Altman and Amodei brief the UN Security Council on AI risk today, in its first meeting devoted specifically to AI safety

The UN Security Council holds its first meeting devoted specifically to AI-safety risk today, convened by France during the UN General Assembly's high-level week. OpenAI's Sam Altman is confirmed to attend in person; Anthropic's Dario Amodei is expected to join and press the same argument as his Sept. 12 essay, We Must Pace the Frontier -- the one that, with same-day endorsements from Altman, Elon Musk and Demis Hassabis attached, is now the subject of a federal antitrust lawsuit accusing the four labs of an illegal agreement to slow their own products. China's DeepSeek and Moonshot were also invited to brief the Council; DeepSeek's own founder, Liang Wenfeng, is not expected to show.

This is the Council's seventh meeting engaging AI's connection to international security since its first formal debate in July 2023, but the first one built specifically around the risk of losing control of increasingly capable systems rather than AI's broader effects on conflict, disinformation or cyber operations. France's own concept note names the shift directly, warning that 'autonomous AI systems could undertake actions with serious international security consequences, such as attacks on critical infrastructure,' and pointing specifically at the danger of a system capable of recursive self-improvement outrunning any human's ability to supervise it. The note poses four guiding questions rather than a proposal: what risks rapid AI advances pose to peace and security, how governments and companies can address them, what diplomatic tools exist, and how evaluation and verification can keep pace with development. Yoshua Bengio, co-chair of the UN's Independent International Scientific Panel on AI, and Hugging Face CEO Clement Delangue are also confirmed to brief the 15-member Council -- voices from research and open-source software alongside the two companies actually building the frontier systems the meeting is about.

Amodei's own proposal is more specific than France's questions. It asks that outside evaluators get employee-level facility access at frontier labs -- a step Anthropic says it has already taken unilaterally -- that the arrangement spread through binding international rules rather than company-by-company goodwill, and that labs accept a verified speed limit on how fast their models' own self-improvement capability grows, explicitly modeled on Cold War-era SALT missile treaties between the US and Soviet Union. Two days after the essay published, China's state-run Global Times called it exactly that kind of geopolitics wearing a safety label.

“The article is packed with containment provisions targeting China and is, in essence, a 'Cold War playbook' for the AI sector.” — Global Times editorial, Sept. 14 2026

Beijing's Foreign Ministry followed with a more measured version of the same objection days later: fearmongering and confrontation, it said, 'will only disrupt the process of global AI governance which serves no one's interest.' The sequence is now five events in eleven days, and today's briefing is the first one happening on a stage neither company controls. Anthropic disclosed this week that Claude now automates 26% of its own AI research and development -- a genuine acceleration metric, published the same stretch its CEO is asking the industry, and now a UN body, to agree to slow down. Anthropic's own framing treats the 26% figure as evidence its safety infrastructure can keep pace with its own capability growth, not as a contradiction of the pacing argument -- but it is the argument's obvious pressure point.

What's different about today isn't the argument -- it's the venue. Amodei has been making this case in essays and interviews for weeks; China has been rejecting it in press statements for just as long. A Security Council briefing doesn't bind anyone -- these sessions produce statements and transcripts, not enforceable resolutions, absent a formal vote nobody has scheduled -- but it puts both sides' arguments on a diplomatic record for the first time, in front of the body that would have to authorize any binding international measure. Russia has already signaled it doesn't think that record belongs here at all, arguing AI security risk is a matter for 'more specialized forums' like the Global Dialogue on AI Governance -- a procedural objection that, if it gains support from other members, could keep this from becoming more than today's one meeting.

None of today's briefers can bind the Council to anything by talking. What they can do is start the paper trail a future resolution, sanctions regime or verification treaty would eventually cite -- and after two weeks of dueling essays and editorials, today is the first time either side has had to make its case in the same room, on the record, in front of the body that would actually have to enforce whatever comes next. (The last time the Council spent this long working through a new technology's security implications -- nuclear non-proliferation in the 1960s -- the gap between first transcript and first treaty ran more than a decade. Nobody involved today is promising AI governance will move faster than that.)