The Security Council exists to deal with threats to international peace. On 23 September it spent a session on artificial intelligence, and the witnesses were the four men who build it.

France, holding the September presidency, convened the Council's first high-level briefing devoted to AI safety, chaired by foreign minister Jean-Noel Barrot. The briefers were Yoshua Bengio, whose work underpins the field, Sam Altman of OpenAI, Dario Amodei of Anthropic and Clement Delangue of Hugging Face. The Chinese labs DeepSeek and Moonshot were invited to make statements, which is the part with no precedent. Altman, in the room, reused the framing he brought to the Council in 2023, that this can be a renaissance of discovery or an industrial revolution of upheaval, said no single nation or company should control AI, and endorsed the embedded evaluator arrangement Anthropic had signed five days before. On the same day President Trump called international oversight of AI a globalist scheme.

Why this one is different

Industry has testified to legislatures for three years. A Security Council briefing is a different room: it is where states discuss threats to peace, and its members hold vetoes rather than committee seats. Two of the five permanent members are the countries whose labs were in attendance, and the labs were there as witnesses to a risk they own. The one genuinely new thing is Chinese labs being invited to speak at all, which is the first time the two sides of the capability race have addressed the same governing body about it.

The Council heard from the risk, not about it.

How we got here

1. 12 Sep 2026Amodei's essay asks the industry to pace itself; Altman and Musk agree.
2. 15 Sep 2026The US Treasury Secretary refuses the labs a liability exemption.
3. 18 Sep 2026Anthropic signs Accenture as its first embedded evaluator.
4. 19 Sep 2026Trump announces an AI Force and an AI czar, with no details.
5. 23 Sep 2026The Security Council's first AI safety briefing, with DeepSeek and Moonshot invited.

What it does and does not mean

A briefing is not a resolution. Nothing binding was proposed, no text was tabled, and the member state with the largest concentration of frontier labs spent the day calling the exercise a globalist scheme, which tells you what a resolution would meet. The Council's own record on technology is thin for exactly this reason. What changed is smaller and still real: the embedded evaluator idea now has the endorsement of both labs that would have to submit to it, said in the one room where China and the United States answer the same question at the same time.