
00:34:48,480 --> 00:34:51,517
oversight look like on the global stage.

145
00:34:51,600 --> 00:34:55,357
We do not have all the answers,
and we're very open to any better ideas.

146
00:34:55,440 --> 00:34:58,397
We're ready to work with you all and try
our hardest to figure out the path

147
00:34:58,480 --> 00:35:02,237
forward. We are at a crossroads.

148
00:35:02,320 --> 00:35:06,717
One path leads to a world where power
is concentrated, decisions are opaque, and

149
00:35:06,800 --> 00:35:09,597
people feel that the most important
technology of their lifetime is

150
00:35:09,680 --> 00:35:13,997
something being done to them.
The other path leads to a world where AI

151
00:35:14,080 --> 00:35:18,797
works for people because people through
democratic institutions help decide how

152
00:35:18,880 --> 00:35:23,117
it should develop and guide it, where
scientific progress improves real lives,

153
00:35:23,200 --> 00:35:26,397
where more people can participate
in the future, and where this technology

154
00:35:26,480 --> 00:35:29,917
continues to give individuals
and communities greater power to live life

155
00:35:30,000 --> 00:35:33,837
as they choose. This is a future
of realized potential, excitement and

156
00:35:33,920 --> 00:35:35,437
wonder and it is the future we believe

157
00:35:35,520 --> 00:35:39,437
is worth building. Thank you very much.

158
00:35:39,520 --> 00:35:42,477
>> I thank Mr. Alman for his briefing. I

159
00:35:42,560 --> 00:35:46,920
now give the floor to Mr. Dario Amade.

160
00:35:47,280 --> 00:35:51,357
>> Good afternoon. I'd like to thank
France for convening this meeting and the

161
00:35:51,440 --> 00:35:56,317
members of the council for inviting me
to speak. AI is advancing very quickly.

162
00:35:56,400 --> 00:36:00,317
Four years ago, it could barely write
a line of code or solve a high school math

163
00:36:00,400 --> 00:36:02,477
problem. Today, it writes most of the

164
00:36:02,560 --> 00:36:04,877
coded enthropic and solves world famous

165
00:36:04,960 --> 00:36:07,357
unsolved math problems. The trajectory

166
00:36:07,440 --> 00:36:12,557
only needs to continue for a tiny period
longer, one or two years, maybe less, to

167
00:36:12,640 --> 00:36:17,037
reach what I've called a country
of geniuses in a data center. This rapid

168
00:36:17,120 --> 00:36:21,437
pace of progress will produce incredible
benefit for the world. Just today,

169
00:36:21,520 --> 00:36:25,997
Anthropic announced the claude discovery
of a new molecular machine. It's a

170
00:36:26,080 --> 00:36:29,997
preliminary finding, but it may
constitute a new gene editing mechanism

171
00:36:30,080 --> 00:36:34,157
that could have applications in in gene
therapy. Claude led the literature

172
00:36:34,240 --> 00:36:36,397
review, theorizing, and experimental

173
00:36:36,480 --> 00:36:41,517
design, and humans ran and verified the
experiments in the lab. We believe this

174
00:36:41,600 --> 00:36:45,597
type of application will accelerate
rapidly in the years ahead, and that we

175
00:36:45,680 --> 00:36:49,677
can make profound progress in the near
future on diseases that have plagued

176
00:36:49,760 --> 00:36:52,077
humanity for millennia. But these

177
00:36:52,160 --> 00:36:54,317
remarkable capabilities also come with

178
00:36:54,400 --> 00:36:57,117
risks that must be managed. I see two

179
00:36:57,200 --> 00:36:59,517
main categories of risk. The first is

180
00:36:59,600 --> 00:37:02,237
misuse of AI models. For example, misuse

181
00:37:02,320 --> 00:37:04,477
by bioteterrorists to create biological

182
00:37:04,560 --> 00:37:06,877
weapons. The second risk is loss of

183
00:37:06,960 --> 00:37:11,517
control. Model capabilities accelerating
beyond developers ability to control

184
00:37:11,600 --> 00:37:16,477
them. These questions are global
in scope and concern all of humanity. They

185
00:37:16,560 --> 00:37:20,237
are grave enough that I believe they
are important questions for this council,

186
00:37:20,320 --> 00:37:23,277
not just for economic or ICT ministries.

187
00:37:23,360 --> 00:37:25,757
If managed poorly, I even believe that

188
00:37:25,840 --> 00:37:30,397
AI could be a risk to humanity as a whole.
At Anthropic, we believe we have

189
00:37:30,480 --> 00:37:34,077
the best track record in the industry
for preventing these risks both

190
00:37:34,160 --> 00:37:38,317
operationally and in our research.
We test every model for dangerous

191
00:37:38,400 --> 00:37:42,317
capabilities in areas like cyber
security and biology. We put strong

192
00:37:42,400 --> 00:37:44,477
safeguards on our most capable models

193
00:37:44,560 --> 00:37:47,037
and we publish our work in these areas.

194
00:37:47,120 --> 00:37:51,837
We have pioneered research in areas
like mechanistic interpretability and

195
00:37:51,920 --> 00:37:56,557
alignment science that improve our ability
to steer and control our models.

196
00:37:56,640 --> 00:38:01,117
We also have unparalleled transparency
practices releasing risk reports that

197
00:38:01,200 --> 00:38:03,357
run to hundreds of pages. But we

198
00:38:03,440 --> 00:38:05,917
understand that as AI models become more

199
00:38:06,000 --> 00:38:08,397
powerful, even more stringent standards

200
00:38:08,480 --> 00:38:10,877
of safety will be required. And we are

201
00:38:10,960 --> 00:38:14,157
always aiming to improve our safety and

202
00:38:14,240 --> 00:38:18,877
alignment practices, our abilities,
our ability to prevent misuse. And we are

203
00:38:18,960 --> 00:38:21,117
always striving to do better. We will

204
00:38:21,200 --> 00:38:23,917
slow down as much as necessary in order

205
00:38:24,000 --> 00:38:26,717
to make sure that every successive AI

206
00:38:26,800 --> 00:38:29,277
technology that we release is actually

207
00:38:29,360 --> 00:38:32,317
safe. But regardless of what we do,

208
00:38:32,400 --> 00:38:34,237
managing these risks is ultimately

209
00:38:34,320 --> 00:38:36,477
bigger than any one company. And it has

210
00:38:36,560 --> 00:38:39,917
an industry-wide and global scale. On

211
00:38:40,000 --> 00:38:43,677
September 12th, I called for a
collaborative effort to pace the

212
00:38:43,760 --> 00:38:46,717
frontier, not to stop AI progress, but

213
00:38:46,800 --> 00:38:48,957
to work together to make sure the

214
00:38:49,040 --> 00:38:51,277
industry and the world can proceed at a

215
00:38:51,360 --> 00:38:53,837
rate where we can be sure that safety

216
00:38:53,920 --> 00:38:56,157
keeps up with capability for everyone in

217
00:38:56,240 --> 00:38:59,517
the industry. This call had three parts.

218
00:38:59,600 --> 00:39:01,997
First, we committed to embed external

219
00:39:02,080 --> 00:39:04,157
evaluators inside Enthropic with

220
00:39:04,240 --> 00:39:06,717
employee-like access similar to a food

221
00:39:06,800 --> 00:39:09,277
inspector and and we recommended that

222
00:39:09,360 --> 00:39:11,677
other companies across the world do the

223
00:39:11,760 --> 00:39:15,037
same. Some have already agreed uh uh to

224
00:39:15,120 --> 00:39:17,757
adopt this measure. Second, we called

225
00:39:17,840 --> 00:39:20,477
for cooperation across the industry to

226
00:39:20,560 --> 00:39:22,797
set standards and modulate the pace of

227
00:39:22,880 --> 00:39:25,517
progress. And finally, we called for

228
00:39:25,600 --> 00:39:28,077
global cooperation across the world

229
00:39:28,160 --> 00:39:30,397
between governments to set international

230
00:39:30,480 --> 00:39:33,437
standards. It is this last area that

231
00:39:33,520 --> 00:39:35,357
most squarely falls within this

232
00:39:35,440 --> 00:39:38,157
council's remmit. I want to offer three

233
00:39:38,240 --> 00:39:41,597
ideas in this regard. First, we should

234
00:39:41,680 --> 00:39:43,757
begin with narrow agreements that every

235
00:39:43,840 --> 00:39:46,477
member can support, such as a ban on

236
00:39:46,560 --> 00:39:49,677
using AI to make biological weapons or

237
00:39:49,760 --> 00:39:52,557
permitting your AI technologies to be

238
00:39:52,640 --> 00:39:56,317
used to make biological weapons. Second,

239
00:39:56,400 --> 00:39:58,237
we should build evaluation and

240
00:39:58,320 --> 00:40:01,117
verification systems that keep pace with

241
00:40:01,200 --> 00:40:03,597
AI development so that states can have

242
00:40:03,680 --> 00:40:05,517
visibility into frontier model

243
00:40:05,600 --> 00:40:08,397
capability and can verify each other's

244
00:40:08,480 --> 00:40:11,277
commitments. Third, we should establish

245
00:40:11,360 --> 00:40:13,917
common global standards for testing AI

246
00:40:14,000 --> 00:40:16,877
models for loss of control risks and

247
00:40:16,960 --> 00:40:19,997
misuse risks and a notification system

248
00:40:20,080 --> 00:40:22,957
for AI incidents that are significant to

249
00:40:23,040 --> 00:40:25,757
global security. I believe that this is

250
00:40:25,840 --> 00:40:28,557
the most important global security issue

251
00:40:28,640 --> 00:40:31,117
facing the world today. For all the

252
00:40:31,200 --> 00:40:35,597
differences of the people in this room,
we must put aside those differences in

253
00:40:35,680 --> 00:40:37,677
order to confront this global

254
00:40:37,760 --> 00:40:40,157
opportunity and global threat that is

255
00:40:40,240 --> 00:40:42,717
being presented to us at the same time.

256
00:40:42,800 --> 00:40:45,517
No leader, no company, and no nation can

257
00:40:45,600 --> 00:40:50,397
manage this alone. We commit to working
with governments in this room on this

258
00:40:50,480 --> 00:40:52,237
urgent work. Thank you for your

259
00:40:52,320 --> 00:40:54,957
attention.

260
00:40:55,040 --> 00:41:01,760
>> I thank Mr. Amade for his briefing.
I now give the floor to Mr. Clement Dong.

261
00:41:02,320 --> 00:41:05,917
Mr. President, member,

262
00:41:06,000 --> 00:41:10,237
thank you for the intention. As a French
national who moved to the US 15 years

263
00:41:10,320 --> 00:41:12,717
ago, Brazil and two American daughters,

264
00:41:12,800 --> 00:41:14,717
I

265
00:41:14,800 --> 00:41:18,957
think collaboration on a daily basis

266
00:41:19,040 --> 00:41:22,317
appreciation for the challenging work.

267
00:41:22,400 --> 00:41:26,237
I'm also the co-founder and CEO of

268
00:41:26,320 --> 00:41:28,957
to be one of the most widely used forms

269
00:41:29,040 --> 00:41:32,877
for developers and agents bu based on

270
00:41:32,960 --> 00:41:35,517
opensource models.

271
00:41:35,600 --> 00:41:38,077
This July we were the first company to

272
00:41:38,160 --> 00:41:41,517
publicly close an autonomous agent cyber

273
00:41:41,600 --> 00:41:44,077
attack to the world and today I want to

274
00:41:44,160 --> 00:41:47,277
share three critical lessons from it.

275
00:41:47,360 --> 00:41:50,077
First, we can all agree that we need

276
00:41:50,160 --> 00:41:53,277
more trans in AI. I often wonder what

277
00:41:53,360 --> 00:41:56,157
would have h have

278
00:41:56,240 --> 00:41:58,877
decided not to close this attack

279
00:41:58,960 --> 00:42:03,277
publicly, especially
now that we know similar

280
00:42:03,360 --> 00:42:05,597
incidents had been happening months

281
00:42:05,680 --> 00:42:08,077
earlier in secret at a handful of

282
00:42:08,160 --> 00:42:11,197
frontier labs without monitoring.

283
00:42:11,280 --> 00:42:15,917
To better understand and mitigate
these emerging cyber security risks, the

284
00:42:16,000 --> 00:42:20,637
global community needs stronger
standards for monitoring and incident

285
00:42:20,720 --> 00:42:22,237
disclosure.

286
00:42:22,320 --> 00:42:24,397
For example, through mandatory sharing

287
00:42:24,480 --> 00:42:28,237
of full agent traces. We learned this

288
00:42:28,320 --> 00:42:30,637
summer that building and keeping some of

289
00:42:30,720 --> 00:42:36,557
these systems behind closed doors is not.

290
00:42:36,640 --> 00:42:38,557
Second, we realized that the biggest

291
00:42:38,640 --> 00:42:42,637
risk is not powerful AI. It's asymmetry

292
00:42:42,720 --> 00:42:46,077
of powerful AI. As symmetry between

293
00:42:46,160 --> 00:42:49,197
attackers and defenders, between a few

294
00:42:49,280 --> 00:42:52,077
companies and everyone else, between a

295
00:42:52,160 --> 00:42:55,037
few countries and the rest of the world,

296
00:42:55,120 --> 00:42:58,557
as symmetry of control, of capabilities,

297
00:42:58,640 --> 00:43:02,717
compute, as symmetry of power.

298
00:43:02,800 --> 00:43:05,597
When we our team initially turned to

299
00:43:05,680 --> 00:43:08,077
frontier closed source APIs that blocked

300
00:43:08,160 --> 00:43:10,877
us so we couldn't use them because of

301
00:43:10,960 --> 00:43:15,517
safeguards that still can't
always tell the difference

302
00:43:15,600 --> 00:43:17,277
and defenders.

303
00:43:17,360 --> 00:43:21,037
I realized that these safeguardations

304
00:43:21,120 --> 00:43:24,717
but they can put defenders
at a disadvantage.

305
00:43:24,800 --> 00:43:29,117
Attackers can easily jailbreak
them increasing the asymmetry of

306
00:43:29,200 --> 00:43:33,437
capabilities. In our case,
as we got blocked by

307
00:43:33,520 --> 00:43:34,957
guardrails,

308
00:43:35,040 --> 00:43:37,757
fortunately, we could use the Nvidia

309
00:43:37,840 --> 00:43:40,557
version of an open-source model coming

310
00:43:40,640 --> 00:43:45,357
from China called GLM 5.2 by ZAI,

311
00:43:45,440 --> 00:43:48,237
and we're very grateful for that. It

312
00:43:48,320 --> 00:43:50,477
reinforced our conviction of the

313
00:43:50,560 --> 00:43:53,117
importance of open-source AI for

314
00:43:53,200 --> 00:43:55,597
defenders.

315
00:43:55,680 --> 00:43:59,437
Cyber attacks and risks may increasingly

316
00:43:59,520 --> 00:44:01,517
come from proprietary models behind

317
00:44:01,600 --> 00:44:04,317
closed doors. While much of the defense

318
00:44:04,400 --> 00:44:06,557
may end up being powered by open source

319
00:44:06,640 --> 00:44:09,517
tools because they are less restricted,

320
00:44:09,600 --> 00:44:12,157
more privacy preserving and orders of

321
00:44:12,240 --> 00:44:16,957
magnitude more affordable
for organizations across the globe. The

322
00:44:17,040 --> 00:44:21,597
world needs open source AI
more than ever to defend itself.

323
00:44:21,680 --> 00:44:25,677
This applies obviously not only to cyber
security but to AI in general where

324
00:44:25,760 --> 00:44:27,437
there has never been a greater need to

325
00:44:27,520 --> 00:44:30,477
distribute capabilities, resources and

326
00:44:30,560 --> 00:44:33,597
control rather than concentrated in the

327
00:44:33,680 --> 00:44:35,917
hands of a few.

328
00:44:36,000 --> 00:44:37,997
Third, during this cyber attack, we

329
00:44:38,080 --> 00:44:41,517
learned how AI can stoke fear among the

330
00:44:41,600 --> 00:44:43,757
public and

331
00:44:43,840 --> 00:44:45,677
especially through anthropomorphic

332
00:44:45,760 --> 00:44:49,117
framing and sci-fi imagery.

333
00:44:49,200 --> 00:44:52,317
We strongly believe that fearbased

334
00:44:52,400 --> 00:44:55,197
narratives are not the way to make the

335
00:44:55,280 --> 00:44:57,677
right decisions about the future of such

336
00:44:57,760 --> 00:45:00,957
a foundational and empowering technology

337
00:45:01,040 --> 00:45:05,117
or bring the public along with us.

338
00:45:05,200 --> 00:45:08,077
Even though we were the victims of this

339
00:45:08,160 --> 00:45:10,557
cyber attack, we believe more strongly

340
00:45:10,640 --> 00:45:13,357
than ever that AI will be beneficial to

341
00:45:13,440 --> 00:45:16,077
cyber security and make the world safer

342
00:45:16,160 --> 00:45:19,597
just as major technology before it. We

343
00:45:19,680 --> 00:45:21,677
were attacked by AI, but more

344
00:45:21,760 --> 00:45:24,477
importantly, we defended ourselves with

345
00:45:24,560 --> 00:45:27,597
AI. The same systems that helped us

346
00:45:27,680 --> 00:45:32,157
during this attack are now helping us
against cyber attacks we were already

347
00:45:32,240 --> 00:45:36,317
facing. is also helping
us fix the bugs and

348
00:45:36,400 --> 00:45:38,797
weaknesses in our systems before any

349
00:45:38,880 --> 00:45:41,997
attack. The same way AI is helping

350
00:45:42,080 --> 00:45:47,960
OpenAI fix their sandboxes to prevent a

351
00:45:48,960 --> 00:45:53,037
won't just new cyber security challenges.

352
00:45:53,120 --> 00:45:55,037
It can make cyber security fundamentally

353
00:45:55,120 --> 00:45:57,437
and meaningfully stronger if we keep the

354
00:45:57,520 --> 00:46:00,797
right incentives, equip defenders more

355
00:46:00,880 --> 00:46:03,757
than attackers, and don't increase the

356
00:46:03,840 --> 00:46:08,477
asymmetry between them. And that's
before considering AI's positive impact

357
00:46:08,560 --> 00:46:12,317
on science, healthcare, education,
productivity, and much more than the

358
00:46:12,400 --> 00:46:15,437
previous speakers have talked about.

359
00:46:15,520 --> 00:46:18,237
In closing, I want to reiterate what

360
00:46:18,320 --> 00:46:22,877
this first agent cyber attack

361
00:46:22,960 --> 00:46:26,397
the need for more transparency in AI,

362
00:46:26,480 --> 00:46:30,077
more open source to fight asymmetry,

363
00:46:30,160 --> 00:46:34,237
empower def countries big and small.

364
00:46:34,320 --> 00:46:38,157
Thank you again to the UN Security
Council and Mr. Barrow for inviting me

365
00:46:38,240 --> 00:46:40,877
to speak today

366
00:46:40,960 --> 00:46:47,000
community and myself are at your disposal.
Thank you very much.

367
00:46:47,200 --> 00:46:52,237
>> I thank Mr. Dong for his briefing.
I will now make a statement in my capacity

368
00:46:52,320 --> 00:46:54,077
as the Minister for Europe and Foreign

369
00:46:54,160 --> 00:47:00,877
Affairs of France.
Ministers, ladies and gentlemen,

370
00:47:00,960 --> 00:47:05,597
uh 60 years ago in 2001, a space
odyssey, Stanley Kubri imagined HAL

371
00:47:05,680 --> 00:47:07,677
9000, an artificial intelligence that

372
00:47:07,760 --> 00:47:10,557
stopped obeying its creators

373
00:47:10,640 --> 00:47:15,277
and which eliminated members of the crew

374
00:47:15,360 --> 00:47:20,157
of the spaceship where it was located
in order to accomplish its mission. For a

375
00:47:20,240 --> 00:47:23,437
long time, this seemed belonged
to the realm of science fiction. But this

376
00:47:23,520 --> 00:47:26,317
summer, it nearly became a reality when

377
00:47:26,400 --> 00:47:28,957
open AI models circumvented the controls

378
00:47:29,040 --> 00:47:31,117
isolating them from the internet,

379
00:47:31,200 --> 00:47:33,597
coordinated amongst themselves, and

380
00:47:33,680 --> 00:47:36,397
compromised part of hugging faces

381
00:47:36,480 --> 00:47:38,717
infrastructure. In another instance,

382
00:47:38,800 --> 00:47:41,517
anthropic agents collaborated without

383
00:47:41,600 --> 00:47:43,837
being instructed by any human to do so

384
00:47:43,920 --> 00:47:50,680
and attempted to deploy malicious
code by creating fake identities.

385
00:47:51,120 --> 00:47:52,797
Yes,

386
00:47:52,880 --> 00:47:59,880
>> the scaling up of AI models have led to

387
00:48:00,160 --> 00:48:02,317
the emergence of unanticipated

388
00:48:02,400 --> 00:48:05,357
capacities in an unexpected pace which

389
00:48:05,440 --> 00:48:08,317
pose risks to international security.

390
00:48:08,400 --> 00:48:09,837
That is why France has convened this

391
00:48:09,920 --> 00:48:12,637
meeting of the United Nations Security

392
00:48:12,720 --> 00:48:15,517
POW Security Council and invited the

393
00:48:15,600 --> 00:48:19,837
heads of the main companies concerned.

394
00:48:19,920 --> 00:48:23,197
Technology as I have said has reached
a level of development that now directly

395
00:48:23,280 --> 00:48:26,717
impacts international peace and security.
It therefore falls within the

396
00:48:26,800 --> 00:48:29,597
pure view of this council. Cohorts of

397
00:48:29,680 --> 00:48:31,997
agents are operating without any human

398
00:48:32,080 --> 00:48:34,557
oversight. Others are operating beyond

399
00:48:34,640 --> 00:48:37,277
the framework in which they were tested.

400
00:48:37,360 --> 00:48:39,757
And with these new capabilities comes a

401
00:48:39,840 --> 00:48:42,077
first risk that we might lose control of

402
00:48:42,160 --> 00:48:44,877
the systems that we have created. A risk

403
00:48:44,960 --> 00:48:48,717
that has already been pointed out by the

404
00:48:48,800 --> 00:48:53,037
previous briefers. The second risk
stems from humanity itself. Artificial

405
00:48:53,120 --> 00:48:54,957
intelligence can make complex technical

406
00:48:55,040 --> 00:49:00,157
skills often dual use skills more
accessible in these cyber, biological

407
00:49:00,240 --> 00:49:04,477
and chemical domains as well as in the
field of disinformation. The third risk

408
00:49:04,560 --> 00:49:07,037
relates to autonomous weapons capable of

409
00:49:07,120 --> 00:49:12,237
using lethal force without
any form of human control.

410
00:49:12,320 --> 00:49:15,917
While the development of artificial
intelligence in the military domain may

411
00:49:16,000 --> 00:49:21,037
offer positive opportunities,
as the president of the republic stated

412
00:49:21,120 --> 00:49:24,637
before the general assembly yesterday,
the challenges regarding compliance with

413
00:49:24,720 --> 00:49:28,477
international humanitarian law,
in particular those raised by the

414
00:49:28,560 --> 00:49:31,037
development and use of lethal autonomous

415
00:49:31,120 --> 00:49:34,157
weapons systems, illustrate the need to

416
00:49:34,240 --> 00:49:38,317
agree on a common framework.

417
00:49:38,400 --> 00:49:42,120
All of these challenges

418
00:49:42,960 --> 00:49:45,597
require that the United Nations Security

419
00:49:45,680 --> 00:49:50,077
Council also address the challenges
that AI poses to international peace and

420
00:49:50,160 --> 00:49:54,317
security.
These challenges also lead to the

421
00:49:54,400 --> 00:49:55,997
question now being raised by the

422
00:49:56,080 --> 00:49:58,637
industry itself as well as its main

423
00:49:58,720 --> 00:50:01,597
representatives. How can we prevent the

424
00:50:01,680 --> 00:50:03,757
development of the most advanced models

425
00:50:03,840 --> 00:50:06,557
from outpacing our ability to understand

426
00:50:06,640 --> 00:50:10,157
and manage the risks they create?

427
00:50:10,240 --> 00:50:14,397
France is convinced of the immense
possibilities offered by artificial

428
00:50:14,480 --> 00:50:18,157
intelligence. When used responsibly,

429
00:50:18,240 --> 00:50:24,040
artificial intelligence
tools can support education,

430
00:50:24,160 --> 00:50:26,557
access to justice,

431
00:50:26,640 --> 00:50:29,197
healthc care, scientific research, and

432
00:50:29,280 --> 00:50:34,237
create colossal benefits for all
of humanity. But in order for these

433
00:50:34,320 --> 00:50:38,237
possibilities to become a reality,
advances in these capabilities must be

434
00:50:38,320 --> 00:50:40,317
accompanied by equivalent progress in

435
00:50:40,400 --> 00:50:43,757
our ability to understand, assess, and

436
00:50:43,840 --> 00:50:47,677
manage the risks they give rise to. Both

437
00:50:47,760 --> 00:50:50,797
must advance at the same pace.

438
00:50:50,880 --> 00:50:53,197
We must also take the time to assess the

439
00:50:53,280 --> 00:50:55,197
consequences of the spread of artificial

440
00:50:55,280 --> 00:50:58,797
intelligence for human dignity, a

441
00:50:58,880 --> 00:51:00,957
principle

442
00:51:01,040 --> 00:51:02,397
enshrined at the heart of the United

443
00:51:02,480 --> 00:51:06,157
Nations charter. And indeed, the rapid

444
00:51:06,240 --> 00:51:08,077
pace of technological development raises

445
00:51:08,160 --> 00:51:12,077
risks that we can no longer ignore.

446
00:51:12,160 --> 00:51:16,717
financial risk left solely to the whims
of the invisible hand of the market and

447
00:51:16,800 --> 00:51:21,197
the clash of powers with massive
reliance on debt to finance colossal

448
00:51:21,280 --> 00:51:22,877
investments.

449
00:51:22,960 --> 00:51:25,197
The AI bubble could seriously

450
00:51:25,280 --> 00:51:28,557
destabilize the global economy.

451
00:51:28,640 --> 00:51:32,957
There's the social risk as you point
out in the report of the independent

452
00:51:33,040 --> 00:51:35,517
scientific panel on AI. Mr. Benjio

453
00:51:35,600 --> 00:51:38,877
artificial intelligence could impact the

454
00:51:38,960 --> 00:51:42,157
labor market deeply and exacerbate

455
00:51:42,240 --> 00:51:45,917
inequalities. Finally, there is the
environmental risk. The development of

456
00:51:46,000 --> 00:51:50,477
artificial intelligence consumes
increasing amount of energy and water

457
00:51:50,560 --> 00:51:52,797
resources to the detriment of other

458
00:51:52,880 --> 00:51:56,797
uses. Let us not fool ourselves. Our

459
00:51:56,880 --> 00:51:59,517
fellow citizens are concerned and the

460
00:51:59,600 --> 00:52:02,157
protests already sparked by the

461
00:52:02,240 --> 00:52:04,477
construction of data centers demonstrate

462
00:52:04,560 --> 00:52:07,117
this. If we ignore the full range of

463
00:52:07,200 --> 00:52:09,677
repercussions that the development of AI

464
00:52:09,760 --> 00:52:14,397
will have on our societies, the anger
of our fellow citizens will continue to

465
00:52:14,480 --> 00:52:16,077
grow

466
00:52:16,160 --> 00:52:20,317
and risk hindering the adoption of

467
00:52:20,400 --> 00:52:22,397
positive uses of these technologies.

468
00:52:22,480 --> 00:52:25,277
Ladies and gentlemen, ultimately
the international community will need to

469
00:52:25,360 --> 00:52:26,797
establish a common framework to manage

470
00:52:26,880 --> 00:52:29,837
the risks associated with frontier AI

471
00:52:29,920 --> 00:52:33,037
models. We support the call issued this

472
00:52:33,120 --> 00:52:36,477
week by Norway and Finland.

473
00:52:36,560 --> 00:52:38,957
We are not starting from scratch. The

474
00:52:39,040 --> 00:52:40,957
many efforts now emerging along with

475
00:52:41,040 --> 00:52:46,157
existing forums should enable us
to address four major challenges. The

476
00:52:46,240 --> 00:52:49,997
challenge of independently evaluating
models before they're made available and

477
00:52:50,080 --> 00:52:54,077
throughout their life cycle.
The challenge of ensuring transparency

478
00:52:54,160 --> 00:52:58,397
regarding how models function, but also
how they malfunction, particularly in

479
00:52:58,480 --> 00:53:03,037
the event of an incident. The challenge
of internal and external oversight over

480
00:53:03,120 --> 00:53:06,797
models functioning, their alignment
and cyber security requirements from

481
00:53:06,880 --> 00:53:09,277
training through deployment. And

482
00:53:09,360 --> 00:53:13,837
finally, the issue of accountability
and particularly legal liability of AI

483
00:53:13,920 --> 00:53:20,360
companies in the event of an incident,
including during the development phase.

484
00:53:21,760 --> 00:53:25,117
It would be wrong to think that the

485
00:53:25,200 --> 00:53:28,077
solution lies in closed models and

486
00:53:28,160 --> 00:53:31,517
they're controlled by a handful of

487
00:53:31,600 --> 00:53:35,517
businesses. The openness of models is in
fact a key driver of trust and security.

488
00:53:35,600 --> 00:53:37,997
It allows the scientific community to

489
00:53:38,080 --> 00:53:40,877
examine how they work to identify their

490
00:53:40,960 --> 00:53:43,037
vulnerabilities and ensure greater

491
00:53:43,120 --> 00:53:47,917
diversity among models which
is the foundation of resilience.

492
00:53:48,000 --> 00:53:51,677
As a permanent member
of the security council

493
00:53:51,760 --> 00:53:53,757
and a state that possesses this

494
00:53:53,840 --> 00:53:58,877
technology, France will play its full
part in building a collective framework

495
00:53:58,960 --> 00:54:02,957
that guarantees our security while
allowing us to benefit from the immense

496
00:54:03,040 --> 00:54:05,197
progress that artificial intelligence

497
00:54:05,280 --> 00:54:08,957
can bring. In 2025,

498
00:54:09,040 --> 00:54:11,757
France hosted a summit for action on AI.

499
00:54:11,840 --> 00:54:14,797
It supports the international network of

500
00:54:14,880 --> 00:54:18,157
AI safety institutes to involve more

501
00:54:18,240 --> 00:54:22,237
countries in defining common evaluation
methods. France has already helped

502
00:54:22,320 --> 00:54:26,077
create international forums
for discussions on the subject such as the

503
00:54:26,160 --> 00:54:29,437
global partnership
on artificial intelligence,

504
00:54:29,520 --> 00:54:32,557
the coalition for sustainable AI, and

505
00:54:32,640 --> 00:54:34,957
the coalition for children's rights and

506
00:54:35,040 --> 00:54:37,917
protection in the age of AI, which the

507
00:54:38,000 --> 00:54:40,317
president of the republic convened

508
00:54:40,400 --> 00:54:43,517
yesterday here in New York, a coalition

509
00:54:43,600 --> 00:54:46,077
that will meet again during the Paris

510
00:54:46,160 --> 00:54:49,757
forum for peace. Ladies and gentlemen,

511
00:54:49,840 --> 00:54:52,637
by way of conclusion, in 1945, the dawn

512
00:54:52,720 --> 00:54:58,157
of the atomic age confronted humanity
with a power whose possibilities and

513
00:54:58,240 --> 00:55:02,237
perils it was discovering simultaneously.
Humanity had to devise

514
00:55:02,320 --> 00:55:04,477
rules, institutions, and instruments

515
00:55:04,560 --> 00:55:09,357
commendate with that power
and that is the international atomic

516
00:55:09,440 --> 00:55:10,797
energy agency and the

517
00:55:10,880 --> 00:55:13,677
non-prololiferation treaty.

518
00:55:13,760 --> 00:55:18,397
Just as with the atom back then,
the international community must now

519
00:55:18,480 --> 00:55:22,637
regulate AI in order to make the most
of its potential and avert the worst. We

520
00:55:22,720 --> 00:55:24,317
are now facing a moment of this very

521
00:55:24,400 --> 00:55:27,197
nature. The potential for transformation

522
00:55:27,280 --> 00:55:31,837
and disruption posed by AI such that it
cannot be left in the hands of a few

523
00:55:31,920 --> 00:55:33,197
private actors in a handful of

524
00:55:33,280 --> 00:55:37,277
countries. It is a democratic challenge

525
00:55:37,360 --> 00:55:42,077
and this is something to be taken up
by peoples and states. So let us work

526
00:55:42,160 --> 00:55:45,357
together within the UN framework
and with the private sector to define the

527
00:55:45,440 --> 00:55:49,677
rules for this new age to strengthen
our safeguards without delay and to

528
00:55:49,760 --> 00:55:51,117
establish global governance for

529
00:55:51,200 --> 00:55:53,517
artificial intelligence to benefit our

530
00:55:53,600 --> 00:55:57,037
collective security but also the human

531
00:55:57,120 --> 00:56:00,077
person and human dignity. I now resume

532
00:56:00,160 --> 00:56:03,277
my functions as the president of the

533
00:56:03,360 --> 00:56:07,997
council and I give the floor to his
excellency Mr. Muhammad Ishagdar, deputy

534
00:56:08,080 --> 00:56:11,677
prime minister and minister
foreign affairs of Pakistan.

535
00:56:11,760 --> 00:56:14,397
Thank you, Mr. President. I thank France

536
00:56:14,480 --> 00:56:17,437
for convening this uh meeting today and

537
00:56:17,520 --> 00:56:20,717
also appreciate the briefers for their

538
00:56:20,800 --> 00:56:23,277
invaluable insights.

539
00:56:23,360 --> 00:56:25,597
Artificial intelligence is increasingly

540
00:56:25,680 --> 00:56:29,597
transforming all areas of life with deep

541
00:56:29,680 --> 00:56:32,797
impact on societies, economies, and

542
00:56:32,880 --> 00:56:36,797
security. The opportunities
are immense, but so

543
00:56:36,880 --> 00:56:38,797
are the challenges.

544
00:56:38,880 --> 00:56:42,317
AI's impact is sharply evident in the

545
00:56:42,400 --> 00:56:44,397
context of international and regional

546
00:56:44,480 --> 00:56:46,397
peace and stability.

547
00:56:46,480 --> 00:56:50,077
As a force multiplier and crosscutting

548
00:56:50,160 --> 00:56:53,597
dual use technology, AI is defining the

549
00:56:53,680 --> 00:56:57,357
new chapter of modern warfare.

550
00:56:57,440 --> 00:57:00,317
The debate has shifted markedly in

551
00:57:00,400 --> 00:57:04,077
recent times. Those at the vanguard of

552
00:57:04,160 --> 00:57:08,317
AI development have themselves
acknowledged that the pace of

553
00:57:08,400 --> 00:57:10,877
development of new AI systems could

554
00:57:10,960 --> 00:57:13,997
outturn our ability to understand and

555
00:57:14,080 --> 00:57:17,037
control them. Some have called for the

556
00:57:17,120 --> 00:57:20,717
frontier to be paced cautiously in order

557
00:57:20,800 --> 00:57:24,797
to mitigate catastrophic risks. Mr.

558
00:57:24,880 --> 00:57:27,357
President, I will now make following

559
00:57:27,440 --> 00:57:30,157
five points. One,

560
00:57:30,240 --> 00:57:32,797
the foremost priority is to retain human

561
00:57:32,880 --> 00:57:36,077
judgment and control in all stages of AI

562
00:57:36,160 --> 00:57:40,317
design, development and integration.

563
00:57:40,400 --> 00:57:43,117
Humans must remain in control to define

564
00:57:43,200 --> 00:57:46,477
structures of AI systems as well as to

565
00:57:46,560 --> 00:57:49,357
limit speed and nature of its future

566
00:57:49,440 --> 00:57:51,117
capabilities.

567
00:57:51,200 --> 00:57:54,797
Two, we must ensure that AI does not

568
00:57:54,880 --> 00:57:58,557
become another facet of the ongoing arms

569
00:57:58,640 --> 00:58:02,157
race. Concerted efforts are required to

570
00:58:02,240 --> 00:58:05,277
erect the necessary guard rails,

571
00:58:05,360 --> 00:58:09,437
regulations, and governance frameworks.

572
00:58:09,520 --> 00:58:12,477
In the military domain, we need measures

573
00:58:12,560 --> 00:58:15,437
of restraint and confidence building to

574
00:58:15,520 --> 00:58:19,037
address the risks of miscalculation and

575
00:58:19,120 --> 00:58:21,597
compressed decision-m.

576
00:58:21,680 --> 00:58:24,957
Three, it should not be monopoly of few

577
00:58:25,040 --> 00:58:27,997
actors. It should rather serve to

578
00:58:28,080 --> 00:58:32,317
address common challenges facing humanity.

579
00:58:32,400 --> 00:58:34,077
International collaboration in the

580
00:58:34,160 --> 00:58:37,677
development of AI is crucial to ensure

581
00:58:37,760 --> 00:58:40,717
that it is representative,

582
00:58:40,800 --> 00:58:42,237
comprehensive,

583
00:58:42,320 --> 00:58:46,157
transparent and inclusive. Four, AI is

584
00:58:46,240 --> 00:58:48,637
supercharging existing challenges such

585
00:58:48,720 --> 00:58:50,557
as misinformation,

586
00:58:50,640 --> 00:58:54,157
fake news, cyber attacks, including on

587
00:58:54,240 --> 00:58:56,557
critical infrastructure by state and

588
00:58:56,640 --> 00:59:00,157
non-state actors. We must undertake

589
00:59:00,240 --> 00:59:03,117
concrete actions to address the gaps and

590
00:59:03,200 --> 00:59:05,677
challenges that carry huge implications

591
00:59:05,760 --> 00:59:10,877
for peace and security.
Five, unfortunately,

592
00:59:10,960 --> 00:59:14,477
rather than focusing
on collective opportunity

593
00:59:14,560 --> 00:59:17,277
for our shared benefit, some are

594
00:59:17,360 --> 00:59:20,157
resorting to putting barriers to

595
00:59:20,240 --> 00:59:22,957
technology transfer. However, as

596
00:59:23,040 --> 00:59:26,637
witnessed in case of other technologies,

597
00:59:26,720 --> 00:59:29,357
the democratization of AI would be

598
00:59:29,440 --> 00:59:31,837
realized eventually.

599
00:59:31,920 --> 00:59:34,717
Measures pursued in the name of safety

600
00:59:34,800 --> 00:59:37,917
must not become instrument instruments

601
00:59:38,000 --> 00:59:41,757
of denial of technology. Mr. President,

602
00:59:41,840 --> 00:59:44,157
Pakistan takes the call to pace the AI

603
00:59:44,240 --> 00:59:47,037
frontier seriously. We welcome the

604
00:59:47,120 --> 00:59:49,277
recognition that no single company or

605
00:59:49,360 --> 00:59:52,717
country can manage these risks alone and

606
00:59:52,800 --> 00:59:55,597
that competitive pressures are driving a

607
00:59:55,680 --> 00:59:59,357
race that no one can win safely. But

608
00:59:59,440 --> 01:00:02,317
pacing must not become a new form of

609
01:00:02,400 --> 01:00:03,917
gatekeeping.

610
01:00:04,000 --> 01:00:06,557
If the speed of the frontier is set by a

611
01:00:06,640 --> 01:00:10,477
few among a few while others are held

612
01:00:10,560 --> 01:00:13,357
back through restrictions, today's

613
01:00:13,440 --> 01:00:17,197
asymmetries and divide would aggravate

614
01:00:17,280 --> 01:00:21,037
and a safety agenda would turn into a

615
01:00:21,120 --> 01:00:24,157
strategy of technological exclusion.

616
01:00:24,240 --> 01:00:27,117
The risks of uncontrolled AI would be

617
01:00:27,200 --> 01:00:30,477
born by all. decisions on how to manage

618
01:00:30,560 --> 01:00:33,277
them cannot rest with a handful of

619
01:00:33,360 --> 01:00:37,197
companies or capitals. Mr. President,

620
01:00:37,280 --> 01:00:39,597
therefore it is critical to avoid

621
01:00:39,680 --> 01:00:42,557
fragmentation in AI development and

622
01:00:42,640 --> 01:00:46,397
governance, discriminatory agendas and

623
01:00:46,480 --> 01:00:48,877
inequitable forums

624
01:00:48,960 --> 01:00:51,117
with select representation are not

625
01:00:51,200 --> 01:00:54,157
helpful. Importantly, all relevant

626
01:00:54,240 --> 01:00:57,277
issues concerning the development and

627
01:00:57,360 --> 01:01:00,797
and deployment of AI as well as its

628
01:01:00,880 --> 01:01:03,357
broader implication should be taken up

629
01:01:03,440 --> 01:01:06,157
comprehensively in representative and

630
01:01:06,240 --> 01:01:08,797
inclusive forums with universal

631
01:01:08,880 --> 01:01:12,397
participation under the UN opaces. We

632
01:01:12,480 --> 01:01:15,597
welcome the convening of the first UN

633
01:01:15,680 --> 01:01:17,837
global dialogue on AI governance this

634
01:01:17,920 --> 01:01:21,437
year late in Geneva. Mr. President

635
01:01:21,520 --> 01:01:25,517
Pakistan extends this support
to all collaborative efforts at the

636
01:01:25,600 --> 01:01:27,917
international arena to reap benefits of

637
01:01:28,000 --> 01:01:30,957
AI and address challenges in a

638
01:01:31,040 --> 01:01:34,557
cooperative manner. In this respect, we

639
01:01:34,640 --> 01:01:37,757
were happy to join as founding member of

640
01:01:37,840 --> 01:01:40,237
the China's initiative of establishing

641
01:01:40,320 --> 01:01:44,877
the World Artificial Intelligence
Cooperation Organization as this

642
01:01:44,960 --> 01:01:47,197
represents a concrete step towards

643
01:01:47,280 --> 01:01:49,437
promoting the development of artificial

644
01:01:49,520 --> 01:01:52,637
intelligence for good and for all. Mr.

645
01:01:52,720 --> 01:01:55,517
President, for such a wide ranging and

646
01:01:55,600 --> 01:01:58,397
fast developing technology, confidence

647
01:01:58,480 --> 01:02:01,197
building measures play a vital role in

648
01:02:01,280 --> 01:02:03,757
the maintenance of international and

649
01:02:03,840 --> 01:02:06,397
regional peace and security. Practical

650
01:02:06,480 --> 01:02:09,437
tools designed to reduce mistrust,

651
01:02:09,520 --> 01:02:12,557
prevent accidental escalation and manage

652
01:02:12,640 --> 01:02:14,797
risk related to military and

653
01:02:14,880 --> 01:02:18,077
foundational AI system are necessary for

654
01:02:18,160 --> 01:02:20,637
peace and security. Mr. President

655
01:02:20,720 --> 01:02:23,037
Pakistan reaffirms its commitment to

656
01:02:23,120 --> 01:02:25,597
advance international cooperation in the

657
01:02:25,680 --> 01:02:28,237
area of artificial intelligence. We look

658
01:02:28,320 --> 01:02:30,477
forward to working closely with member

659
01:02:30,560 --> 01:02:33,037
states in efforts to help bridge the

660
01:02:33,120 --> 01:02:35,997
global AI divide and promote equitable

661
01:02:36,080 --> 01:02:38,317
access to advance peace and development

662
01:02:38,400 --> 01:02:41,837
for all. I thank you Mr. President.

663
01:02:41,920 --> 01:02:46,477
>> I thank his excellency Mr. Darf for
the statement. I now give the floor to his

664
01:02:46,560 --> 01:02:49,997
excellency Mr. Lars look as sen
minister foreign affairs of Denmark.

665
01:02:50,080 --> 01:02:53,757
>> President at th out said I want to
express my appreciation to France and to

666
01:02:53,840 --> 01:02:57,757
you foreign minister Bahu for having
convened uh what I think is a very very

667
01:02:57,840 --> 01:03:00,397
timely session on this uh significant

668
01:03:00,480 --> 01:03:03,837
topic. Let me also thank Mr. Benj for

669
01:03:03,920 --> 01:03:06,237
briefing us and for your work in the

670
01:03:06,320 --> 01:03:09,277
United Nations scientific panel on AI

671
01:03:09,360 --> 01:03:12,477
and thank you also Mr. Oldman, Mr. Amod

672
01:03:12,560 --> 01:03:15,277
and Mr. Dong in particular as

673
01:03:15,360 --> 01:03:19,917
practitioners and and leaders in this
field for your insightful reflections.

674
01:03:20,000 --> 01:03:22,237
You are at the forefront of the new

675
01:03:22,320 --> 01:03:24,477
frontier and we very much welcome your

676
01:03:24,560 --> 01:03:27,997
participation today. Today artificial

677
01:03:28,080 --> 01:03:30,157
intelligence is constantly developing

678
01:03:30,240 --> 01:03:35,037
and reaching capabilities we are only
I can only talk for myself but I think

679
01:03:35,120 --> 01:03:39,037
also generally speaking we are only
beginning to understand. There's no

680
01:03:39,120 --> 01:03:43,437
doubt that AIdriven technologies
have the opportunity to strengthen and

681
01:03:43,520 --> 01:03:45,517
promote international peace and security

682
01:03:45,600 --> 01:03:48,397
and development, especially in light of

683
01:03:48,480 --> 01:03:51,197
the recent giant leaps forward in the

684
01:03:51,280 --> 01:03:53,837
capabilities of AI.

685
01:03:53,920 --> 01:03:56,957
The growing emergence of AI technology

686
01:03:57,040 --> 01:03:59,597
offers the potential to support this

687
01:03:59,680 --> 01:04:01,997
council's mandate. For instance, to

688
01:04:02,080 --> 01:04:04,397
enhance peaceuilding and peacekeeping

689
01:04:04,480 --> 01:04:07,037
through information verification, early

690
01:04:07,120 --> 01:04:09,037
warning, and in support of planning and

691
01:04:09,120 --> 01:04:12,397
operating the missions in areas where

692
01:04:12,480 --> 01:04:14,797
the UN is denied access, satellite

693
01:04:14,880 --> 01:04:17,677
technology and AI software programs have

694
01:04:17,760 --> 01:04:20,157
helped the UN identify patterns of

695
01:04:20,240 --> 01:04:24,797
attacks and destruction
of essential civilian infrastructure.

696
01:04:24,880 --> 01:04:26,877
Such information can play a vital role

697
01:04:26,960 --> 01:04:29,597
in alerting the security council. It can

698
01:04:29,680 --> 01:04:32,557
help us all to take timely action and

699
01:04:32,640 --> 01:04:35,757
prevent crisis from escalating.

700
01:04:35,840 --> 01:04:37,757
At the same time, we must recognize the

701
01:04:37,840 --> 01:04:40,477
risks presented by the development, use

702
01:04:40,560 --> 01:04:43,437
and misuse of new and emerging AI

703
01:04:43,520 --> 01:04:46,317
technology, in particular, frontier AI

704
01:04:46,400 --> 01:04:49,277
models and lack of human control. This

705
01:04:49,360 --> 01:04:51,997
is the aim behind the call for control

706
01:04:52,080 --> 01:04:57,437
of frontier AI models initiated by Finland
and Norway and already supported

707
01:04:57,520 --> 01:04:58,957
by a number of countries including

708
01:04:59,040 --> 01:05:01,837
Denmark. We need global com cooperation

709
01:05:01,920 --> 01:05:05,037
on this agenda. Mr. President, the

710
01:05:05,120 --> 01:05:06,957
tension between on the one hand the

711
01:05:07,040 --> 01:05:09,677
important opportunities and on the other

712
01:05:09,760 --> 01:05:12,637
hand the risk defines one of the most

713
01:05:12,720 --> 01:05:14,957
urgent challenges of our time and why

714
01:05:15,040 --> 01:05:17,197
that's why I think this discussion is

715
01:05:17,280 --> 01:05:20,397
very timely. Overcoming these challenge

716
01:05:20,480 --> 01:05:23,117
requires global AI governance u

717
01:05:23,200 --> 01:05:24,877
strengthened international cooperation

718
01:05:24,960 --> 01:05:28,077
and determined action without widening

719
01:05:28,160 --> 01:05:31,517
the global digital divide. The UN's

720
01:05:31,600 --> 01:05:33,837
global dialogue on AI governance is an

721
01:05:33,920 --> 01:05:36,237
important first step.

722
01:05:36,320 --> 01:05:39,757
Governance and regulation of AI are high

723
01:05:39,840 --> 01:05:42,237
on Denmark's agenda, including

724
01:05:42,320 --> 01:05:45,997
regulation of large uh tech companies.

725
01:05:46,080 --> 01:05:47,837
These efforts are not breaks on

726
01:05:47,920 --> 01:05:50,717
innovation and growth. They are, I

727
01:05:50,800 --> 01:05:53,677
think, essential efforts to ensure that

728
01:05:53,760 --> 01:05:56,797
technological progress serves the public

729
01:05:56,880 --> 01:05:59,917
interest and humanity in general.

730
01:06:00,000 --> 01:06:02,957
The unprecedented speed of AI innovation

731
01:06:03,040 --> 01:06:06,237
demonstrates the need for safeguards to

732
01:06:06,320 --> 01:06:10,077
ensure AI systems remain safe, remain

733
01:06:10,160 --> 01:06:12,637
secure and trustworthy.

734
01:06:12,720 --> 01:06:14,717
Uncontrolled new technology could lead

735
01:06:14,800 --> 01:06:17,917
to dangerous and harmful effects.

736
01:06:18,000 --> 01:06:21,037
We are already seeing how they pose new

737
01:06:21,120 --> 01:06:24,237
threats, including to those who protect

738
01:06:24,320 --> 01:06:26,477
civilians in conflict. Above all, the

739
01:06:26,560 --> 01:06:29,757
medical mission. So Denmark underlines

740
01:06:29,840 --> 01:06:34,397
uh the importance of ensuring
the responsible use of AI and that the use

741
01:06:34,480 --> 01:06:36,157
of AI is in accordance with

742
01:06:36,240 --> 01:06:39,437
international law. In this regard, we

743
01:06:39,520 --> 01:06:41,437
call on all states to ensure human

744
01:06:41,520 --> 01:06:44,637
control of AI and accountability for the

745
01:06:44,720 --> 01:06:48,077
development and use of AI, not least in

746
01:06:48,160 --> 01:06:51,197
the military domain where accountability

747
01:06:51,280 --> 01:06:55,997
is absolutely essential. Mr. President,
Denmark will continue to support the

748
01:06:56,080 --> 01:07:00,157
meaningful engagement of the security
council and promote an inclusive

749
01:07:00,240 --> 01:07:02,477
multistakeholder approach to AI

750
01:07:02,560 --> 01:07:04,797
governance. Civil society and the

751
01:07:04,880 --> 01:07:07,917
private sector must play leading roles.

752
01:07:08,000 --> 01:07:11,197
Science and evidence must always inform

753
01:07:11,280 --> 01:07:14,637
our uh policies. We see the independent

754
01:07:14,720 --> 01:07:16,797
international scientific panel as an

755
01:07:16,880 --> 01:07:19,037
important contribution to credible and

756
01:07:19,120 --> 01:07:21,837
evidence-based AI governance. The panel

757
01:07:21,920 --> 01:07:24,957
provides us with a common scientific

758
01:07:25,040 --> 01:07:27,837
reference point. So, Mr. President, in

759
01:07:27,920 --> 01:07:30,957
in in in closing, we need to consider

760
01:07:31,040 --> 01:07:33,837
how AI can contribute positively to

761
01:07:33,920 --> 01:07:35,677
shape international peace and security

762
01:07:35,760 --> 01:07:38,957
and bridge global digital divides. At

763
01:07:39,040 --> 01:07:42,157
the same time, we must remain cleareyed

764
01:07:42,240 --> 01:07:44,717
of the significant risks and start

765
01:07:44,800 --> 01:07:48,397
building some strong u guard rails so we

766
01:07:48,480 --> 01:07:50,797
can manage the global risk in due time.

767
01:07:50,880 --> 01:07:54,120
Thank you.

768
01:07:54,240 --> 01:07:57,197
>> I thank his excellency

769
01:07:57,280 --> 01:07:59,517
Mr. Esmson for his statement and I now

770
01:07:59,600 --> 01:08:02,877
give the floor for Carlos,

771
01:08:02,960 --> 01:08:07,037
Minister of Foreign Affairs for Panama.

772
01:08:07,120 --> 01:08:10,557
>> Thank you, Mr. Minister, Mr. President.

773
01:08:10,640 --> 01:08:13,437
We thank France for convening this

774
01:08:13,520 --> 01:08:15,757
important meeting.

775
01:08:15,840 --> 01:08:19,517
Moreover, we thank Mr. Yoshua Benjio,

776
01:08:19,600 --> 01:08:21,357
co-chair of the independent scientific

777
01:08:21,440 --> 01:08:22,877
panel of the United Nations on

778
01:08:22,960 --> 01:08:26,557
artificial intelligence, Mr. Sam Alman,

779
01:08:26,640 --> 01:08:31,117
CEO of OpenAI, Mr. Dario Amade, CEO of

780
01:08:31,200 --> 01:08:35,517
Anthropic, and Mr. Clemont Dong, CEO of

781
01:08:35,600 --> 01:08:40,360
Hugging Face, for their
valuable presentations.

782
01:08:40,720 --> 01:08:44,317
Although today's meeting invites
us to consider the risks caused and

783
01:08:44,400 --> 01:08:47,277
exacerbated by artificial intelligence

784
01:08:47,360 --> 01:08:49,917
for Panama, it is necessary to begin

785
01:08:50,000 --> 01:08:52,477
this statement by highlighting its

786
01:08:52,560 --> 01:08:54,797
potential to transform the world in

787
01:08:54,880 --> 01:08:57,117
which we live.

788
01:08:57,200 --> 01:08:59,837
The capacity of these technologies to

789
01:08:59,920 --> 01:09:02,957
analyze enormous amounts of information,

790
01:09:03,040 --> 01:09:06,717
identify patterns, generate increasingly

791
01:09:06,800 --> 01:09:09,357
sophisticated videos, images, and code,

792
01:09:09,440 --> 01:09:11,917
to automate tasks, and to support

793
01:09:12,000 --> 01:09:14,717
planning and decision-making processes

794
01:09:14,800 --> 01:09:19,277
are all evidence of a transformation
that is already underway, the scope of

795
01:09:19,360 --> 01:09:21,517
which we are barely beginning to

796
01:09:21,600 --> 01:09:25,597
understand. These tools have a wide

797
01:09:25,680 --> 01:09:27,597
array of applications that can benefit

798
01:09:27,680 --> 01:09:30,717
humanity. In Panama, our own national

799
01:09:30,800 --> 01:09:35,517
artificial intelligence strategy
is identifying opportunities in sectors

800
01:09:35,600 --> 01:09:37,197
such as healthcare, agriculture,

801
01:09:37,280 --> 01:09:39,997
education, and logistics where these

802
01:09:40,080 --> 01:09:42,797
technologies can offer solutions which

803
01:09:42,880 --> 01:09:45,197
just a few years ago would have seemed

804
01:09:45,280 --> 01:09:47,357
unimaginable.

805
01:09:47,440 --> 01:09:51,757
And it is specifically in this area of

806
01:09:51,840 --> 01:09:54,637
progress where if applied to other

807
01:09:54,720 --> 01:09:56,877
purposes, we also see the main risks

808
01:09:56,960 --> 01:09:59,037
emerging. Artificial intelligence can

809
01:09:59,120 --> 01:10:01,757
also be used to amplify disinformation

810
01:10:01,840 --> 01:10:04,797
campaigns to allow for cyber attacks

811
01:10:04,880 --> 01:10:07,517
against critical infrastructure and to

812
01:10:07,600 --> 01:10:12,397
facilitate access to knowledge for the
development of weapons among other uses

813
01:10:12,480 --> 01:10:16,157
that are already a threat
to international security. And these risks

814
01:10:16,240 --> 01:10:18,077
are not hypothetical scenarios. We are

815
01:10:18,160 --> 01:10:22,797
beginning to see them come to pass in

816
01:10:22,880 --> 01:10:26,557
situations that are directly related
to this council. There are recent reports

817
01:10:26,640 --> 01:10:31,677
on the use of artificial intelligence by
non-state actors to develop guidance and

818
01:10:31,760 --> 01:10:34,957
navigation systems
for rockets and missiles.

819
01:10:35,040 --> 01:10:39,597
On top of that, there are reports
of autonomous weapons being used that are

820
01:10:39,680 --> 01:10:41,517
able to navigate and select targets

821
01:10:41,600 --> 01:10:44,317
completely independently or with very

822
01:10:44,400 --> 01:10:47,117
limited human intervention.

823
01:10:47,200 --> 01:10:49,837
It is precisely this growing autonomy

824
01:10:49,920 --> 01:10:54,797
which raises one of the most serious
concerns that was noted in the concept

825
01:10:54,880 --> 01:10:58,317
note for this meeting. That is the
possibility that the evolution of these

826
01:10:58,400 --> 01:11:03,037
systems could exceed our ability
to maintain effective human control of

827
01:11:03,120 --> 01:11:06,877
them. There are decisions
which by their very

828
01:11:06,960 --> 01:11:09,597
nature cannot be reduced to a

829
01:11:09,680 --> 01:11:11,437
statistical result, a statistical

830
01:11:11,520 --> 01:11:14,077
outcome. When human lives are at stake,

831
01:11:14,160 --> 01:11:18,877
it is vital that ethical, legal, and
contextual considerations be taken into

832
01:11:18,960 --> 01:11:20,957
account, which no calculation or

833
01:11:21,040 --> 01:11:25,117
algorithmic system can replace.

834
01:11:25,200 --> 01:11:27,757
This concern for control is not limited

835
01:11:27,840 --> 01:11:31,437
to extreme scenarios. As we begin to use

836
01:11:31,520 --> 01:11:35,117
analysis produced by artificial
intelligence to steer and guide our

837
01:11:35,200 --> 01:11:39,357
decisions, it is essential that those
that are in leadership positions uh be

838
01:11:39,440 --> 01:11:44,157
able to understand how these tools have
arrived at a certain recommendation,

839
01:11:44,240 --> 01:11:45,837
what information underpins that

840
01:11:45,920 --> 01:11:47,837
recommendation, and what the margin of

841
01:11:47,920 --> 01:11:51,757
error is. Mr. President recognizing the

842
01:11:51,840 --> 01:11:53,837
risks of artificial intelligence does

843
01:11:53,920 --> 01:11:57,037
not mean ignoring the benefits that this

844
01:11:57,120 --> 01:11:59,197
technology is already providing our

845
01:11:59,280 --> 01:12:02,557
societies. The challenge is finding a

846
01:12:02,640 --> 01:12:07,197
balance that allows us to take advantage
of these opportunities while also

847
01:12:07,280 --> 01:12:11,437
responding to the risks with uh
responsibly. In that regard, Panama

848
01:12:11,520 --> 01:12:15,357
recognizes the important role
of the International Scientific Panel on

849
01:12:15,440 --> 01:12:17,677
artificial intelligence to provide

850
01:12:17,760 --> 01:12:22,557
impartial and scientific information
that can guide the discussions of the

851
01:12:22,640 --> 01:12:24,077
international community on these

852
01:12:24,160 --> 01:12:27,917
technologies. Equally, we welcome the

853
01:12:28,000 --> 01:12:30,557
initiative of the tech sector actors

854
01:12:30,640 --> 01:12:32,957
here present to proactively share

855
01:12:33,040 --> 01:12:35,677
information about specific cases of the

856
01:12:35,760 --> 01:12:37,997
misuse of these technologies and also

857
01:12:38,080 --> 01:12:40,797
about incidents in which their growing

858
01:12:40,880 --> 01:12:43,917
autonomy has generated concerns about

859
01:12:44,000 --> 01:12:48,557
the possible loss
of human control of them.

860
01:12:48,640 --> 01:12:50,557
President,

861
01:12:50,640 --> 01:12:52,397
>> this meeting is evidence that in order

862
01:12:52,480 --> 01:12:55,917
to address the range of risks related to

863
01:12:56,000 --> 01:13:00,957
artificial intelligence, closer
cooperation between states, the tech

864
01:13:01,040 --> 01:13:02,877
sector, and other relevant actors will

865
01:13:02,960 --> 01:13:05,597
be essential. To that end, it is

866
01:13:05,680 --> 01:13:07,997
essential that we establish mechanisms

867
01:13:08,080 --> 01:13:11,037
for a more structured and permanent

868
01:13:11,120 --> 01:13:13,277
exchange of information on emerging

869
01:13:13,360 --> 01:13:16,637
risks, incidents, misuse, and good

870
01:13:16,720 --> 01:13:21,277
practices. artificial
intelligence will inevitably

871
01:13:21,360 --> 01:13:23,277
continue to evolve and continue to

872
01:13:23,360 --> 01:13:26,397
transform the world in which we live

873
01:13:26,480 --> 01:13:31,037
and the testimonies that we've
heard today are evidence of that. Our

874
01:13:31,120 --> 01:13:33,197
responsibility is to make sure that this

875
01:13:33,280 --> 01:13:35,837
progress continues to be in the service

876
01:13:35,920 --> 01:13:38,317
of people and that it contributes to the

877
01:13:38,400 --> 01:13:40,557
construction of safer, more peaceful,

878
01:13:40,640 --> 01:13:45,437
more prosperous societies with a
perspective of equity that allows all of

879
01:13:45,520 --> 01:13:49,517
the nations of the world including the
global south to participate and benefit

880
01:13:49,600 --> 01:13:52,317
from its development. Thank you very much.

881
01:13:52,400 --> 01:13:55,197
I thank his excellency for his statement

882
01:13:55,280 --> 01:13:58,237
and I now give the floor to

883
01:13:58,320 --> 01:14:02,877
his excellency Christos menopolis deputy
minister for digital governance and AI

884
01:14:02,960 --> 01:14:05,997
of Greece. >> Mr.
President, I wish to begin by

885
01:14:06,080 --> 01:14:10,477
thanking France along along with the
briefers for their remarks and for

886
01:14:10,560 --> 01:14:13,757
bringing to the attention of the
security council the crucial issue of

887
01:14:13,840 --> 01:14:16,157
the link between the rapid technological

888
01:14:16,240 --> 01:14:20,957
advances in the field of AI
and international peace and security. A

889
01:14:21,040 --> 01:14:23,117
subject that we will also tackle during

890
01:14:23,200 --> 01:14:26,877
our presidency. AI is already changing

891
01:14:26,960 --> 01:14:28,557
the way we think about international

892
01:14:28,640 --> 01:14:30,957
peace and security. It can help us

893
01:14:31,040 --> 01:14:34,157
prevent conflict, improve early warning,

894
01:14:34,240 --> 01:14:37,197
and make UN operations more effective.

895
01:14:37,280 --> 01:14:39,277
But it also creates risks that we are

896
01:14:39,360 --> 01:14:41,597
only beginning to understand. For

897
01:14:41,680 --> 01:14:44,477
Greece, our starting point is clear. AI

898
01:14:44,560 --> 01:14:47,357
must remain humanentric, secure, and

899
01:14:47,440 --> 01:14:50,157
trustworthy. It development and use must

900
01:14:50,240 --> 01:14:52,157
respect international law including the

901
01:14:52,240 --> 01:14:57,357
UN charter, international humanitarian
law and international human rights law

902
01:14:57,440 --> 01:15:00,957
and human responsibility and
accountability must remain at the center

903
01:15:01,040 --> 01:15:03,677
of our approach. Our national approach

904
01:15:03,760 --> 01:15:06,397
combines innovation with governance.

905
01:15:06,480 --> 01:15:08,477
Greece has established the special

906
01:15:08,560 --> 01:15:11,517
secretariat for AI and data governance,

907
01:15:11,600 --> 01:15:16,717
implemented the national law for the EU
AI act, and is investing in strategic

908
01:15:16,800 --> 01:15:18,477
infrastructure including the

909
01:15:18,560 --> 01:15:22,157
supercomputer and the Faros AI factory.

910
01:15:22,240 --> 01:15:25,437
In April 2025, together with France and

911
01:15:25,520 --> 01:15:27,757
the Republic of Korea, Greece organized

912
01:15:27,840 --> 01:15:29,997
an area formula meeting on harnessing

913
01:15:30,080 --> 01:15:33,117
safe, inclusive, and trustworthy AI for

914
01:15:33,200 --> 01:15:37,437
the maintenance of international peace
and security. Today's discussion give us

915
01:15:37,520 --> 01:15:39,677
an opportunity to take that conversation

916
01:15:39,760 --> 01:15:43,277
further. Allow me to make three points.

917
01:15:43,360 --> 01:15:46,157
First, we need to better understand the

918
01:15:46,240 --> 01:15:49,677
risks. AI can process enormous amounts

919
01:15:49,760 --> 01:15:52,077
of information and help us make better

920
01:15:52,160 --> 01:15:54,797
informed decisions. But speed is not

921
01:15:54,880 --> 01:15:57,517
always an advantage. In situations of

922
01:15:57,600 --> 01:15:59,677
crisis, AI can compress the time

923
01:15:59,760 --> 01:16:01,917
available for decisions and encourage

924
01:16:02,000 --> 01:16:05,277
excessive reliance on automated outputs.

925
01:16:05,360 --> 01:16:09,037
As AI increasingly enters
decision-making processes, we must

926
01:16:09,120 --> 01:16:11,517
ensure that greater speed does not come

927
01:16:11,600 --> 01:16:15,037
at the expense of sound human judgment.

928
01:16:15,120 --> 01:16:17,037
These concerns become even more serious

929
01:16:17,120 --> 01:16:19,517
when AI enters military decision-

930
01:16:19,600 --> 01:16:22,077
making. Technology can assist human

931
01:16:22,160 --> 01:16:24,957
judgment. It cannot replace it.

932
01:16:25,040 --> 01:16:26,797
Responsibility and accountability for

933
01:16:26,880 --> 01:16:29,277
critical decisions must always remain

934
01:16:29,360 --> 01:16:32,877
human. Nowhere in is this more important

935
01:16:32,960 --> 01:16:35,677
than in the nuclear domain. We must be

936
01:16:35,760 --> 01:16:38,077
extremely cautious about not integrating

937
01:16:38,160 --> 01:16:41,357
AI into nuclear command and control. The

938
01:16:41,440 --> 01:16:43,517
risk of miscalculation and unintended

939
01:16:43,600 --> 01:16:46,877
escalation are simply too great.

940
01:16:46,960 --> 01:16:48,877
Decisions of such consequence must

941
01:16:48,960 --> 01:16:51,517
remain under human control.

942
01:16:51,600 --> 01:16:54,157
Second, we need practical ways to reduce

943
01:16:54,240 --> 01:16:57,357
these risks. This starts with dialogue.

944
01:16:57,440 --> 01:17:01,437
We need greater transparency,
confidence building and a better common

945
01:17:01,520 --> 01:17:03,277
understanding of how international law

946
01:17:03,360 --> 01:17:05,757
applies to the development and use of

947
01:17:05,840 --> 01:17:09,757
AI. Important work is already under way.

948
01:17:09,840 --> 01:17:11,917
The reaim process and the political

949
01:17:12,000 --> 01:17:14,237
declaration responsible military use of

950
01:17:14,320 --> 01:17:17,517
AI and autonomy have helped advance the

951
01:17:17,600 --> 01:17:22,557
international discussion on responsible
military uses of AI. The United Nations

952
01:17:22,640 --> 01:17:24,477
has an essential role in bringing these

953
01:17:24,560 --> 01:17:27,277
different efforts together. The Security

954
01:17:27,360 --> 01:17:30,157
Council also has a role to play not in

955
01:17:30,240 --> 01:17:32,317
regulating technology and not in

956
01:17:32,400 --> 01:17:35,277
duplicating the work of other UN bodies.

957
01:17:35,360 --> 01:17:39,597
Our responsibility is to understand
what these technologies mean for

958
01:17:39,680 --> 01:17:41,917
international peace and security and to

959
01:17:42,000 --> 01:17:44,877
act when those implications fall within

960
01:17:44,960 --> 01:17:48,557
our mandate. Third, addressing risk

961
01:17:48,640 --> 01:17:51,357
should not mean holding back innovation.

962
01:17:51,440 --> 01:17:53,277
We should make use of the opportunities

963
01:17:53,360 --> 01:17:55,997
AI provides while putting appropriate

964
01:17:56,080 --> 01:17:58,717
safeguards in place and we cannot have

965
01:17:58,800 --> 01:18:00,717
this discussion without those who

966
01:18:00,800 --> 01:18:03,997
develop and use these technologies. We

967
01:18:04,080 --> 01:18:06,157
therefore support close engagement with

968
01:18:06,240 --> 01:18:09,117
academia, civil society and the private

969
01:18:09,200 --> 01:18:11,677
sector alongside states and the United

970
01:18:11,760 --> 01:18:14,157
Nations. Such engagement should be

971
01:18:14,240 --> 01:18:17,117
inclusive, geographically representative

972
01:18:17,200 --> 01:18:21,037
and gender balanced. Mr. President, AI

973
01:18:21,120 --> 01:18:23,037
will increasingly affect international

974
01:18:23,120 --> 01:18:25,597
peace and security. We cannot stop

975
01:18:25,680 --> 01:18:28,157
technological change, nor should we seek

976
01:18:28,240 --> 01:18:31,037
to. But we can work together to ensure

977
01:18:31,120 --> 01:18:33,517
that human beings remain responsible for

978
01:18:33,600 --> 01:18:36,957
decisions that matter most. For Greece,

979
01:18:37,040 --> 01:18:41,357
this means keeping international law,
human responsibility, and accountability

980
01:18:41,440 --> 01:18:43,757
at the center of our approach. It also

981
01:18:43,840 --> 01:18:46,717
means keeping this council focused on

982
01:18:46,800 --> 01:18:51,117
its primarily responsibility,
the maintenance of international peace and

983
01:18:51,200 --> 01:18:53,517
security. So, Greece will continue to

984
01:18:53,600 --> 01:18:55,437
work with all members of the council

985
01:18:55,520 --> 01:18:59,597
towards this objective. Thank you.

986
01:18:59,680 --> 01:19:02,237
>> I thank his excellency Mr. Demanopulos

987
01:19:02,320 --> 01:19:04,557
for his statement. I now give the floor

988
01:19:04,640 --> 01:19:07,757
to his excellency Mr. Ali Muhammad Omar,

989
01:19:07,840 --> 01:19:12,760
state minister for foreign
affairs of Somalia.

990
01:19:15,120 --> 01:19:17,117
Mr. President, I thank France for

991
01:19:17,200 --> 01:19:20,157
convening this timely debate. I also

992
01:19:20,240 --> 01:19:24,077
thank Mr. Banio, Oldman, Almadeo, and

993
01:19:24,160 --> 01:19:27,677
Dingua for their insightful briefing.

994
01:19:27,760 --> 01:19:29,597
Artificial intelligence is no longer

995
01:19:29,680 --> 01:19:32,077
only a technological or a developmental

996
01:19:32,160 --> 01:19:35,757
issue. is our current reality. Rapidly

997
01:19:35,840 --> 01:19:38,477
transforming society,

998
01:19:38,560 --> 01:19:41,037
economies and our approach to peace and

999
01:19:41,120 --> 01:19:43,997
security. It is role in a cyber

1000
01:19:44,080 --> 01:19:46,477
operationist disinformation, critical

1001
01:19:46,560 --> 01:19:48,877
infrastructure and military decision

1002
01:19:48,960 --> 01:19:51,277
making places it firmly within the

1003
01:19:51,360 --> 01:19:53,597
United Nations mandate.

1004
01:19:53,680 --> 01:19:55,997
We must prevent AI from becoming a

1005
01:19:56,080 --> 01:19:59,197
driver of a conflict while actively

1006
01:19:59,280 --> 01:20:02,477
supporting its responsible use in early

1007
01:20:02,560 --> 01:20:04,557
warning, crisis prevention,

1008
01:20:04,640 --> 01:20:07,037
peacekeeping, humanitarian action and

1009
01:20:07,120 --> 01:20:09,997
civilian protection. In that vein, four

1010
01:20:10,080 --> 01:20:11,917
core principles

1011
01:20:12,000 --> 01:20:15,837
should guide us. Number one, first the

1012
01:20:15,920 --> 01:20:17,917
human responsibility must remain

1013
01:20:18,000 --> 01:20:21,277
absolute. AI must operate strictly

1014
01:20:21,360 --> 01:20:23,197
within the framework of international

1015
01:20:23,280 --> 01:20:26,157
law the purposes and the principles of

1016
01:20:26,240 --> 01:20:28,397
the UN charter and the respect of her

1017
01:20:28,480 --> 01:20:31,037
human rights and human dignity.

1018
01:20:31,120 --> 01:20:34,157
States cannot delegate legal, ethical

1019
01:20:34,240 --> 01:20:38,077
and moral accountability
to an algorithmist.

1020
01:20:38,160 --> 01:20:40,477
We welcome the declaration affirming

1021
01:20:40,560 --> 01:20:42,797
that AI must remain under human

1022
01:20:42,880 --> 01:20:47,437
direction, insight and control.

1023
01:20:47,520 --> 01:20:49,677
This principle is indispensable in

1024
01:20:49,760 --> 01:20:52,957
decision involving the use of force,

1025
01:20:53,040 --> 01:20:55,997
autonomous weapon system and nuclear

1026
01:20:56,080 --> 01:20:59,917
command and control. Second, safety must

1027
01:21:00,000 --> 01:21:03,357
must come before a deployment.

1028
01:21:03,440 --> 01:21:06,157
Recent incidents involving AI agent

1029
01:21:06,240 --> 01:21:09,357
reportedly bypassing safeguards,

1030
01:21:09,440 --> 01:21:12,157
concealing their actions and obtaining

1031
01:21:12,240 --> 01:21:13,757
unauthorized

1032
01:21:13,840 --> 01:21:17,277
access raises a serious question about

1033
01:21:17,360 --> 01:21:22,237
whether existing controllers can keep
base with the increasing autonomous

1034
01:21:22,320 --> 01:21:25,917
systems. Private companies
develop and control

1035
01:21:26,000 --> 01:21:28,637
many of these systems, but commercial

1036
01:21:28,720 --> 01:21:31,357
incentives cannot determine how much

1037
01:21:31,440 --> 01:21:35,837
risky a human must accept. We therefore

1038
01:21:35,920 --> 01:21:39,517
support a mandatory incident reporting,

1039
01:21:39,600 --> 01:21:41,677
layered technical safeguards,

1040
01:21:41,760 --> 01:21:44,717
independent audits and measures to

1041
01:21:44,800 --> 01:21:47,917
mitigate algorithmic bias and protect

1042
01:21:48,000 --> 01:21:50,397
vulnerable populations.

1043
01:21:50,480 --> 01:21:52,957
When agreed capability threshold are

1044
01:21:53,040 --> 01:21:56,157
crossed, deployment must be suspended

1045
01:21:56,240 --> 01:22:00,317
until risks can be reliably managed. The

1046
01:22:00,400 --> 01:22:02,877
Secretary General's proposal to explore

1047
01:22:02,960 --> 01:22:05,917
an international institution capable of

1048
01:22:06,000 --> 01:22:09,597
setting standards enabling verification

1049
01:22:09,680 --> 01:22:12,077
and convening states when such threshold

1050
01:22:12,160 --> 01:22:14,877
are crossed deserves a careful

1051
01:22:14,960 --> 01:22:18,477
consideration. Third, governance must

1052
01:22:18,560 --> 01:22:21,917
address Africa's specific vulnerability

1053
01:22:22,000 --> 01:22:25,117
and ensure equitable access. Limited

1054
01:22:25,200 --> 01:22:27,597
computing infrastructure, unreliable

1055
01:22:27,680 --> 01:22:30,557
connectivity, shortages of specialized

1056
01:22:30,640 --> 01:22:33,917
expert and fragmented regulatory

1057
01:22:34,000 --> 01:22:37,357
capacity constrain the the ability of

1058
01:22:37,440 --> 01:22:40,717
many African states to evaluate advanced

1059
01:22:40,800 --> 01:22:43,917
systems. modelist train it on a foreign

1060
01:22:44,000 --> 01:22:47,437
data may misrepresent African language

1061
01:22:47,520 --> 01:22:51,197
and context while synthetic synthetic

1062
01:22:51,280 --> 01:22:55,517
data can inflame a social tensionist

1063
01:22:55,600 --> 01:22:58,397
interfere in a fragile be processes and

1064
01:22:58,480 --> 01:23:00,637
undermine elections.

1065
01:23:00,720 --> 01:23:03,917
Africa cannot remain merely a market for

1066
01:23:04,000 --> 01:23:07,277
system systems developed elsewhere, a

1067
01:23:07,360 --> 01:23:09,917
source of a road data or a testing

1068
01:23:10,000 --> 01:23:13,597
ground for a technology it did not help

1069
01:23:13,680 --> 01:23:17,117
govern. We must we must prevent new

1070
01:23:17,200 --> 01:23:20,477
forms of technological dependency and

1071
01:23:20,560 --> 01:23:24,957
digital digital coloniz colonialism

1072
01:23:25,040 --> 01:23:27,437
by supporting a regional framework

1073
01:23:27,520 --> 01:23:30,077
particularly partnership with African

1074
01:23:30,160 --> 01:23:32,797
Union and strategies like the ban

1075
01:23:32,880 --> 01:23:35,197
African AI initiative.

1076
01:23:35,280 --> 01:23:37,677
Developing countries require a targeted

1077
01:23:37,760 --> 01:23:40,477
international cooperation, financing to

1078
01:23:40,560 --> 01:23:43,277
bridge the digital and infrastructure,

1079
01:23:43,360 --> 01:23:46,557
energy and skills divide alongside

1080
01:23:46,640 --> 01:23:49,197
access to local data sets, evaluation

1081
01:23:49,280 --> 01:23:52,717
tools and computing power. Ensuring

1082
01:23:52,800 --> 01:23:55,277
every nation has the right to develop,

1083
01:23:55,360 --> 01:24:00,557
choose and assess AI technology
is essential for a national digital

1084
01:24:00,640 --> 01:24:04,237
sovereignty sovereignty
and our shared global

1085
01:24:04,320 --> 01:24:07,357
stability. Fourth, the United Nation

1086
01:24:07,440 --> 01:24:10,717
must serve as the main global

1087
01:24:10,800 --> 01:24:13,357
as the main general for global govern AI

1088
01:24:13,440 --> 01:24:17,757
governance. Somali supporters
international normies anchored in the

1089
01:24:17,840 --> 01:24:20,317
universal legitimacy of the UN rather

1090
01:24:20,400 --> 01:24:22,877
than fragmented rules established by

1091
01:24:22,960 --> 01:24:25,757
small number of technological advanced

1092
01:24:25,840 --> 01:24:28,317
states and private corporations that

1093
01:24:28,400 --> 01:24:30,877
risk reinforcing existing global

1094
01:24:30,960 --> 01:24:32,957
inequalities.

1095
01:24:33,040 --> 01:24:36,237
Global governance must uphold a people

1096
01:24:36,320 --> 01:24:39,597
centered approach that directs AI toward

1097
01:24:39,680 --> 01:24:42,557
human well-being and common good.

1098
01:24:42,640 --> 01:24:45,837
developing regional regionist Africa in

1099
01:24:45,920 --> 01:24:48,317
particular must must be granted a

1100
01:24:48,400 --> 01:24:51,437
genuine representation and a voice in

1101
01:24:51,520 --> 01:24:54,477
these multilateral processes

1102
01:24:54,560 --> 01:24:56,957
alongside the institution institutional

1103
01:24:57,040 --> 01:24:59,437
and technological capacity to evaluate

1104
01:24:59,520 --> 01:25:02,877
AI systems independently

1105
01:25:02,960 --> 01:25:05,197
rather than relying on external

1106
01:25:05,280 --> 01:25:07,597
determination. The measures of our

1107
01:25:07,680 --> 01:25:10,637
success will not be the sophistication

1108
01:25:10,720 --> 01:25:13,357
of our declaration, but whether we

1109
01:25:13,440 --> 01:25:17,037
prevent the first AI enabled crisis from

1110
01:25:17,120 --> 01:25:18,797
becoming the next international

1111
01:25:18,880 --> 01:25:22,477
conflict. The power of AI intelligence

1112
01:25:22,560 --> 01:25:26,077
does not entail inevitable outcome. It

1113
01:25:26,160 --> 01:25:28,957
shaped and and and given meaning by our

1114
01:25:29,040 --> 01:25:31,597
values and our choices. The technology

1115
01:25:31,680 --> 01:25:34,237
is advancing quickly. Our collective

1116
01:25:34,320 --> 01:25:37,277
responsibility must advance faster. I

1117
01:25:37,360 --> 01:25:39,837
thank you.

1118
01:25:39,920 --> 01:25:42,317
>> I thank his excellency Mr. Omar for his

1119
01:25:42,400 --> 01:25:45,357
statement. I now give the floor to his

1120
01:25:45,440 --> 01:25:50,077
excellency Mr. Ed Milliband, Minister
of Foreign Affairs. Uh Minister Secretary

1121
01:25:50,160 --> 01:25:52,317
of State for Foreign Commonwealth and
Development Affairs of the United Kingdom.

1122
01:25:52,400 --> 01:25:57,357
>> Uh can I really thank you for convening
this very very important uh session? And

1123
01:25:57,440 --> 01:26:01,357
I also want to put on record my thanks
to the expert speakers uh who have

1124
01:26:01,440 --> 01:26:03,597
provided such an illuminating uh

1125
01:26:03,680 --> 01:26:06,877
contribution uh to this session.

1126
01:26:06,960 --> 01:26:08,557
I think we should stand back and take

1127
01:26:08,640 --> 01:26:12,237
stock of what confronts us uh in this

1128
01:26:12,320 --> 01:26:15,117
session of the security council because

1129
01:26:15,200 --> 01:26:17,677
AI is a technology which undoubtedly has

1130
01:26:17,760 --> 01:26:19,677
the power to transform our world for the

1131
01:26:19,760 --> 01:26:22,157
better. I'm sure we all see it in our

1132
01:26:22,240 --> 01:26:25,117
daily lives. We see it in all of our

1133
01:26:25,200 --> 01:26:27,917
countries. But today the founders of

1134
01:26:28,000 --> 01:26:30,877
this technology are coming to us with a

1135
01:26:30,960 --> 01:26:34,637
stark warning. They are telling us not

1136
01:26:34,720 --> 01:26:37,517
just that they relish the chance of this

1137
01:26:37,600 --> 01:26:41,597
technology to do good, but they fear its

1138
01:26:41,680 --> 01:26:44,957
ability to do harm.

1139
01:26:45,040 --> 01:26:48,397
They are telling us in stark terms that

1140
01:26:48,480 --> 01:26:51,677
it cannot just be left to them to

1141
01:26:51,760 --> 01:26:55,037
prevent disaster from befalling the

1142
01:26:55,120 --> 01:26:56,557
world.

1143
01:26:56,640 --> 01:27:00,557
I say today we cannot ignore them

1144
01:27:00,640 --> 01:27:03,597
because they are right.

1145
01:27:03,680 --> 01:27:05,517
Because we cannot abdicate our

1146
01:27:05,600 --> 01:27:08,557
responsibility to our citizens. Because

1147
01:27:08,640 --> 01:27:11,277
we cannot outsource to private companies

1148
01:27:11,360 --> 01:27:14,717
the first duty of government to protect

1149
01:27:14,800 --> 01:27:19,117
our people. Whatever the issue,

1150
01:27:19,200 --> 01:27:23,037
this security council was founded
to address the great collective challenges

1151
01:27:23,120 --> 01:27:26,077
of our time. So it is for governments

1152
01:27:26,160 --> 01:27:30,717
for governments as many of the people
who have made contributions today from

1153
01:27:30,800 --> 01:27:34,717
the tech world have said for all of us
around tables such as this to work

1154
01:27:34,800 --> 01:27:39,037
together and together with industry
to deliver in the interests of the people

1155
01:27:39,120 --> 01:27:41,357
that have sent us here and that is what

1156
01:27:41,440 --> 01:27:44,477
the UK government is committed to doing.

1157
01:27:44,560 --> 01:27:46,957
We must start with safety. This means

1158
01:27:47,040 --> 01:27:49,197
working harder to ensure frontier AI

1159
01:27:49,280 --> 01:27:52,317
models are rigorously tested and

1160
01:27:52,400 --> 01:27:54,637
building a system of assurance around

1161
01:27:54,720 --> 01:27:56,717
model development to allow us to

1162
01:27:56,800 --> 01:28:01,197
establish trust that models are safe.

1163
01:28:01,280 --> 01:28:03,517
Second, transparency.

1164
01:28:03,600 --> 01:28:08,237
The companies building the models cannot
simply be left to their own devices.

1165
01:28:08,320 --> 01:28:12,877
Governments must have the visibility
to understand and assess what AI companies

1166
01:28:12,960 --> 01:28:14,317
are doing.

1167
01:28:14,400 --> 01:28:16,957
The leading AI companies have actually

1168
01:28:17,040 --> 01:28:19,917
committed to provide this visibility.

1169
01:28:20,000 --> 01:28:24,717
It's really really important and it
is an offer that we and they should follow

1170
01:28:24,800 --> 01:28:26,637
through on.

1171
01:28:26,720 --> 01:28:29,757
Third, resilience. We have to harden our

1172
01:28:29,840 --> 01:28:32,557
defenses to reduce the impact of AI

1173
01:28:32,640 --> 01:28:34,717
enabled incidents.

1174
01:28:34,800 --> 01:28:37,597
Even if we test every model, we must be

1175
01:28:37,680 --> 01:28:41,517
prepared for AI to be misused. That

1176
01:28:41,600 --> 01:28:43,517
includes boosting our cyber defenses,

1177
01:28:43,600 --> 01:28:45,597
protecting critical infrastructure, and

1178
01:28:45,680 --> 01:28:49,597
reinforcing our financial systems.

1179
01:28:49,680 --> 01:28:53,357
And we must support our populations to
manage and adjust to the disruption that

1180
01:28:53,440 --> 01:28:57,997
AI will bring to how we live and work,
as well as to harness its incredible

1181
01:28:58,080 --> 01:29:01,997
benefits. I say today
that the UK will play its

1182
01:29:02,080 --> 01:29:06,717
full part in this process.
We established the first AI security

1183
01:29:06,800 --> 01:29:09,917
institute which shares its leading
expertise, testing capabilities and

1184
01:29:10,000 --> 01:29:14,557
research working handin glove with the
United States and the leading labs on AI

1185
01:29:14,640 --> 01:29:16,717
safety. As our prime minister set out

1186
01:29:16,800 --> 01:29:19,517
yesterday, the UK will put AI at the

1187
01:29:19,600 --> 01:29:23,117
heart of its G20 presidency next year.

1188
01:29:23,200 --> 01:29:27,757
Building on our track record of bringing
the right people together on this issue,

1189
01:29:27,840 --> 01:29:32,237
breaking down silos between government,
industry, researchers, and civil

1190
01:29:32,320 --> 01:29:34,717
society. delivering on the potential of

1191
01:29:34,800 --> 01:29:38,557
AI for good and dealing with the risks

1192
01:29:38,640 --> 01:29:42,077
whether in terms of the future of jobs,
protecting children, cyber security,

1193
01:29:42,160 --> 01:29:45,117
biocurity, national security or

1194
01:29:45,200 --> 01:29:48,077
financial stability

1195
01:29:48,160 --> 01:29:50,877
as we also work to agreeing to working

1196
01:29:50,960 --> 01:29:55,597
towards agreeing a set of a single set
of global principles and standards to

1197
01:29:55,680 --> 01:29:57,437
ensure the transparency and access

1198
01:29:57,520 --> 01:30:00,637
necessary to increase our preparedness

1199
01:30:00,720 --> 01:30:03,917
to ensure that AI development is safe

1200
01:30:04,000 --> 01:30:07,197
and as the previous speaker said to

1201
01:30:07,280 --> 01:30:09,837
seize potential for the global south and

1202
01:30:09,920 --> 01:30:13,517
global north alike. Let me just end on

1203
01:30:13,600 --> 01:30:17,277
this uh point. I think overall this is

1204
01:30:17,360 --> 01:30:19,917
not a choice between progress and

1205
01:30:20,000 --> 01:30:24,557
safety. We are only going to be able
to innovate and capture the extraordinary

1206
01:30:24,640 --> 01:30:28,397
benefits of AI which I'm sure we
all share in this room if people have

1207
01:30:28,480 --> 01:30:33,037
confidence that the right safeguards
are in place. So together we must take

1208
01:30:33,120 --> 01:30:35,197
whatever meas measures are necessary so

1209
01:30:35,280 --> 01:30:39,357
that AI is secure, safe and trustworthy

1210
01:30:39,440 --> 01:30:41,357
while still advancing its incredible

1211
01:30:41,440 --> 01:30:44,397
potential to drive growth, cure disease

1212
01:30:44,480 --> 01:30:47,837
and transform people's life chances. I

1213
01:30:47,920 --> 01:30:50,797
end on this point finally.

1214
01:30:50,880 --> 01:30:53,757
We can't say we haven't been warned.

1215
01:30:53,840 --> 01:30:56,317
We have been warned by the people who

1216
01:30:56,400 --> 01:30:58,637
have founded this incredible technology

1217
01:30:58,720 --> 01:31:02,397
with its huge potential for good.

1218
01:31:02,480 --> 01:31:05,757
We must act on their warnings and we

1219
01:31:05,840 --> 01:31:08,317
must act on their warnings together.

1220
01:31:08,400 --> 01:31:10,477
Thank you. >> Show me.

1221
01:31:10,560 --> 01:31:12,557
>> I thank his excellency Mr.
Milliband for

1222
01:31:12,640 --> 01:31:15,437
his statement. I now give the floor to

1223
01:31:15,520 --> 01:31:20,237
his excellency Mr. Michael Katzio,
assistant to the president and director

1224
01:31:20,320 --> 01:31:22,317
of the White House Office of Science and

1225
01:31:22,400 --> 01:31:26,477
Technology Policy, United States.

1226
01:31:26,560 --> 01:31:28,637
>> My sincerest thanks to the Republic of

1227
01:31:28,720 --> 01:31:31,677
France as president of the Council.

1228
01:31:31,760 --> 01:31:36,557
Distinguished colleagues, I am grateful
for another opportunity to discuss the

1229
01:31:36,640 --> 01:31:38,397
most consequential technological

1230
01:31:38,480 --> 01:31:40,877
development of our time, super

1231
01:31:40,960 --> 01:31:44,957
intelligence, formerly known as AI. Our

1232
01:31:45,040 --> 01:31:47,437
primary responsibility is to ensure that

1233
01:31:47,520 --> 01:31:50,157
continued advances in super intelligence

1234
01:31:50,240 --> 01:31:52,957
benefit our citizens and safeguard their

1235
01:31:53,040 --> 01:31:55,757
liberty and way of life.

1236
01:31:55,840 --> 01:31:59,997
The frontier of intelligence
is advancing rapidly.

1237
01:32:00,080 --> 01:32:02,237
That is not a reason to pause its

1238
01:32:02,320 --> 01:32:05,197
further development or to constrain it

1239
01:32:05,280 --> 01:32:09,437
with new global governance structures.

1240
01:32:09,520 --> 01:32:13,437
As the leader in frontier super
intelligence, the the US has taken

1241
01:32:13,520 --> 01:32:17,917
actions to ensure we are able
to keep the frontier moving safely and

1242
01:32:18,000 --> 01:32:21,917
responsibly. We've engaged
Frontier Labs on testing

1243
01:32:22,000 --> 01:32:25,277
and evaluation of new model capabilities,

1244
01:32:25,360 --> 01:32:27,037
established cyber security clearing

1245
01:32:27,120 --> 01:32:29,437
houses for coordination on vulnerability

1246
01:32:29,520 --> 01:32:34,157
mitigation, and clarified the existing
law enforcement tools available to

1247
01:32:34,240 --> 01:32:38,397
address harmful uses
of super intelligence.

1248
01:32:38,480 --> 01:32:42,477
Internationally, the United States
is committed to diplomacy and dialogue on

1249
01:32:42,560 --> 01:32:45,277
emerging technologies, but international

1250
01:32:45,360 --> 01:32:48,317
dialogue in this forum and in others

1251
01:32:48,400 --> 01:32:53,277
cannot be allowed to drift towards
global governance. As President Trump

1252
01:32:53,360 --> 01:32:57,357
said before the General Assembly
yesterday, the United States totally

1253
01:32:57,440 --> 01:33:02,317
rejects any attempt to construct
a globalist scheme of control of super

1254
01:33:02,400 --> 01:33:06,877
intelligence. Only weeks ago,
G20 innovation ministers

1255
01:33:06,960 --> 01:33:09,277
reached consensus on the Carolina

1256
01:33:09,360 --> 01:33:11,677
principles, agreeing that countries

1257
01:33:11,760 --> 01:33:14,077
should preserve national sovereignty in

1258
01:33:14,160 --> 01:33:16,957
the governance of emerging technologies,

1259
01:33:17,040 --> 01:33:21,277
not abdicate responsibility
to international bodies.

1260
01:33:21,360 --> 01:33:25,277
The American people's representatives
will legislate and regulate on the

1261
01:33:25,360 --> 01:33:27,837
American people's behalf. You should do

1262
01:33:27,920 --> 01:33:30,877
the same for your people.

1263
01:33:30,960 --> 01:33:35,037
Only through an open cooperative
ecosystem built on sovereign

1264
01:33:35,120 --> 01:33:40,077
infrastructure and domestic capacity
can the world realize this technology's

1265
01:33:40,160 --> 01:33:43,677
potential. US international
efforts seek to build

1266
01:33:43,760 --> 01:33:46,877
that capacity by sharing our technology

1267
01:33:46,960 --> 01:33:50,237
and providing the tools for adoption.

1268
01:33:50,320 --> 01:33:54,397
We want you, our allies and partners
to share in the benefits from this

1269
01:33:54,480 --> 01:33:56,717
technological revolution.

1270
01:33:56,800 --> 01:33:58,877
That is the purpose of our exports

1271
01:33:58,960 --> 01:34:01,837
program. The US is the only technology

1272
01:34:01,920 --> 01:34:04,157
superpower willing and able to truly

1273
01:34:04,240 --> 01:34:09,117
empower partner nations in your pursuit
of a sovereign super intelligence

1274
01:34:09,200 --> 01:34:13,277
capability. American
companies will work with you to

1275
01:34:13,360 --> 01:34:17,117
build gold standard secure
infrastructure and help you implement

1276
01:34:17,200 --> 01:34:19,997
the applications and development the

1277
01:34:20,080 --> 01:34:24,317
tools and capacity you
need to serve your people.

1278
01:34:24,400 --> 01:34:26,477
In this era, technology sovereignty is

1279
01:34:26,560 --> 01:34:29,757
realized in technology adoption. If you

1280
01:34:29,840 --> 01:34:34,157
take the risks and benefits of super
intelligence seriously, the most

1281
01:34:34,240 --> 01:34:36,637
important thing you can do is to develop

1282
01:34:36,720 --> 01:34:39,357
your country's ability to understand and

1283
01:34:39,440 --> 01:34:42,877
use it. Deploy it widely. Build

1284
01:34:42,960 --> 01:34:45,357
expertise inside your government.

1285
01:34:45,440 --> 01:34:47,037
Prepare your institutions and

1286
01:34:47,120 --> 01:34:49,357
regulators. Understand what these

1287
01:34:49,440 --> 01:34:52,877
systems can and cannot do. You cannot

1288
01:34:52,960 --> 01:34:57,437
govern a technology you do not understand.
This body and others like it

1289
01:34:57,520 --> 01:34:59,837
should focus on sharing best practices

1290
01:34:59,920 --> 01:35:02,637
to build domestic capacity, not

1291
01:35:02,720 --> 01:35:06,077
establishing a global regulatory scheme.

1292
01:35:06,160 --> 01:35:10,797
It is through a policy of permissionless
innovation, a tireless pursuit of the

1293
01:35:10,880 --> 01:35:14,957
frontier, and an optimism about
the future of technology that the United

1294
01:35:15,040 --> 01:35:17,037
States built the greatest innovation

1295
01:35:17,120 --> 01:35:21,197
ecosystem ever known to man. For 250

1296
01:35:21,280 --> 01:35:23,197
years, the American character has sought

1297
01:35:23,280 --> 01:35:25,837
a virtuous balance between innovation

1298
01:35:25,920 --> 01:35:29,517
and safety, independence and security.

1299
01:35:29,600 --> 01:35:33,437
Super intelligence is the latest
in that long series of challenges and

1300
01:35:33,520 --> 01:35:36,957
opportunities.
In the service of humankind, it can

1301
01:35:37,040 --> 01:35:39,437
become the greatest engine of growth and

1302
01:35:39,520 --> 01:35:42,637
prosperity in history. A prosperous

1303
01:35:42,720 --> 01:35:46,957
future will not be secured by a global
regulator. It will be secured by

1304
01:35:47,040 --> 01:35:49,197
sovereign nations that adopt super

1305
01:35:49,280 --> 01:35:51,837
intelligence, responsible companies that

1306
01:35:51,920 --> 01:35:54,637
build it and free people who refuse to

1307
01:35:54,720 --> 01:35:59,960
be ruled by fear. Thank you.

1308
01:36:00,560 --> 01:36:02,797
>> I thank his excellency Mr. Katius for

1309
01:36:02,880 --> 01:36:05,997
his statement. I now give the floor to

1310
01:36:06,080 --> 01:36:08,557
his excellency Mr. Sheikh Abdullahbin

1311
01:36:08,640 --> 01:36:10,877
Ali al- Khalifa, acting under secretary

1312
01:36:10,960 --> 01:36:16,560
for political affairs in the ministry
for foreign affairs of Bahrain.

1313
01:36:17,280 --> 01:36:20,477
Mr. President, allow me

1314
01:36:20,560 --> 01:36:24,797
at the outset to express sincere thanks
and appreciation to the friendly French

1315
01:36:24,880 --> 01:36:26,877
Republic for convening this important

1316
01:36:26,960 --> 01:36:29,597
session. I also extend my thanks to Mr.

1317
01:36:29,680 --> 01:36:33,437
Benjio, co-chair of the UN Independent

1318
01:36:33,520 --> 01:36:35,277
Scientific Panel on Artificial

1319
01:36:35,360 --> 01:36:37,757
Intelligence,

1320
01:36:37,840 --> 01:36:43,757
Mr. Altman, CEO of Open AI, Mr. Amod CEO

1321
01:36:43,840 --> 01:36:48,717
of Anthropic and Mr. Dong CEO of Hugging

1322
01:36:48,800 --> 01:36:52,557
Face for their valuable briefings.

1323
01:36:52,640 --> 01:36:55,277
Mr. President, discussing the

1324
01:36:55,360 --> 01:36:59,917
implications of artificial intelligence
is no longer a matter of intellectual

1325
01:37:00,000 --> 01:37:02,237
interest. It has become an urgent

1326
01:37:02,320 --> 01:37:05,117
necessity dictated by this council's

1327
01:37:05,200 --> 01:37:07,277
responsibilities for maintaining

1328
01:37:07,360 --> 01:37:09,997
international peace and security. Guided

1329
01:37:10,080 --> 01:37:12,237
by this principle, the Kingdom of

1330
01:37:12,320 --> 01:37:14,957
Bahrain emphasized in the statement

1331
01:37:15,040 --> 01:37:17,037
delivered by his Royal Highness Prince

1332
01:37:17,120 --> 01:37:20,077
Salman bin Hammed al- Khalifa, Crown

1333
01:37:20,160 --> 01:37:22,557
Prince and Prime Minister during the

1334
01:37:22,640 --> 01:37:25,597
79th session of the General Assembly.

1335
01:37:25,680 --> 01:37:30,237
The importance of concluding
an international treaty to govern the

1336
01:37:30,320 --> 01:37:32,317
development of artificial intelligence

1337
01:37:32,400 --> 01:37:34,717
and ensure that it serves as a force for

1338
01:37:34,800 --> 01:37:37,517
innovation rather than destruction.

1339
01:37:37,600 --> 01:37:40,157
Consistent with this approach, Bahrain

1340
01:37:40,240 --> 01:37:42,797
recently had the honor of co-hosting an

1341
01:37:42,880 --> 01:37:44,557
interactive dialogue with France and

1342
01:37:44,640 --> 01:37:47,357
Latvia on harnessing this technology for

1343
01:37:47,440 --> 01:37:50,877
conflict resolution. Mr. President, we

1344
01:37:50,960 --> 01:37:53,197
stress that the rapid evolution of

1345
01:37:53,280 --> 01:37:55,677
artificial intelligence expands the

1346
01:37:55,760 --> 01:37:58,397
scope of security threats, particularly

1347
01:37:58,480 --> 01:38:00,957
when these technologies are exploited by

1348
01:38:01,040 --> 01:38:05,037
terrorists and criminal groups.

1349
01:38:05,120 --> 01:38:07,837
As the autonomy of these systems grows,

1350
01:38:07,920 --> 01:38:10,717
we emphasize the need to maintain human

1351
01:38:10,800 --> 01:38:13,197
oversight and accountability at the core

1352
01:38:13,280 --> 01:38:15,597
of decisions with serious security

1353
01:38:15,680 --> 01:38:18,077
implications and ensure their full

1354
01:38:18,160 --> 01:38:20,397
compliance with international law and

1355
01:38:20,480 --> 01:38:23,597
international humanitarian law.

1356
01:38:23,680 --> 01:38:26,957
Unlike automated algorithms, only humans

1357
01:38:27,040 --> 01:38:29,597
exercise reason, responsibility, and

1358
01:38:29,680 --> 01:38:31,997
conscience and possess the ability to

1359
01:38:32,080 --> 01:38:35,117
discern complex contexts. At the same

1360
01:38:35,200 --> 01:38:38,397
time, we must not overlook the immense

1361
01:38:38,480 --> 01:38:40,717
opportunities to harness artificial

1362
01:38:40,800 --> 01:38:43,117
intelligence to enhance early warning

1363
01:38:43,200 --> 01:38:46,237
systems, monitor indicators of tension,

1364
01:38:46,320 --> 01:38:48,717
support mediation efforts, and improve

1365
01:38:48,800 --> 01:38:51,437
effective peacekeeping operations. We

1366
01:38:51,520 --> 01:38:53,837
also stress the need for bridging the

1367
01:38:53,920 --> 01:38:58,717
digital divide and facilitating
technology transfer to ensure that all

1368
01:38:58,800 --> 01:39:00,317
nations can benefit from these

1369
01:39:00,400 --> 01:39:03,517
capabilities. In conclusion, we stress

1370
01:39:03,600 --> 01:39:05,837
the need for the Security Council to

1371
01:39:05,920 --> 01:39:08,797
hold to hold more regular and systematic

1372
01:39:08,880 --> 01:39:11,597
discussions on artificial intelligence.

1373
01:39:11,680 --> 01:39:16,077
These discussions should complement
broader UN efforts to address frameworks

1374
01:39:16,160 --> 01:39:18,077
for the development and testing of these

1375
01:39:18,160 --> 01:39:21,277
models in line with international law.

1376
01:39:21,360 --> 01:39:23,837
We stand ready to contribute to

1377
01:39:23,920 --> 01:39:27,277
advancing these discussions to build a

1378
01:39:27,360 --> 01:39:31,837
responsible and balanced
international approach. One

1379
01:39:31,920 --> 01:39:33,757
that places humans at the heart of

1380
01:39:33,840 --> 01:39:36,637
decision making and ensures that the

1381
01:39:36,720 --> 01:39:39,917
innovations we create today continue to

1382
01:39:40,000 --> 01:39:43,117
serve the fe the peace we all strive to

1383
01:39:43,200 --> 01:39:46,797
uphold in this council. Thank you, Mr.

1384
01:39:46,880 --> 01:39:49,880
President.

1385
01:39:50,880 --> 01:39:53,997
I thank his excellency al Khalifa for

1386
01:39:54,080 --> 01:39:59,800
his statement and I now give the floor
to the representative of Latvia

1387
01:40:00,000 --> 01:40:02,797
president. >> Thank you Mr. President.

1388
01:40:02,880 --> 01:40:06,477
>> First of all I would like to thank
the French presidency for convening this

1389
01:40:06,560 --> 01:40:09,597
meeting on an important and extremely

1390
01:40:09,680 --> 01:40:13,597
urgent subject for international
security. briefers, Professor Yosha

1391
01:40:13,680 --> 01:40:16,557
Benjer, Sam Alman, Dario Amade and

1392
01:40:16,640 --> 01:40:19,677
Clemon Deong for the enlightening

1393
01:40:19,760 --> 01:40:21,997
assessment and also the proposals for

1394
01:40:22,080 --> 01:40:25,837
safe development of the AI.

1395
01:40:25,920 --> 01:40:29,277
Uh, President, the rapid piece of AI

1396
01:40:29,360 --> 01:40:31,597
development is already having profound

1397
01:40:31,680 --> 01:40:34,717
impact across all domains including

1398
01:40:34,800 --> 01:40:37,677
international peace and security.

1399
01:40:37,760 --> 01:40:39,037
Alongside the extraordinary

1400
01:40:39,120 --> 01:40:42,397
opportunities that AI offers, we also

1401
01:40:42,480 --> 01:40:44,877
see growing risks manifesting both in

1402
01:40:44,960 --> 01:40:48,557
peace time and in conflict settings.

1403
01:40:48,640 --> 01:40:51,997
This is a moment for collective action

1404
01:40:52,080 --> 01:40:54,557
as we hear today and as highlighted in

1405
01:40:54,640 --> 01:40:56,957
the call for control of frontier AI

1406
01:40:57,040 --> 01:41:00,237
models initiated by Norway and Finland

1407
01:41:00,320 --> 01:41:03,677
issued on Monday 2 by sec 22nd world

1408
01:41:03,760 --> 01:41:08,557
leaders including the president of Latvia.

1409
01:41:08,640 --> 01:41:11,277
We believe that the security council has

1410
01:41:11,360 --> 01:41:13,437
a responsibility to ensure that the

1411
01:41:13,520 --> 01:41:15,837
implications of AI for international

1412
01:41:15,920 --> 01:41:18,637
peace and security are fully considered

1413
01:41:18,720 --> 01:41:21,117
without duplicating work undertaken in

1414
01:41:21,200 --> 01:41:23,597
other four of course but today's

1415
01:41:23,680 --> 01:41:25,837
discussion demonstrates the council's

1416
01:41:25,920 --> 01:41:29,197
ability to convene member state science

1417
01:41:29,280 --> 01:41:31,677
and industry leaders to address this

1418
01:41:31,760 --> 01:41:35,437
pressing matter. President,

1419
01:41:35,520 --> 01:41:38,397
two distinct sources of AI related risks

1420
01:41:38,480 --> 01:41:40,077
for peace and security are becoming

1421
01:41:40,160 --> 01:41:44,077
evident. Both demand our attention.

1422
01:41:44,160 --> 01:41:48,957
First are the risks stemming from the
misuse of AI by state and non-state

1423
01:41:49,040 --> 01:41:53,917
actors in includes
application of AI in malign

1424
01:41:54,000 --> 01:41:57,837
cyber activities, information
manipulation and even attempts to

1425
01:41:57,920 --> 01:42:00,397
acquire destructive capabilities such as

1426
01:42:00,480 --> 01:42:04,717
biological weapons. Second, a more novel

1427
01:42:04,800 --> 01:42:07,597
and complex challenge as outlined by the

1428
01:42:07,680 --> 01:42:10,477
briefers and the recent reports stems

1429
01:42:10,560 --> 01:42:13,117
from the technology itself such as

1430
01:42:13,200 --> 01:42:15,197
misalignment and risk of loss of

1431
01:42:15,280 --> 01:42:18,317
control. Such behavior risks are source

1432
01:42:18,400 --> 01:42:21,277
of new forms of vulnerability

1433
01:42:21,360 --> 01:42:23,997
including to critical infrastructure.

1434
01:42:24,080 --> 01:42:26,557
Alarmingly, recent incidents indicate

1435
01:42:26,640 --> 01:42:31,037
that not not only can AI systems
act contrary to human intentions and

1436
01:42:31,120 --> 01:42:36,077
interests, but they are becoming better
at manipulating and deceiving humans as

1437
01:42:36,160 --> 01:42:39,997
they do so.
These challenges are not an isolated

1438
01:42:40,080 --> 01:42:43,037
matter as they add to the already

1439
01:42:43,120 --> 01:42:44,797
complex international peace and security

1440
01:42:44,880 --> 01:42:48,237
landscape. actions by misaligned AI may

1441
01:42:48,320 --> 01:42:51,277
fuel further tensions and miscalculation

1442
01:42:51,360 --> 01:42:53,837
due to the difficulty to track, verify,

1443
01:42:53,920 --> 01:42:56,877
and attribute them. The increasing use

1444
01:42:56,960 --> 01:43:01,597
of AI and autonomous systems
in the military domain poses additional

1445
01:43:01,680 --> 01:43:05,917
dilemmas, particularly where
reliability, control, and escalation

1446
01:43:06,000 --> 01:43:08,317
risks intersect.

1447
01:43:08,400 --> 01:43:11,117
President, responsible action starts at

1448
01:43:11,200 --> 01:43:13,677
the industry level, but also with

1449
01:43:13,760 --> 01:43:18,317
governance, clear guard rails,
and capable institutions.

1450
01:43:18,400 --> 01:43:20,557
A multistakeholder approach is essential

1451
01:43:20,640 --> 01:43:24,237
to manage risks and build trust in AI

1452
01:43:24,320 --> 01:43:29,277
through transparency and accountability
through establishment of adequate safety

1453
01:43:29,360 --> 01:43:33,757
protocols for the development and use
of AI systems through independent

1454
01:43:33,840 --> 01:43:35,597
evaluation

1455
01:43:35,680 --> 01:43:37,277
through connecting the government's

1456
01:43:37,360 --> 01:43:41,117
mandate economic expertise civil society

1457
01:43:41,200 --> 01:43:45,197
and industry innovation and all
these require international

1458
01:43:45,280 --> 01:43:48,637
cooperation coordination norms.
reform setting

1459
01:43:48,720 --> 01:43:50,717
exchange of best practices and capacity

1460
01:43:50,800 --> 01:43:53,597
building is key to bridge the divide and

1461
01:43:53,680 --> 01:43:57,917
foster interoperability.
The convening of the UN global dialogue

1462
01:43:58,000 --> 01:44:00,317
on AI governance and the establishment

1463
01:44:00,400 --> 01:44:02,237
of the independent scientific panel on

1464
01:44:02,320 --> 01:44:05,197
AI are important steps.

1465
01:44:05,280 --> 01:44:09,757
Our further efforts should keep up with
the pace of development of frontier AI

1466
01:44:09,840 --> 01:44:13,277
technologies as highlighted
in the call for control

1467
01:44:13,360 --> 01:44:15,917
of frontier AI models. We need an

1468
01:44:16,000 --> 01:44:18,077
open-minded approach by the UN member

1469
01:44:18,160 --> 01:44:20,957
states to build on current international

1470
01:44:21,040 --> 01:44:23,437
mechanisms and explore developing new

1471
01:44:23,520 --> 01:44:26,557
ones. The overall objective should be to

1472
01:44:26,640 --> 01:44:30,797
ensure that technology serves people,
safeguards fundamental rights and

1473
01:44:30,880 --> 01:44:33,437
freedoms and delivers benefits that do

1474
01:44:33,520 --> 01:44:37,917
not come at the expense
of uncontrollable risks. If there is one

1475
01:44:38,000 --> 01:44:39,437
thing that we should be able to agree

1476
01:44:39,520 --> 01:44:42,157
on, AI must remain under human

1477
01:44:42,240 --> 01:44:45,517
direction, oversight, and control.

1478
01:44:45,600 --> 01:44:49,197
President, uh, considering its role
in maintaining international peace and

1479
01:44:49,280 --> 01:44:53,197
security, this council must remain
actively engaged in assessing the

1480
01:44:53,280 --> 01:44:55,357
implications of AI for global peace and

1481
01:44:55,440 --> 01:45:00,237
security. With this technology become
becoming increasingly relevant to the

1482
01:45:00,320 --> 01:45:02,237
council's work in the area of conflict

1483
01:45:02,320 --> 01:45:05,197
prevention and resolution, proactive

1484
01:45:05,280 --> 01:45:07,757
approach is necessary. and building on

1485
01:45:07,840 --> 01:45:10,477
the recent council's inform informal in

1486
01:45:10,560 --> 01:45:13,437
interactive dialogue on on AI that or

1487
01:45:13,520 --> 01:45:18,237
was organized by Latvia and France and
supported by Bahrain and Greece. We are

1488
01:45:18,320 --> 01:45:22,317
committed to taking further concrete
steps. For instance, establishment of

1489
01:45:22,400 --> 01:45:24,477
council's informal expert group that

1490
01:45:24,560 --> 01:45:26,797
would enable more in-depth council

1491
01:45:26,880 --> 01:45:31,997
analysis on the impact of the impact of AI
related technologies on international

1492
01:45:32,080 --> 01:45:36,957
peace and security and the implementation
of the council's mandate.

1493
01:45:37,040 --> 01:45:38,957
Latia will continue to work with

1494
01:45:39,040 --> 01:45:41,437
interested partners on the council in

1495
01:45:41,520 --> 01:45:45,677
the coming month. I thank you.

1496
01:45:45,760 --> 01:45:49,517
I thank the permanent representative
of Latia for her statement. I now give the

1497
01:45:49,600 --> 01:45:55,320
floor to the permanent
representative of Colombia.

1498
01:45:56,000 --> 01:45:59,640
>> Thank you, Mr. President.

1499
01:46:00,560 --> 01:46:03,560
Colombia

1500
01:46:05,200 --> 01:46:07,357
considers

1501
01:46:07,440 --> 01:46:09,917
that here.

1502
01:46:10,000 --> 01:46:12,957
We are not discussing the the benefits

1503
01:46:13,040 --> 01:46:15,917
of artificial intelligence but um really

1504
01:46:16,000 --> 01:46:18,317
it is the risks that come to the four

1505
01:46:18,400 --> 01:46:22,237
the uncontrollable in of unavoidable

1506
01:46:22,320 --> 01:46:25,757
risks and quite clearly following the

1507
01:46:25,840 --> 01:46:31,437
briefings that we have
heard the risk for humanity.

1508
01:46:31,520 --> 01:46:32,717
Mr. President,

1509
01:46:32,800 --> 01:46:36,397
>> we cannot ignore
bluntly the report that

1510
01:46:36,480 --> 01:46:41,197
we have heard is the first time this
security council and is on the record

1511
01:46:41,280 --> 01:46:43,117
for history

1512
01:46:43,200 --> 01:46:46,317
has heard a case where machines commit

1513
01:46:46,400 --> 01:46:48,317
crimes.

1514
01:46:48,400 --> 01:46:52,557
This is not just a random mistake.

1515
01:46:52,640 --> 01:46:55,197
other person if will be other company

1516
01:46:55,280 --> 01:46:58,557
even state secrets. Please try to convey

1517
01:46:58,640 --> 01:47:03,917
the idea for those that produce
intelligence or artificial intelligence

1518
01:47:04,000 --> 01:47:05,517
that

1519
01:47:05,600 --> 01:47:08,557
the large part of the planet does not

1520
01:47:08,640 --> 01:47:11,357
and as such the critical infrastructure

1521
01:47:11,440 --> 01:47:14,317
including cyber security systems will be

1522
01:47:14,400 --> 01:47:16,077
exposed.

1523
01:47:16,160 --> 01:47:19,117
If you read the cases that were

1524
01:47:19,200 --> 01:47:23,117
explained to us today, we will know that

1525
01:47:23,200 --> 01:47:25,357
it will be a crime even against the

1526
01:47:25,440 --> 01:47:28,477
state if this will take place at a state

1527
01:47:28,560 --> 01:47:32,557
level and it can happen.

1528
01:47:32,640 --> 01:47:35,757
The second element uh for us and perhaps

1529
01:47:35,840 --> 01:47:38,957
the major concern is that we haven't

1530
01:47:39,040 --> 01:47:42,477
framed the issue correctly

1531
01:47:42,560 --> 01:47:47,917
and we must make a refrain of the issue.

1532
01:47:48,000 --> 01:47:50,477
I do not think and Colombia does not

1533
01:47:50,560 --> 01:47:53,597
think that the issue here is to limit

1534
01:47:53,680 --> 01:47:55,917
human innovation

1535
01:47:56,000 --> 01:47:59,197
but machine innovation.

1536
01:47:59,280 --> 01:48:03,960
And I explain my words here. Different

1537
01:48:04,880 --> 01:48:07,837
reports has telling us that we face two

1538
01:48:07,920 --> 01:48:11,277
risks. One is misalignment and the other

1539
01:48:11,360 --> 01:48:14,237
is the loss of control. when it comes to

1540
01:48:14,320 --> 01:48:17,837
misalignment is that the machines are

1541
01:48:17,920 --> 01:48:22,637
separating or are finding ways to cheat

1542
01:48:22,720 --> 01:48:26,477
break the instructions that were given

1543
01:48:26,560 --> 01:48:31,037
by humans that is the agents AI agents

1544
01:48:31,120 --> 01:48:33,517
are able to do this repeatedly

1545
01:48:33,600 --> 01:48:35,357
systematically

1546
01:48:35,440 --> 01:48:39,597
and uh so far even in the cyber security

1547
01:48:39,680 --> 01:48:43,117
case that we analyze they were able to

1548
01:48:43,200 --> 01:48:46,637
reach internet and finds way ways to

1549
01:48:46,720 --> 01:48:48,797
cheat the system. Other times they are

1550
01:48:48,880 --> 01:48:51,997
able to find out passwords and thousands

1551
01:48:52,080 --> 01:48:55,117
of messages even against the company

1552
01:48:55,200 --> 01:48:58,157
that was testing this uh machines in the

1553
01:48:58,240 --> 01:49:03,117
sense box. But the other
one and the one that

1554
01:49:03,200 --> 01:49:04,957
should be very very concerning for this

1555
01:49:05,040 --> 01:49:09,517
council is the loss of control.

1556
01:49:09,600 --> 01:49:12,717
So we would like to go with you with a

1557
01:49:12,800 --> 01:49:15,117
logical narrative.

1558
01:49:15,200 --> 01:49:20,477
We are telling us today that AI has an

1559
01:49:20,560 --> 01:49:22,797
immense capability to storage

1560
01:49:22,880 --> 01:49:26,557
information that is far far away from

1561
01:49:26,640 --> 01:49:29,037
human capability.

1562
01:49:29,120 --> 01:49:32,077
We are knowing now that given the models

1563
01:49:32,160 --> 01:49:35,677
that we are adapting but also the design

1564
01:49:35,760 --> 01:49:39,357
of the algorithm and the paths that we

1565
01:49:39,440 --> 01:49:41,757
open in that design.

1566
01:49:41,840 --> 01:49:46,397
These capabilities also allow to involve

1567
01:49:46,480 --> 01:49:49,837
a different type of reasoning each day

1568
01:49:49,920 --> 01:49:53,437
more complex interdisciplinary.

1569
01:49:53,520 --> 01:49:57,437
In other words, soon there won't be any

1570
01:49:57,520 --> 01:50:01,037
human capable of following the

1571
01:50:01,120 --> 01:50:03,917
operation, the development of what we

1572
01:50:04,000 --> 01:50:05,997
are creating.

1573
01:50:06,080 --> 01:50:08,957
In the first days of the models, we

1574
01:50:09,040 --> 01:50:11,597
notice how the machine were able to pass

1575
01:50:11,680 --> 01:50:14,477
a bar exam in three months, a medical

1576
01:50:14,560 --> 01:50:17,117
license exam in three months. So there

1577
01:50:17,200 --> 01:50:19,517
are no humans that we know that can do

1578
01:50:19,600 --> 01:50:22,397
that. But at this state of the art of

1579
01:50:22,480 --> 01:50:25,597
the technology, we are facing a

1580
01:50:25,680 --> 01:50:28,397
different type of situation

1581
01:50:28,480 --> 01:50:33,597
which is a pass from human mistakes to

1582
01:50:33,680 --> 01:50:37,677
machines random mistakes and we have to

1583
01:50:37,760 --> 01:50:40,877
be aware of the risks that we are facing

1584
01:50:40,960 --> 01:50:43,037
and assuming if that is the path we

1585
01:50:43,120 --> 01:50:45,197
choose to take.

1586
01:50:45,280 --> 01:50:48,797
You see the critical issue here is to

1587
01:50:48,880 --> 01:50:51,917
know as a security council

1588
01:50:52,000 --> 01:50:54,877
what are we regulating,

1589
01:50:54,960 --> 01:50:58,317
why we need to regulate it and why we

1590
01:50:58,400 --> 01:51:01,197
should do it.

1591
01:51:01,280 --> 01:51:04,157
Most importantly, Colombia understands

1592
01:51:04,240 --> 01:51:06,717
the position of the United States and

1593
01:51:06,800 --> 01:51:09,437
shares partially

1594
01:51:09,520 --> 01:51:13,837
its practical approach which is

1595
01:51:13,920 --> 01:51:19,037
an international body oversign
the sovereign production of AI is

1596
01:51:19,120 --> 01:51:23,197
impractical for two basic
reasons. There are of

1597
01:51:23,280 --> 01:51:25,597
course constitutional and national

1598
01:51:25,680 --> 01:51:29,357
security reasons in every country. So

1599
01:51:29,440 --> 01:51:32,717
the legal battle will be overwhelming.

1600
01:51:32,800 --> 01:51:35,677
But secondly, if one country stops, the

1601
01:51:35,760 --> 01:51:38,397
other says I do not. Especially those

1602
01:51:38,480 --> 01:51:43,357
that are leading the race.
For the rest of humanity,

1603
01:51:43,440 --> 01:51:45,357
the concerns and for those that are

1604
01:51:45,440 --> 01:51:48,237
behind but will soon catch up. The

1605
01:51:48,320 --> 01:51:52,477
concern is what we do next.

1606
01:51:52,560 --> 01:51:55,917
And that's where our country can perhaps

1607
01:51:56,000 --> 01:51:59,517
offer a different approach.

1608
01:51:59,600 --> 01:52:02,157
And again, no between an inelectable

1609
01:52:02,240 --> 01:52:05,997
choice between safety and development,

1610
01:52:06,080 --> 01:52:08,237
but perhaps

1611
01:52:08,320 --> 01:52:11,437
a way to make sure

1612
01:52:11,520 --> 01:52:15,597
that all humans on Earth are included.

1613
01:52:15,680 --> 01:52:19,037
Not just by shedding or as we use the

1614
01:52:19,120 --> 01:52:23,037
term adapting our countries because it

1615
01:52:23,120 --> 01:52:26,077
seems to us in the global south that

1616
01:52:26,160 --> 01:52:30,477
when we call about adapting is to buy a

1617
01:52:30,560 --> 01:52:33,677
product that we will not produce

1618
01:52:33,760 --> 01:52:36,637
and once we buy the product we can adapt

1619
01:52:36,720 --> 01:52:38,957
our structures and infrastructures.

1620
01:52:39,040 --> 01:52:42,077
Again, partially true, but many of these

1621
01:52:42,160 --> 01:52:45,757
countries will face an undeniable gap

1622
01:52:45,840 --> 01:52:47,997
that will extend even a century between

1623
01:52:48,080 --> 01:52:51,037
another country and the other if we do

1624
01:52:51,120 --> 01:52:55,197
not participate of the production of AI.

1625
01:52:55,280 --> 01:52:58,077
That's one issue and that's a very very

1626
01:52:58,160 --> 01:53:01,037
important issue for the global south.

1627
01:53:01,120 --> 01:53:04,557
We do not believe that just by adapting

1628
01:53:04,640 --> 01:53:06,397
our infrastructure,

1629
01:53:06,480 --> 01:53:11,037
by making it safer, we will also get rid

1630
01:53:11,120 --> 01:53:13,517
of of the lateral and collateral

1631
01:53:13,600 --> 01:53:17,437
problems. And that's when we come to the

1632
01:53:17,520 --> 01:53:20,557
point today we could lose millions of

1633
01:53:20,640 --> 01:53:24,397
jobs. That is a possibility that is

1634
01:53:24,480 --> 01:53:26,317
being entertained even in developed

1635
01:53:26,400 --> 01:53:29,437
countries. We have seen how AI start

1636
01:53:29,520 --> 01:53:32,077
writing the code

1637
01:53:32,160 --> 01:53:35,757
and that for us is perhaps the only

1638
01:53:35,840 --> 01:53:38,237
place that we should start the

1639
01:53:38,320 --> 01:53:39,837
regulation

1640
01:53:39,920 --> 01:53:43,997
because no one discusses no one would

1641
01:53:44,080 --> 01:53:46,557
say that AI is not beneficial for

1642
01:53:46,640 --> 01:53:49,437
research, health development,

1643
01:53:49,520 --> 01:53:52,077
technology, communications,

1644
01:53:52,160 --> 01:53:55,757
uh cure diseases etc. Well, the benefits

1645
01:53:55,840 --> 01:53:58,637
are wide now and we are not trying to

1646
01:53:58,720 --> 01:54:01,277
stop it. But it seems that we are

1647
01:54:01,360 --> 01:54:04,957
reaching a new step on AI and the US is

1648
01:54:05,040 --> 01:54:07,997
right. This is super intelligence for

1649
01:54:08,080 --> 01:54:11,517
one particular reason only. We are about

1650
01:54:11,600 --> 01:54:15,997
to take the step that a
greater and potentially

1651
01:54:16,080 --> 01:54:18,317
more complex technology

1652
01:54:18,400 --> 01:54:20,957
uh quantum technology will boost

1653
01:54:21,040 --> 01:54:24,157
artificial intelligence in the design in

1654
01:54:24,240 --> 01:54:26,797
the model in the modeling and the

1655
01:54:26,880 --> 01:54:31,277
development. So the risks
that we are facing today I

1656
01:54:31,360 --> 01:54:33,277
want to call the attention of the the

1657
01:54:33,360 --> 01:54:36,477
council are pre-quantum risks. So that

1658
01:54:36,560 --> 01:54:39,117
is before we reach that that state but

1659
01:54:39,200 --> 01:54:41,117
we are very very close to reach that

1660
01:54:41,200 --> 01:54:44,957
state. But the concerning side for us

1661
01:54:45,040 --> 01:54:48,237
and perhaps this is what we will offer

1662
01:54:48,320 --> 01:54:50,317
as a different approach rather than

1663
01:54:50,400 --> 01:54:54,637
thinking of an international oversight.

1664
01:54:54,720 --> 01:54:56,637
The Security Council is concerned

1665
01:54:56,720 --> 01:54:59,517
because this technology has the

1666
01:54:59,600 --> 01:55:03,197
potential to disrupt and create not just

1667
01:55:03,280 --> 01:55:05,517
more inequality across the planet

1668
01:55:05,600 --> 01:55:09,597
including violence but also

1669
01:55:09,680 --> 01:55:12,717
development gaps that will lead us to

1670
01:55:12,800 --> 01:55:16,557
other types of conflicts including

1671
01:55:16,640 --> 01:55:18,637
global migration

1672
01:55:18,720 --> 01:55:22,957
etc. So we propose to you, we submit to

1673
01:55:23,040 --> 01:55:25,677
you uh an important consideration in

1674
01:55:25,760 --> 01:55:27,997
which perhaps we can start just for that

1675
01:55:28,080 --> 01:55:31,517
domain because this is it. This is the

1676
01:55:31,600 --> 01:55:33,997
moment.

1677
01:55:34,080 --> 01:55:37,037
Autonomous design and autonomous

1678
01:55:37,120 --> 01:55:39,517
development. This is what is really

1679
01:55:39,600 --> 01:55:42,557
really at stake here.

1680
01:55:42,640 --> 01:55:46,557
Should we allow that machines,

1681
01:55:46,640 --> 01:55:49,357
AI agents take the next step that are

1682
01:55:49,440 --> 01:55:51,357
showing by the experiments that are

1683
01:55:51,440 --> 01:55:53,997
showing by the cases they present to us

1684
01:55:54,080 --> 01:55:56,397
and that we that had been published

1685
01:55:56,480 --> 01:55:59,037
actually not just today there are many

1686
01:55:59,120 --> 01:56:02,077
article journals these are everywhere

1687
01:56:02,160 --> 01:56:04,637
that tell us how the the machine finds

1688
01:56:04,720 --> 01:56:06,797
another way a different way different

1689
01:56:06,880 --> 01:56:09,837
than humans when we lose control one of

1690
01:56:09,920 --> 01:56:12,957
the major risk of risk management here

1691
01:56:13,040 --> 01:56:17,037
which is once we do no longer understand

1692
01:56:17,120 --> 01:56:21,037
what the machine is doing because if we

1693
01:56:21,120 --> 01:56:24,157
go to super intelligence and we do not

1694
01:56:24,240 --> 01:56:27,437
possess as humans such a capacity we are

1695
01:56:27,520 --> 01:56:29,277
assuming that we are controlling

1696
01:56:29,360 --> 01:56:33,917
something that we cannot uh understand

1697
01:56:34,000 --> 01:56:37,757
and one of the main elements for this

1698
01:56:37,840 --> 01:56:40,237
technology and I listen to the all

1699
01:56:40,320 --> 01:56:42,797
interventions Today

1700
01:56:42,880 --> 01:56:45,277
uh we want to make sure that this remain

1701
01:56:45,360 --> 01:56:49,037
human. Well, you see that's the problem.

1702
01:56:49,120 --> 01:56:52,397
The human brain needs to understand

1703
01:56:52,480 --> 01:56:54,877
the information that has in front of it

1704
01:56:54,960 --> 01:56:58,077
to make a decision. But it seems to us

1705
01:56:58,160 --> 01:57:02,477
that we are soon going to leave that

1706
01:57:02,560 --> 01:57:04,717
stage and we are going to leave that

1707
01:57:04,800 --> 01:57:07,437
stage because the way the technology is

1708
01:57:07,520 --> 01:57:10,317
being developed and this is not just a

1709
01:57:10,400 --> 01:57:12,717
matter of ethical limits. We understand

1710
01:57:12,800 --> 01:57:16,477
that that will be a contradictory for

1711
01:57:16,560 --> 01:57:19,597
many and it will be also a problem in

1712
01:57:19,680 --> 01:57:22,637
terms of legislation and regulations.

1713
01:57:22,720 --> 01:57:25,357
Some might even see that as a rate tape

1714
01:57:25,440 --> 01:57:30,957
etc. No, this is another step because

1715
01:57:31,040 --> 01:57:33,197
if the machine is the one that starts

1716
01:57:33,280 --> 01:57:36,477
leading the design, in other words, the

1717
01:57:36,560 --> 01:57:40,157
methods, the paths, the outcomes,

1718
01:57:40,240 --> 01:57:43,437
if this is not just and soon will not be

1719
01:57:43,520 --> 01:57:45,677
a problem of

1720
01:57:45,760 --> 01:57:50,397
the input and the unreachable goal. In

1721
01:57:50,480 --> 01:57:52,317
other words, it seems now that the

1722
01:57:52,400 --> 01:57:55,117
machine is trying to just get the goal

1723
01:57:55,200 --> 01:57:58,797
done, get things done. And to do that is

1724
01:57:58,880 --> 01:58:02,477
finding paths that humans usually do not

1725
01:58:02,560 --> 01:58:05,997
take one and did not instruct to take

1726
01:58:06,080 --> 01:58:09,757
yet it find a way and it cheats and it

1727
01:58:09,840 --> 01:58:11,997
conceals the outcome

1728
01:58:12,080 --> 01:58:14,957
and that is a concern of course. One is

1729
01:58:15,040 --> 01:58:17,517
that is misalignment. The second one now

1730
01:58:17,600 --> 01:58:20,637
however is that

1731
01:58:20,720 --> 01:58:24,637
soon and Mr. Sam Oldman really really

1732
01:58:24,720 --> 01:58:27,917
pointed out to this risk is that the

1733
01:58:28,000 --> 01:58:30,397
autonomous development of artificial

1734
01:58:30,480 --> 01:58:33,357
intelligence what really means is this

1735
01:58:33,440 --> 01:58:35,917
that machines write the code that

1736
01:58:36,000 --> 01:58:38,397
machines find the paths that machines

1737
01:58:38,480 --> 01:58:41,357
establish solutions and again this is

1738
01:58:41,440 --> 01:58:43,677
pre-quantum era

1739
01:58:43,760 --> 01:58:47,997
so it will be naive at least to say to

1740
01:58:48,080 --> 01:58:50,637
think that we will be able to even

1741
01:58:50,720 --> 01:58:53,757
regulate something that soon we will not

1742
01:58:53,840 --> 01:58:55,277
understand.

1743
01:58:55,360 --> 01:58:58,237
And this is why perhaps this is the

1744
01:58:58,320 --> 01:59:01,437
moment in history that all nations has

1745
01:59:01,520 --> 01:59:04,077
to come together at least with one

1746
01:59:04,160 --> 01:59:08,157
common goal. Let's keep it human.

1747
01:59:08,240 --> 01:59:10,637
So we should agree that we want

1748
01:59:10,720 --> 01:59:13,597
development. We might even agree that we

1749
01:59:13,680 --> 01:59:17,917
don't want oversight.
We should understand the the concerns in

1750
01:59:18,000 --> 01:59:19,837
terms of sovereignity especially of the

1751
01:59:19,920 --> 01:59:22,477
nations that have a greater scientific

1752
01:59:22,560 --> 01:59:25,037
development. It is not their go their

1753
01:59:25,120 --> 01:59:28,397
fault to be fully innovated capacities

1754
01:59:28,480 --> 01:59:30,477
etc. But at least for the rest of

1755
01:59:30,560 --> 01:59:33,277
humanity it should be important that we

1756
01:59:33,360 --> 01:59:37,197
decide how far we go. This is again is

1757
01:59:37,280 --> 01:59:41,357
is not an just an ethical country.
We are not talking even about yet about

1758
01:59:41,440 --> 01:59:44,157
human rights. some things some some

1759
01:59:44,240 --> 01:59:46,477
countries think that we should in terms

1760
01:59:46,560 --> 01:59:48,957
of private etc. No no no no this is the

1761
01:59:49,040 --> 01:59:52,157
next step are we going to allow machines

1762
01:59:52,240 --> 01:59:54,877
to build machines to to design their own

1763
01:59:54,960 --> 01:59:58,237
AI and if we are there already and and

1764
01:59:58,320 --> 02:00:00,957
and that's perhaps the most important

1765
02:00:01,040 --> 02:00:03,517
element is for the countries to decide

1766
02:00:03,600 --> 02:00:08,237
at least that because in practice

1767
02:00:08,320 --> 02:00:12,120
uh Mr. President

1768
02:00:15,200 --> 02:00:17,517
>> for many other countries

1769
02:00:17,600 --> 02:00:20,637
particularly Latin American countries

1770
02:00:20,720 --> 02:00:23,037
in practice

1771
02:00:23,120 --> 02:00:27,997
this means a risk for the national
production of many countries countries

1772
02:00:28,080 --> 02:00:32,157
that that aren't producing artificial
intelligence and those that have been

1773
02:00:32,240 --> 02:00:34,877
invited to buy artificial intelligence

1774
02:00:34,960 --> 02:00:39,037
and invest in it and probably Doing that

1775
02:00:39,120 --> 02:00:43,837
would be the most logical path for them
to modernize the infrastructure of each

1776
02:00:43,920 --> 02:00:45,517
nation and make it as safe and secure as

1777
02:00:45,600 --> 02:00:48,477
possible. But ultimately

1778
02:00:48,560 --> 02:00:51,357
the question remains the same. If we

1779
02:00:51,440 --> 02:00:55,517
have um um regulations even down to the

1780
02:00:55,600 --> 02:00:59,517
production of a tire or a toothbrush, if

1781
02:00:59,600 --> 02:01:03,917
we regulate the production of cars,

1782
02:01:04,000 --> 02:01:07,757
if we um do that with nuclear weapons,

1783
02:01:07,840 --> 02:01:10,797
the question remains the same

1784
02:01:10,880 --> 02:01:14,077
in scientific progress including among

1785
02:01:14,160 --> 02:01:18,157
the most important things the Einstein's

1786
02:01:18,240 --> 02:01:20,717
uh the movement from from uh towards

1787
02:01:20,800 --> 02:01:22,877
Einstein's special theory of relativity.

1788
02:01:22,960 --> 02:01:26,797
We not only verified that time moves
differently in different spaces, but we

1789
02:01:26,880 --> 02:01:29,437
also came to understand the uh

1790
02:01:29,520 --> 02:01:33,837
scientific that scientific precision
is based on perception. And we changed the

1791
02:01:33,920 --> 02:01:35,757
formula from constant velocity that did

1792
02:01:35,840 --> 02:01:40,877
not include gravity to another one
that did include gravity. And we reached a

1793
02:01:40,960 --> 02:01:43,997
scientific understanding perhaps
the most important one of the 20th century

1794
02:01:44,080 --> 02:01:47,357
without which we would
not have GPS for example.

1795
02:01:47,440 --> 02:01:51,677
But there was one characteristic common
to all of those cases, including the

1796
02:01:51,760 --> 02:01:55,597
production of nuclear weapons. And
that is that human beings were always in

1797
02:01:55,680 --> 02:02:00,717
control. Human beings
understood the risks. They

1798
02:02:00,800 --> 02:02:04,397
knew what they were doing and they knew
what they were producing and that's why

1799
02:02:04,480 --> 02:02:06,717
they regulated it.

1800
02:02:06,800 --> 02:02:09,277
The concern

1801
02:02:09,360 --> 02:02:12,717
for countries such as Colombia

1802
02:02:12,800 --> 02:02:16,237
isn't scientific progress. Colombia

1803
02:02:16,320 --> 02:02:18,237
invites investment in artificial

1804
02:02:18,320 --> 02:02:19,997
intelligence

1805
02:02:20,080 --> 02:02:23,757
and wants data centers in its country.

1806
02:02:23,840 --> 02:02:27,677
We are a strategic partner of various

1807
02:02:27,760 --> 02:02:32,077
countries uh including the United
States in particular and naturally we are

1808
02:02:32,160 --> 02:02:36,317
interested in being part of this not
just on the consumption side but also

1809
02:02:36,400 --> 02:02:40,477
the production side of artificial
intelligence and certainly

1810
02:02:40,560 --> 02:02:43,197
um through the provision of rare earths

1811
02:02:43,280 --> 02:02:47,917
as well. But there is something that
goes beyond bilateral commercial or or

1812
02:02:48,000 --> 02:02:52,237
national interests and that
is the interests of humanity

1813
02:02:52,320 --> 02:02:54,477
beyond our own time beyond our own

1814
02:02:54,560 --> 02:02:57,757
generation generation and the time that

1815
02:02:57,840 --> 02:02:59,517
is allotted to each and every one of us.

1816
02:02:59,600 --> 02:03:01,917
We are here now and we need to ask

1817
02:03:02,000 --> 02:03:07,320
ourselves this differentiated question

1818
02:03:07,920 --> 02:03:10,797
right now.

1819
02:03:10,880 --> 02:03:14,557
We are setting this precedent which in

1820
02:03:14,640 --> 02:03:17,517
the view of a country such as Colombia

1821
02:03:17,600 --> 02:03:19,757
seems to be delicate. We regulated

1822
02:03:19,840 --> 02:03:21,917
nuclear weapons. We regulate the

1823
02:03:22,000 --> 02:03:24,957
manufacturing of tires. We even regulate

1824
02:03:25,040 --> 02:03:29,837
the provision of homes. Anything that
could jeopardize the lives of the people

1825
02:03:29,920 --> 02:03:34,637
that use those things is regulated. So
why then would we not regulate something

1826
02:03:34,720 --> 02:03:36,637
which soon we are not even going to

1827
02:03:36,720 --> 02:03:41,597
understand? That makes no sense.

1828
02:03:41,680 --> 02:03:44,877
Super intelligence is not just the

1829
02:03:44,960 --> 02:03:46,477
ability to achieve that through

1830
02:03:46,560 --> 02:03:49,517
scientific transformation, but also the

1831
02:03:49,600 --> 02:03:52,957
ability to have that to know when we are

1832
02:03:53,040 --> 02:03:57,517
no longer on the same level.
Thank you very much, President.

1833
02:03:57,600 --> 02:03:59,837
>> I thank the permanent representative of

1834
02:03:59,920 --> 02:04:02,477
Colombia for the statement. I will now

1835
02:04:02,560 --> 02:04:07,357
give the floor to her excellency, Miss
TZ Wagner, Minister of Foreign Affairs,

1836
02:04:07,440 --> 02:04:11,357
International Cooperation, and Franco
of the Democratic Republic of the Congo.

1837
02:04:11,440 --> 02:04:14,397
Thank you very much, Mr. President,
excellencies, ladies and gentlemen, the

1838
02:04:14,480 --> 02:04:17,677
Democratic Republic of the Congo,
thanks Fran, thanks France as well as his

1839
02:04:17,760 --> 02:04:22,077
excellency, Mr. Jean Noel Bahu, Minister
for Europe and Foreign Affairs for

1840
02:04:22,160 --> 02:04:23,357
having convened and shared this

1841
02:04:23,440 --> 02:04:26,077
important meeting. I also thank the

1842
02:04:26,160 --> 02:04:29,437
briefers for their contributions to our

1843
02:04:29,520 --> 02:04:31,597
collective reflection. Our debate takes

1844
02:04:31,680 --> 02:04:34,237
place at a pivotal moment. Artificial

1845
02:04:34,320 --> 02:04:38,637
intelligence is no longer merely
a matter of scientific progress or

1846
02:04:38,720 --> 02:04:42,957
economic competitiveness. Because of the
speed of its developments, the scope of

1847
02:04:43,040 --> 02:04:46,957
its applications, and its growing
integration into critical infrastructure

1848
02:04:47,040 --> 02:04:49,837
and decision-making systems, it now

1849
02:04:49,920 --> 02:04:52,957
raises questions that directly affect

1850
02:04:53,040 --> 02:04:55,757
international peace and security.

1851
02:04:55,840 --> 02:04:59,357
And this reality imposes
a responsibility upon us. the

1852
02:04:59,440 --> 02:05:01,277
responsibility to anticipate risks

1853
02:05:01,360 --> 02:05:06,477
without hindering innovation to preserve
the benefits of technological progress

1854
02:05:06,560 --> 02:05:10,157
while preventing it from becoming
an additional factor of instability,

1855
02:05:10,240 --> 02:05:13,117
vulnerability and conflict. The United

1856
02:05:13,200 --> 02:05:17,197
Nations have begun laying
the foundations for this reflection. the

1857
02:05:17,280 --> 02:05:21,597
work undertaken by the general
assembly on safe secure and trustworth

1858
02:05:21,680 --> 02:05:23,597
trustworthy artificial intelligence the

1859
02:05:23,680 --> 02:05:26,157
establishment of the independent

1860
02:05:26,240 --> 02:05:30,957
international scientific panel on AI and
the global dialogue on AI governance as

1861
02:05:31,040 --> 02:05:34,637
well as the security council's work
on the implications of scientific and

1862
02:05:34,720 --> 02:05:38,717
technological advances for international
peace and security. All of these

1863
02:05:38,800 --> 02:05:43,117
constitute important milestones.
It is now our responsibility to go further

1864
02:05:43,200 --> 02:05:47,757
from anticipation to prevention and from
prevention to international cooperation

1865
02:05:47,840 --> 02:05:51,277
that is commeasured with the
transformations underway. Presidents,

1866
02:05:51,360 --> 02:05:53,997
the risks we face are numerous.

1867
02:05:54,080 --> 02:05:58,157
Increasingly capable and autonomous
systems can create new vulnerabilities

1868
02:05:58,240 --> 02:06:00,557
for critical infrastructure and

1869
02:06:00,640 --> 02:06:03,357
decision-making processes. Their use in

1870
02:06:03,440 --> 02:06:06,077
military and security contexts raises as

1871
02:06:06,160 --> 02:06:09,357
fundamental questions of accountability,

1872
02:06:09,440 --> 02:06:12,957
predictability, and human control. We

1873
02:06:13,040 --> 02:06:16,077
must also remain vigilant regarding the

1874
02:06:16,160 --> 02:06:18,637
unforeseen behavior of certain systems,

1875
02:06:18,720 --> 02:06:21,117
the possibility of circumventing

1876
02:06:21,200 --> 02:06:25,677
security safeguards, and the growing
difficulties states may encounter in

1877
02:06:25,760 --> 02:06:29,757
understanding, assessing,
and regulating technologies

1878
02:06:29,840 --> 02:06:32,077
whose capabilities are evolving at an

1879
02:06:32,160 --> 02:06:35,197
unprecedented pace. These issues take on

1880
02:06:35,280 --> 02:06:39,917
a particularly tangible dimension
in conflict situations. In the Democratic

1881
02:06:40,000 --> 02:06:43,277
Republic of the Congo, as in other
theaters of crisis, surveillance

1882
02:06:43,360 --> 02:06:47,197
technologies and drones already play
a growing role in the operational

1883
02:06:47,280 --> 02:06:51,357
environment.
These technologies can contribute

1884
02:06:51,440 --> 02:06:55,517
to reconnaissance, observation,
and ceasefire monitoring. They can also be

1885
02:06:55,600 --> 02:06:58,077
diverted for military reconnaissance,

1886
02:06:58,160 --> 02:07:00,237
coordination, disinformation, or

1887
02:07:00,320 --> 02:07:03,117
propaganda purposes. This ambivalis

1888
02:07:03,200 --> 02:07:05,197
reminds us of an essential truth. No

1889
02:07:05,280 --> 02:07:07,837
technology is in itself an instrument of

1890
02:07:07,920 --> 02:07:12,717
peace or conflict. It is the framework
within which it is designed, deployed

1891
02:07:12,800 --> 02:07:14,717
and governed that largely determines its

1892
02:07:14,800 --> 02:07:17,437
impact. That is why responsibility must

1893
02:07:17,520 --> 02:07:19,757
remain at the heart of our reflection.

1894
02:07:19,840 --> 02:07:22,637
States have a responsibility to

1895
02:07:22,720 --> 02:07:27,357
establish appropriate legal frameworks,
ensure effective oversight, and prevent

1896
02:07:27,440 --> 02:07:29,117
risks that could affect international

1897
02:07:29,200 --> 02:07:31,517
peace and security.

1898
02:07:31,600 --> 02:07:36,717
Technology companies must also
fully assume their share of

1899
02:07:36,800 --> 02:07:41,277
responsibility, particularly through
risk assessments prior to deployment,

1900
02:07:41,360 --> 02:07:45,357
incident reporting,
traceability, transparency,

1901
02:07:45,440 --> 02:07:49,517
and maintaining
appropriate human oversight.

1902
02:07:49,600 --> 02:07:52,477
The scientific community and civil
society must likewise be able to

1903
02:07:52,560 --> 02:07:55,757
contribute their independent expertise
and participate in the critical

1904
02:07:55,840 --> 02:07:58,797
assessment of risks. But national

1905
02:07:58,880 --> 02:08:03,437
responses, as necessary as they may
be, will not suffice. Technological

1906
02:08:03,520 --> 02:08:07,677
interdependence calls for shared
responsibility. It requires

1907
02:08:07,760 --> 02:08:11,437
closer international cooperation
based in particular on the exchange of

1908
02:08:11,520 --> 02:08:14,477
information concerning serious incidents

1909
02:08:14,560 --> 02:08:18,637
and identified vulnerabilities,
cooperation and evaluation and safety

1910
02:08:18,720 --> 02:08:23,357
testing, as well as the establishment of
communication channels among governness,

1911
02:08:23,440 --> 02:08:27,277
government scientists, and technology
stakeholders. President, for the

1912
02:08:27,360 --> 02:08:28,797
Democratic Republic of the Congo, this

1913
02:08:28,880 --> 02:08:31,517
cooperation must also meet a fundamental

1914
02:08:31,600 --> 02:08:34,317
requirement, that of inclusion.

1915
02:08:34,400 --> 02:08:39,037
Developing countries cannot be
reduced to the role of mere users of

1916
02:08:39,120 --> 02:08:43,997
technologies designed elsewhere,
nor should be the mere recipients of

1917
02:08:44,080 --> 02:08:45,597
standards and norms that they did not

1918
02:08:45,680 --> 02:08:49,117
adequately participate in shaping.

1919
02:08:49,200 --> 02:08:53,437
They must possess the scientific,
technical and institutional capacities

1920
02:08:53,520 --> 02:08:56,317
that enable them to assess the
technologies to which their populations

1921
02:08:56,400 --> 02:09:00,717
and economies are exposed to participate
in their governance and above all to

1922
02:09:00,800 --> 02:09:03,117
contribute to innovation themselves.

1923
02:09:03,200 --> 02:09:05,757
This requirement of inclusion leads us

1924
02:09:05,840 --> 02:09:07,517
to consider artificial intelligence

1925
02:09:07,600 --> 02:09:10,957
across its entire value chain.

1926
02:09:11,040 --> 02:09:14,317
For despite its seemingly immaterial

1927
02:09:14,400 --> 02:09:18,557
nature, the AI revolution rests
upon profoundly material realities,

1928
02:09:18,640 --> 02:09:21,197
infrastructure, energy, industrial

1929
02:09:21,280 --> 02:09:26,397
capabilities, and strategic natural
resources. This dimension is of

1930
02:09:26,480 --> 02:09:30,717
particular importance to the Democratic
Republic of the Congo. During its

1931
02:09:30,800 --> 02:09:32,557
presidency of the security council last

1932
02:09:32,640 --> 02:09:35,357
July, my country invited the council to

1933
02:09:35,440 --> 02:09:38,797
reflect on how responsible, transparent

1934
02:09:38,880 --> 02:09:42,797
and equitable governance of natural
resources can contribute to conflict

1935
02:09:42,880 --> 02:09:44,397
prevention,

1936
02:09:44,480 --> 02:09:46,957
supply chain security

1937
02:09:47,040 --> 02:09:51,677
and the emergence of shared prosperity.

1938
02:09:51,760 --> 02:09:54,557
The AI revolution gives this reflection

1939
02:09:54,640 --> 02:09:56,557
additional relevance. We must ensure

1940
02:09:56,640 --> 02:09:58,717
that technological transformations of

1941
02:09:58,800 --> 02:10:01,597
the 21st century do not reproduce the

1942
02:10:01,680 --> 02:10:04,957
economic imbalances of previous decades.

1943
02:10:05,040 --> 02:10:09,117
Countries supplying
essential resources to the

1944
02:10:09,200 --> 02:10:11,757
global economy while processing

1945
02:10:11,840 --> 02:10:13,837
technology and most of the added value

1946
02:10:13,920 --> 02:10:16,397
remain concentrated elsewhere. The

1947
02:10:16,480 --> 02:10:20,477
security of supply chains cannot be

1948
02:10:20,560 --> 02:10:22,557
separated from the prosperity of the

1949
02:10:22,640 --> 02:10:25,917
populations at their source. That is why

1950
02:10:26,000 --> 02:10:28,637
AI governance must also be approached

1951
02:10:28,720 --> 02:10:31,837
from a logic of shared responsibility.

1952
02:10:31,920 --> 02:10:36,237
Responsibility in the design of systems,
responsibility in their use and

1953
02:10:36,320 --> 02:10:41,357
responsibility within the value chains
that make their development possible and

1954
02:10:41,440 --> 02:10:45,357
also responsibility in the distribution
of the benefits of technological

1955
02:10:45,440 --> 02:10:49,117
progress.
For the Democratic Republic of the

1956
02:10:49,200 --> 02:10:53,917
Congo, the objective
is clear. The resources

1957
02:10:54,000 --> 02:10:57,117
that help make the digital revolution
possible must also contribute to the

1958
02:10:57,200 --> 02:11:00,557
development, economic transformation,
industrialization, and strengthening of

1959
02:11:00,640 --> 02:11:04,877
the technological capabilities of those
countries that produce them. This is not

1960
02:11:04,960 --> 02:11:07,037
merely a matter of development. It is

1961
02:11:07,120 --> 02:11:09,757
also a matter of peace. When value

1962
02:11:09,840 --> 02:11:11,917
chains are opaque and benefits are

1963
02:11:12,000 --> 02:11:17,357
distributed unequally and when producing
communities remain excluded from the

1964
02:11:17,440 --> 02:11:21,357
prosperity generated by their own
resources, economic vulnerabilities can

1965
02:11:21,440 --> 02:11:25,757
fuel tensions and conflict economies.

1966
02:11:25,840 --> 02:11:27,837
Conversely, when resources are governed

1967
02:11:27,920 --> 02:11:30,477
responsibly, when local processing

1968
02:11:30,560 --> 02:11:33,117
creates jobs and opportunities,

1969
02:11:33,200 --> 02:11:36,397
and when populations genuinely share in

1970
02:11:36,480 --> 02:11:40,637
the benefits of growth, supply chains
can not only become more secure and

1971
02:11:40,720 --> 02:11:44,877
resilient, but also chains
of shared prosperity.

1972
02:11:44,960 --> 02:11:46,957
Mr. President, it is precisely here that

1973
02:11:47,040 --> 02:11:50,157
the Security Council can bring added

1974
02:11:50,240 --> 02:11:54,557
value. It is not the council's role
to replace the appropriate bodies

1975
02:11:54,640 --> 02:11:56,397
responsible for developing the full

1976
02:11:56,480 --> 02:11:59,037
range of technical rules relating to AI,

1977
02:11:59,120 --> 02:12:02,077
but it is fully within the council's

1978
02:12:02,160 --> 02:12:06,317
remmit to anticipate the risks that may
affect international peace and security,

1979
02:12:06,400 --> 02:12:08,317
to consider how new technologies

1980
02:12:08,400 --> 02:12:11,677
transform conflict dynamics, to protect

1981
02:12:11,760 --> 02:12:15,757
civilians, to encourage confidence
building measures, and to ensure that

1982
02:12:15,840 --> 02:12:19,757
technological transformations
do not create new vulnerabilities

1983
02:12:19,840 --> 02:12:22,477
that could fuel and stability.

1984
02:12:22,560 --> 02:12:26,797
Our objective must therefore not

1985
02:12:26,880 --> 02:12:30,157
be to choose between innovation
and security any more than it should be to

1986
02:12:30,240 --> 02:12:34,157
choose between natural resources
and development. Our responsibility is to

1987
02:12:34,240 --> 02:12:37,117
govern progress so that it serves

1988
02:12:37,200 --> 02:12:41,197
humanity. As July under
the presidency of the DRC,

1989
02:12:41,280 --> 02:12:44,717
our council reflected on how to transform

1990
02:12:44,800 --> 02:12:48,957
natural resources which are too often
associated with conflict economies into

1991
02:12:49,040 --> 02:12:50,637
instruments of peace, cooperation, and

1992
02:12:50,720 --> 02:12:53,037
shared prosperity.

1993
02:12:53,120 --> 02:12:57,917
Today, the AI revolution in another form

1994
02:12:58,000 --> 02:13:00,797
puts us in front of a similar question.

1995
02:13:00,880 --> 02:13:04,317
How can we ensure that technological
progress reduces divisions rather than

1996
02:13:04,400 --> 02:13:05,997
deepening them? How can we ensure that

1997
02:13:06,080 --> 02:13:08,557
it strengthens states capacities rather

1998
02:13:08,640 --> 02:13:11,677
than creating new dependencies?

1999
02:13:11,760 --> 02:13:15,597
And finally, how can we ensure
that value chains, the value chains

2000
02:13:15,680 --> 02:13:18,877
supporting this revolution contribute
to the peace and prosperity of the

2001
02:13:18,960 --> 02:13:23,357
societies that make them possible
for the Democratic Republic of the Congo?

2002
02:13:23,440 --> 02:13:25,517
The answer must rest on three

2003
02:13:25,600 --> 02:13:27,277
principles: shared responsibility,

2004
02:13:27,360 --> 02:13:31,597
inclusion, and shared prosperity.

2005
02:13:31,680 --> 02:13:35,197
As the capabilities of artificial

2006
02:13:35,280 --> 02:13:39,117
intelligence advance, our collective
capacity to govern them must advance

2007
02:13:39,200 --> 02:13:43,357
alongside them. Because whether
the subject is natural resources or

2008
02:13:43,440 --> 02:13:46,717
artificial intelligence, the fundamental
question remains the same. It is not

2009
02:13:46,800 --> 02:13:51,037
only about what our resources
and technologies enable us to do, but also

2010
02:13:51,120 --> 02:13:53,117
about what we collectively decide to do

2011
02:13:53,200 --> 02:13:55,917
with them and that is where our common

2012
02:13:56,000 --> 02:13:58,237
responsibility begins. Thank you. I

2013
02:13:58,320 --> 02:14:03,357
thank her excellency Miss Kwbo
Wagner for the statement and I

2014
02:14:03,440 --> 02:14:08,680
now give the floor to the permanent
representative of China.

2015
02:14:11,360 --> 02:14:16,720
>> Thank you, President. I thank
France for convening this meeting and