Altman and Amodei Call for International Coordination on AI Safety at UN Security Council

A debate session is not the same as a Council decision

Sam Altman, OpenAI’s chief executive, and Dario Amodei, Anthropic’s chief executive, took part in a United Nations Security Council session devoted to artificial intelligence and its implications for international security. The sources provided place the meeting in the context of the UN General Assembly and say France organized it during its rotating presidency of the Council. They also mention the participation of Yoshua Bengio and Clément Delangue.

The issue that brought the executives before the Council was how to respond to the risks posed by more capable AI systems, including the possibility that rapid development could make human control more difficult. Available reports describe remarks and appeals for cooperation. Those should not be confused with an adopted rule, a binding commitment by states, or a formal Council decision.

That distinction matters. A briefing session can give an issue international visibility and allow Council members to hear from experts and company leaders without the body adopting any measures. The material provided does not show that the meeting produced a resolution, a negotiation timetable, or an agreed international oversight mechanism.

Sam Altman’s case for coordination and human control

In his remarks to the Council, Altman discussed AI safety, the need to preserve human control, and international cooperation. OpenAI’s page presents his words as remarks delivered to the Council. A subsequent EFE report summarized his appeal as a call for cooperation to guide technology companies through complementary standards at both national and global levels.

The central idea conveyed by these sources is that the challenge is not confined to one company or one jurisdiction. If AI systems’ capabilities advance rapidly, the rules and safety practices of a single country may not be enough to address risks that cross borders. In this framing, cooperation would help align expectations and prevent safeguards from depending solely on isolated decisions.

That does not amount to a detailed treaty proposal. The information provided does not specify draft rules, who would administer a compliance system, or what penalties might apply. Nor does it allow us to attribute to Altman—without consulting his full remarks—a specific plan for licensing, inspections, or mandatory limits. It is important to distinguish the general principle of cooperation, which the sources do document, from instruments they do not describe.

Dario Amodei: common standards for cross-border risks

Available news reports attribute a related argument to Amodei: AI risks cannot be managed fully by a single company or a single country. They also describe his call for common global standards to assess both the misuse of these systems and the risk of losing control over them. Bloomberg Línea’s coverage summarizes Altman and Amodei as advocating cooperation on standards in light of the possibility that AI systems could improve themselves and exceed humans’ ability to control them.

The overlap supported by these sources is broad, not necessarily complete: both executives advocated international cooperation and attention to risks associated with more capable systems. The material provided does not offer a full, point-by-point comparison of their remarks. It is therefore not possible to conclude that they presented a joint plan, disagreed over a specific measure, or reached agreement with each other on a particular policy.

It is also important to distinguish a warning about possible risks from a claim that those risks have already materialized. The summaries provided discuss scenarios involving loss of control and the possibility of systems with increasing capabilities. They do not establish that an AI system has surpassed human control or that there is scientific consensus on when that might happen. The session put these risks up for discussion; it did not resolve them.

What was proposed—and what is not documented as agreed

Taken together, the proposals described in the coverage can be summarized in three ideas: strengthen cooperation among countries, develop safety standards that can be coordinated internationally, and address both misuse and risks related to loss of control. These are appeals to guide action and debate. They are not, by themselves, rules adopted by governments.

The distinction becomes clearer if we separate three levels. First are the executives’ statements: OpenAI’s source publishes Altman’s remarks, while news reports summarize Amodei’s position. Second are the policy proposals described in the press: cooperation and standards. Third are institutional outcomes: a resolution, a mandate, a body, or a government commitment. The supplied material allows us to report the first two levels with caution, but does not document an outcome at the third.

A transcript of the session is among the available sources, although its own description warns that it was generated automatically and is not an official document or record. It can help confirm who spoke and review what was said, but it is wise not to present wording as a verbatim quotation unless it has been checked against the audio or video. This article therefore summarizes the positions without attributing exact phrases that have not been verified in the sources described.

There is also not enough evidence to interpret the idea of slowing or decelerating development as a precise policy agreed at the session. Pre-meeting coverage refers to debate about a possible slowdown for safety reasons, but that does not prove that Altman and Amodei called for a moratorium with specific conditions before the Council. Supporting that interpretation would require verified wording and details about its scope, duration, and responsible parties.

The UN’s role—and the limits of what is known

The Security Council is an intergovernmental forum whose mandate is connected to international peace and security. A session can help examine whether certain AI risks have implications for that area and bring state representatives, experts, and companies into conversation. However, the presence of technology executives does not turn their proposals into UN policy: decisions belong to the Council’s members and depend on institutional procedures.

The session is also part of a broader discussion about international coordination and regulation. The UN can provide a forum for sharing assessments and discussing possible responses, while governments retain the ability to adopt national measures and negotiate agreements. Companies, in turn, can propose practices or standards, but they do not replace states or multilateral bodies.

The conclusion supported by the sources is therefore limited: Altman and Amodei presented arguments to the Council in favor of cooperation and safety standards, and the discussion addressed human control and potential risks from advanced systems. The supplied material does not show that countries agreed on a binding instrument, that the Council ordered AI development to slow, or that permanent international oversight was established.

The main uncertainty concerns the level of detail in the remarks and any institutional follow-up. Resolving it would require checking the full statements and consulting official documentation about any subsequent decision. Until such evidence is available, the most rigorous reading is that there was a political appeal and an international debate—not a new global AI safety regime.

Keep exploring

Corrections and transparency

If you spot incorrect or outdated information, send us a correction with the page and source we should review.