The United Nations Security Council has long served as a venue for deliberations over existential threats—nuclear weapons, climate change, pandemics. The Council is the UN body charged with primary responsibility for international peace and security, which makes its agenda a rough barometer of how governments collectively frame global risk. On September 23, 2026, it added a new item to that list: artificial intelligence that might outgrow human control.

Sam Altman, chief executive of OpenAI, and Dario Amodei, chief executive of Anthropic, appeared before the Security Council during the annual UN General Assembly session, the September gathering in New York that brings heads of state and government together in one place. Both urged world leaders to build cooperative frameworks around AI safety before the technology's trajectory outruns the institutions meant to govern it.

What the executives said

Altman's message centered on shared stakes. He argued that nations hold common interests in managing powerful new technologies, and that those shared interests—not competition—should form the foundation for international coordination.

Amodei went further, warning that AI could pose genuine risks to humanity if the development process lacks adequate safeguards. His appearance at the UN followed a September 12 essay in which he publicly called for a more measured pace of AI capability development and advocated placing independent evaluators inside labs to provide accountability that internal teams cannot deliver themselves.

The session carried significance beyond its speaker list. It was the first UN Security Council briefing focused specifically on frontier AI safety risks—the term generally denotes the most capable AI systems at the current edge of development—a signal that the conversation has migrated from academic conferences and Silicon Valley podcasts into the machinery of formal international diplomacy.

Other prominent voices joined the discussion. Clément Delangue, CEO of Hugging Face, participated alongside Yoshua Bengio, one of the foundational researchers in modern deep learning and a consistent advocate for caution in AI development. Chinese AI developers were also invited to contribute—a notable detail given the geopolitical tensions threading through the entire conversation.

Diplomacy, disagreement, and recursive self-improvement

The specific risk that animated much of the session was recursive self-improvement: the possibility that advanced AI systems could iteratively enhance their own capabilities in ways that exceed meaningful human oversight.

The meeting also sits inside a broader diplomatic moment. U.S.-China dialogue on AI regulation has been quietly expanding, and the inclusion of Chinese developers in the UN session reflected an attempt to make the conversation genuinely global rather than a Western standards-setting exercise dressed up as multilateralism.

That attempt faces a direct political headwind. U.S. President Donald Trump has publicly rejected what he characterized as a globalist scheme to control AI, a framing that positions international coordination as a threat to American technological leadership rather than a mechanism for managing shared risk.

Bengio's presence added a different dimension. He operates outside the commercial AI ecosystem and has no product roadmap to protect. His participation alongside the CEOs of OpenAI and Anthropic signals a rare convergence between academic safety researchers and industry leaders on the need for international mechanisms.

What comes next

The Security Council session did not produce binding agreements or formal resolutions. That distinction matters: under the UN Charter, Security Council resolutions can carry binding force on member states, making formal Council action one of the strongest levers available in international governance.

For OpenAI and Anthropic specifically, the advocacy for global standards cuts both ways. On one hand, industry-aligned companies that help shape international frameworks often end up with standards that reflect their existing practices, which is a competitive advantage. On the other hand, any serious multilateral regime would likely bring the kind of independent evaluation and oversight that Amodei described in his September 12 essay, and that creates operational friction that does not exist today.

Any international AI safety framework that excludes China is incomplete by definition, given the scale of Chinese AI development. But including China means negotiating across fundamentally different regulatory philosophies and political systems.