UN Security Council Debates Global AI Oversight as US Rejects Multilateral Authority
The UN Security Council convened a session on advanced AI risk on September 23-24, bringing frontier-lab chief executives face to face with national governments for one of the most direct airings yet of a question the industry has mostly left to individual jurisdictions: who, if anyone, should set binding rules for systems that already operate across borders. The meeting produced no agreement and no mechanism with teeth, but it did produce something rarer — an on-the-record split between the companies building the most powerful AI systems and the government that hosts most of them.
Sam Altman told the Council that decisions about AI's trajectory "cannot be made by labs in San Francisco alone" and pushed for common capability-assessment benchmarks industry and governments could share. Dario Amodei went further, telling delegates that AI "could be a risk to humanity as a whole" if development is managed poorly, and called for international testing and verification standards. AI researcher Yoshua Bengio proposed licensing requirements for frontier systems deployed in critical sectors, while Hugging Face CEO Clément Delangue pushed for mandatory incident disclosure and global monitoring — and said his own company had turned to a Chinese model for a defensive task because comparable U.S. tools carried more restrictions.
The White House response came from Michael Kratsios, head of the Office of Science and Technology Policy, who rejected the entire premise. "The US totally rejects all efforts by international bodies to assert centralized control & global governance of AI," he wrote, calling concern over existential risk and social equity an "ideological fixation" that obstructs progress. In the room, he put it more plainly: a prosperous future "will not be secured by a global regulator." The exchange left little ambiguity about where the administration's AI-dominance strategy — built on exporting American models and standards rather than negotiating shared ones — stands relative to the companies it is meant to champion.
"National action is essential, but global coordination is indispensable." — António Guterres, UN Secretary-General
The session sits inside machinery the UN already built for exactly this purpose: an Independent International Scientific Panel on AI and a standing Global Dialogue on AI Governance, both established in August 2025, plus a preliminary risk report the panel issued on July 1. Guterres used the Security Council platform to press for cooperation specifically between the US and China to avoid what he called a "race to the bottom," floating proposals including an AI Child Safety Pledge and a Global Fund to help developing countries build AI capacity — soft, voluntary instruments rather than enforceable law.
What makes the split notable is who is on which side. Altman and Amodei have each resisted specific binding rules in domestic fights — over state disclosure mandates, liability standards, licensing — even as they now ask for international ones. That is not necessarily contradiction: for companies selling the same models into dozens of jurisdictions, a single harmonized standard is operationally cheaper than compliance with the fifty-state and multi-continent patchwork now forming around them. Beijing, meanwhile, has embraced the UN process for reasons of its own, using it to build soft power among the Global South while the US disengages — meaning the likeliest near-term outcome isn't a global rulebook but a widening gap between two governance blocs, with the industry's own leadership on record wanting the opposite.