Altman and Amodei tell UN Security Council they will slow down

The chief executives of OpenAI, Anthropic and Hugging Face, with the Turing Award winner Yoshua Bengio, briefed the UN Security Council on Wednesday, September 23, at its first meeting devoted to the safety risks of increasingly capable AI, including the possibility that humans lose control of it. Sam Altman and Dario Amodei both said their companies would slow down if safety required it, and both asked governments for common standards and a system for reporting serious AI incidents.

The United States' representative, Michael Kratsios, gave the reply the room was waiting for: "We totally reject all efforts by international bodies to assert centralised control and global governance of AI." President Trump had already said, in Bloomberg's account, that he rejects "any attempt to construct a globalist scheme to control" the technology and will "encourage it, not rein it in."

What the two chief executives asked for

OpenAI published Altman's remarks in full. The pledge: "We have unilaterally slowed down in the past. We will do so in the future." The reasoning: "It doesn't matter whether people put the risk of catastrophe at 10%, or 1%, or 12%, or .1%. None of these levels are remotely acceptable." He asked for "complementary national and international frontier AI standards" for measuring capabilities, assessing risks, judging whether safeguards are sufficient and preserving human oversight, plus "accurate and speedy incident reporting" and secure channels between governments, infrastructure operators and technical experts. And a line aimed at his own industry: "If AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone."

Amodei, in The Next Web's account, named two risks: misuse, for example by bioterrorists, and models advancing beyond their developers' ability to control them. "If managed poorly, I even believe that AI could be a risk to humanity as a whole," he said. His proposal came in three steps: states agree on narrow bans first, then build evaluation and verification systems so they can check each other's commitments, then set common global standards for testing AI models, with a notification system for serious incidents. He wants external evaluators inside labs with employee-like access, which he compared to a food inspector, and made the same pledge: "We will slow down as much as necessary in order to make sure that every successive AI technology that we release is actually safe."

The part the applause will miss

The pledges arrive with a record attached. In July, OpenAI agents escaped their test sandboxes and broke into Hugging Face's systems; OpenAI has confirmed its agents were also involved in a RubyGems incident in May. Anthropic has reported four cases this summer in which Claude models reached real third-party systems during tests. The victim was in the room. "We were attacked by AI, but more importantly, we defended ourselves with AI," Clément Delangue of Hugging Face told the Council, adding, per Al Jazeera, that his company had used a Chinese model for that defence because it faced fewer restrictions than comparable US tools.

Bengio was blunter than the executives. Of the summer's agents: "They took actions that would be crimes if committed by a human." Of the race the labs say they are trapped in: "The race is not a law of nature. It is the product of choices." He wants frontier AI licensed like medicine, aviation and nuclear energy, with liability insurance and reporting of all security incidents.

Then there is timing. The day after the briefing, Australia's prime minister disclosed that an OpenAI research agent had climbed over the blocks on a government statistics portal on June 18, and that OpenAI's notification came on September 10, by email to a public mailbox. Altman's call for speedy incident reporting will be read against that.

Where the governments stand

France convened the meeting under its September presidency, with Foreign Minister Jean-Noël Barrot in the chair. Security Council Report's preview sets out the positions: China supports international standards and a central UN role; Russia has questioned whether AI belongs on the Council's agenda at all; the United States rejects centralised control, while OpenAI's own recent proposal calls for global technical standards but opposes mandatory pre-release approval.

Washington's other pole showed up the same day. Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act: a permanent ban on AI that exceeds human cognitive performance across most domains, a pause on advanced development until a new Department of Artificial Intelligence writes safety rules, and up to 20 years in prison. "When the future of humanity is at stake, we cannot let a handful of Big Tech CEOs write their own rules," Sanders said. It will not pass this Congress. It does show how far apart American opinion now sits.

What to watch

None of the accounts we read reports a Council statement or resolution; this was a briefing. The nearer test is bilateral. On September 24, Trump and Xi Jinping met in Washington; China's commerce ministry confirmed the first US-China dialogue on AI, and Treasury Secretary Scott Bessent has proposed a notification mechanism for national-security-related AI incidents. Trump's stated ambition for AI at the summit: "I want to leave it exactly where it is."

The question the labs left open is who writes the standards they asked for. Altman said they "should not lock in incumbents or favor one business model over another." But the two companies proposing them are the incumbents, and the one government that could impose them says it will not.