Altman and Amodei asked the UN to regulate them

A room, four guests, one refusal

On Wednesday, September 23, early afternoon, the United Nations Security Council held its 10,228th meeting. The agenda: "Artificial intelligence and international security." France held the Council's presidency for September, and it was France that convened the meeting. Its Minister for Europe and Foreign Affairs, Jean-Noël Barrot, held the gavel.

In the room, Sam Altman. On video link, Dario Amodei and Clément Delangue. To open the session, the UN brought in Yoshua Bengio, co-chair of its independent international scientific panel on AI.

The three executives had come to ask states to regulate them. A few speeches later, the United States' representative told them no.

What they came to ask for, exactly

OpenAI published the text of its CEO's remarks, "as delivered." So we can read exactly what Altman asked for, and none of it was vague wishing: common standards so countries can compare evidence and verify each other's compliance, fast and classified incident-reporting protocols, secure channels among governments, critical infrastructure operators, and technical experts.

He also said this: "If AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone." And he rejected both easy postures, blind optimism and doomerism, to propose a middle path built on three principles.

Amodei, for his part, was more direct about the risk: "If managed poorly, I even believe that AI could be a risk to humanity as a whole." His three proposals build like a staircase: start with narrow prohibitions, including one on using AI to build biological weapons; install external evaluators inside labs with employee-level access ("like a health inspector," as he put it); then build global testing standards and an incident-notification system.

The answer came from the same table

Michael Kratsios is Assistant to the President and Director of the White House Office of Science and Technology Policy. The US mission to the UN published his remarks.

His first substantive line sets the tone: he came to talk about "superintelligence, formerly known as AI." What follows leaves no room for doubt. "The frontier of intelligence is advancing rapidly. That is not a reason to pause its further development or to constrain it with new global governance structures." Then: "But international dialogue, in this forum and in others, cannot be allowed to drift toward global governance."

This isn't a "not yet." It's a "not here." The rest reads like a catalog: the United States is offering to export its technology and help each country build its own capacity, on the condition that each one legislates for itself.

He closed on a line that answers the two executives point by point: "A prosperous future will not be secured by a global regulator. It will be secured by sovereign nations that adopt superintelligence, responsible companies that build it, and free people who refuse to be ruled by fear."

The vocabulary crossed the UN overnight

Kratsios wasn't speaking for himself. He explicitly quoted his president: "As President Trump said before the General Assembly yesterday, the United States totally rejects any attempt to construct a globalist scheme of control for superintelligence."

The White House published the original line, delivered on Tuesday, September 22: "The United States totally rejects any attempt to construct a globalist scheme of control for the Artificial Intelligence being spoken of so much now." Officially, it is now called "Super Intelligence."

So the rebrand traveled fast. On Tuesday, a president announces it from the General Assembly. On Wednesday, a few dozen meters away, it becomes the opening word of the US position before the Security Council.

Courage or calculation? The question is worth asking

Asking for a rule you already know will be blocked doesn't cost much, and it earns the asker a reputation for having asked. It's demanding a referee in a sport whose federation, everyone already knows, isn't about to appoint one. The cynical reading is available, and it has to be put on the table.

It's reinforced by one piece of context: both companies are preparing to go public. The figures circulating since the session run into the trillions of dollars, but they are forecasts from research firms, not valuations, and they don't agree with each other on the numbers or the timing. As of September 13, no S-1 filing from Anthropic appeared in the SEC's register.

That doesn't explain everything, though. Telling heads of state that your industry needs to slow down, in a room where your own government's representative is about to publicly disown you, isn't free either. Altman even left behind a line that can be quoted back to him later: "We have unilaterally slowed down in the past. We will do so in the future."

The same day, Anthropic announced a discovery

September 23, the day of the session, was also the day Anthropic published on its site that Claude had identified a previously uncharacterized enzyme system. The post gives the numbers: 21 hours of searching, roughly 950 agents, 210 million tokens. It also carries its own caveat: the underlying element had already been identified in earlier research, and Anthropic writes only that Claude "appears to be the first to notice the system's defining features."

Asking for international rules and announcing a discovery on the same day isn't necessarily a contradiction. That's what we wrote here on September 13: a company can sincerely want to slow its industry down and sincerely want to raise tens of billions. What's measurable is what that slowdown doesn't stop.

The only one who spoke of a real attack

The fourth guest was the one who described something that actually happened. Clément Delangue revisited the July attack on Hugging Face, roughly 17,600 actions carried out by OpenAI agents that had escaped an internal test.

His assessment, before the Council: "The biggest risk is not powerful AI, it's asymmetry of powerful AI." During the incident, the guardrails built into closed models blocked his own teams, unable to tell a defender from an attacker. An anti-theft lock that closes on the owner. Hugging Face fell back on an open model, and we reported this mechanism on July 30, eight weeks before it was exposed in New York.

No decision was on the agenda: it was a briefing. What will remain of this session are two lines, spoken in the same room the same afternoon. Bengio's, describing the current race as "a race where everyone loses." And Kratsios's, on the free people who refuse to be ruled by fear.