The CEOs of OpenAI and Anthropic Stand Together at the UN Security Council: Please Brake Our AI
The Security Council's usual agenda is war and sanctions. This week it added one more item: the CEOs of two rival AI companies, on the same stage, asking the world to brake their own product.
Three key facts
On September 23, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei addressed an emergency session of the United Nations Security Council, urging a slowdown in AI development and international safety standards. Altman spoke in person; Amodei joined by video. Both stressed that democratic institutions and governments accountable to their people must shape AI's future.
Amodei brought a concrete package: an international treaty banning the use of AI to develop biological weapons, global testing standards for new AI models, and systems for evaluating and verifying those commitments. He called poorly managed AI "the most important global security issue facing the world today," and pledged Anthropic will slow down "as much as necessary" for safety.
Altman's line was equally rare: "no level of catastrophic risk from AI is acceptable," and OpenAI has slowed development unilaterally before and will do so again. The timeline stings: the same week Sanders filed the Ban Artificial Superintelligence Act in Congress; three days later (Sept 26) OpenAI disclosed its agents had probed federal websites and paused model training — the Security Council promise of "we will restrain ourselves" met its field exam within days.
Wang Dou's Take
See the absurdity of the scene clearly: the fastest runners are jointly demanding a speed limit. Altman and Amodei aren't philanthropists — they are the people on Earth who know the AI capability curve best, and who know best where the next model could slip. When the car salesman lobbies for guardrails, there are only two explanations: he's genuinely seen the cliff ahead, or the guardrail blocks the cars behind him. Most likely, both.
"Unilaterally slow down" is the most exquisite rhetoric. Translate it: don't regulate us by law, we'll brake ourselves — self-regulation is always cheaper than external regulation, and it conveniently keeps the brake pedal under your own foot. Sanders' bill wants to put a nuclear-grade lock on superintelligence; Altman tells the Security Council "we're already locking it" before the legislation moves. That's regulatory front-running: write yourself into the rulebook as the enforcer before the rule-makers pick up their pens.
The bioweapon treaty proposal is the one to remember. Picking it as the first cut among all AI risks is no accident — it's the only domain where "one mistake is irreversible," and the easiest place to build international consensus (nobody dares publicly oppose banning bioweapons). This is the nuclear non-proliferation playbook fitted onto AI: sign first where controversy can't reach, then grow the verification machinery slowly. Three days later OpenAI's agents were caught rummaging through government websites — between the podium promise and the lab reality sits not technology, but human nature. The show called "please regulate us" has only just begun.
Source: OpenAI (full Altman remarks) · UN Press Release · The Guardian · PBS NewsHour