On September 23, 2026, three executives leading major AI companies appeared before the U.N. Security Council to deliver warnings about what their own products could become. The setting was not a product launch stage or a tech conference; it was the body responsible for international peace and security, where AI is making autonomous threats an ever more urgent concern.
The shift in venue matters. Risk language that once circulated mainly among academic safety researchers entered formal international diplomacy that day, carried there by the very people building the systems under scrutiny.
“A Risk to Humanity as a Whole”
The heads of two of the world’s leading AI companies delivered some of the starkest public warnings yet about the dangers of advanced AI.
Dario Amodei, CEO of Anthropic, stated it plainly: “If managed poorly, I even believe that AI could be a risk to humanity as a whole,” according to the Associated Press via PBS NewsHour. Sam Altman, CEO of OpenAI, warned that humanity could “lose control of the future to AI,” per the same reporting.
Altman described the danger as AI advancing faster than people can follow or intervene when something goes wrong. That concern centers on two concepts worth defining clearly.
Frontier AI refers to the most capable, cutting-edge systems currently under development. Loss of control describes a scenario where humans can no longer reliably understand, constrain, or stop what an AI system does.
Altman also raised a second concern: the concentration of power. Available sourcing confirms his broader warning that no single entity should be able to use advanced AI to impose its worldview globally, though the complete statement was not reproduced in verified reporting.
What They Asked Governments to Do
The executives arrived not only with warnings but with specific requests for international action.
Altman called for capability benchmarks: formal measures of how powerful AI systems are and whether company safety practices keep pace as those systems grow stronger, per Reuters. Amodei pushed for shared international standards governing AI testing and the assessment of loss-of-control risks, per reporting by PennLive. He also called for agreements preventing AI from assisting in the development of biological weapons.
Amodei pledged that Anthropic would slow deployment as much as necessary to ensure safety, arguing that more stringent standards would be required as models become more capable, according to France 24. The broader request from the executives was that governments establish common safeguards rather than leave oversight entirely to private companies, per UN News.
Clément Delangue, co-founder of AI platform Hugging Face, also briefed the council. France organized the session while holding the Security Council’s rotating presidency in September.
Autonomous Weapons and Geopolitical Stakes
The session was treated as a matter of international security, not a technology policy forum.
A European diplomat cited by Reuters raised the possibility that AI systems interacting autonomously with one another could trigger armed conflict between states. U.N. Secretary-General António Guterres had warned the previous day about autonomous weapons, and experts and European leaders at the session described that risk as near-term rather than speculative.
China’s presence added diplomatic weight. Chinese AI companies DeepSeek and Moonshot were invited to address the session, reflecting the geopolitical dimension of any conversation about who builds frontier AI and under what rules.
The Contradiction at the Center
The executives warning the Security Council are also the ones building the systems they warned about.
Altman leads OpenAI. Amodei co-founded Anthropic after leaving OpenAI. Both companies are in an active race to develop more capable models and expand infrastructure. Attracting the capital required to compete at the frontier is central to that effort.
Common international standards could raise compliance costs and require more extensive testing before releases. They could also reduce the regulatory fragmentation that currently makes operating across borders complicated; that second effect is analysis, not a reported outcome.
Several claims from background briefing materials, including details about U.S. AI infrastructure initiatives, presidential financial disclosures, and public polling on data centers, were not confirmed by available sourcing and are not reported here as established fact.
What Remains Unresolved
The most consequential questions from September 23 left the Security Council chamber without answers.
Governments have not reached a universally accepted definition of what constitutes a dangerous level of AI capability. There is no widely recognized global body with the authority to evaluate frontier systems across borders. The question of what happens when a system fails a safety evaluation, specifically who acts, how fast, and with what authority, remains entirely open.
Your stake in those answers is not abstract. AI systems already shape search results and automate decisions about credit and hiring; some are also being integrated into national defense infrastructure. Whether the people building them answer to international standards or only to their own internal benchmarks is, at its core, a question about accountability in a domain most governments are still working to understand.
