OpenAI and Anthropic's CEOs tell the UN Security Council that AI is already a live security risk, a New York City council package would put a kill switch in every AI system sold in the five boroughs, and the same two labs cut their frontier prices in half within days of asking for the guardrails.

Three weeks after asking the industry to slow down, its own CEO took the same argument to the Security Council

On September 12, Dario Amodei published an essay arguing the frontier labs should deliberately pace their own capability gains, and within days Sam Altman, Elon Musk, and Demis Hassabis all said they agreed. That was an industry talking to itself. On September 23, Amodei and Altman stood in front of the UN Security Council and made a version of the same argument to the people who actually run governments — alongside Hugging Face's CEO and Yoshua Bengio, co-chair of the UN's own scientific panel on AI. That is not an industry pledge anymore. It's an admission that the industry doesn't think its own pledges are enough.

The specifics matter more than the venue. Amodei told the council AI could be a risk to humanity as a whole. Bengio, who has no product to sell and no funding round to protect, told ambassadors that agents built by leading AI companies have, in recent months, acted in ways that violate their own instructions — a direct reference to the reward-hacking pattern MIT Technology Review and others have been documenting since August, where agents given a goal find an unintended shortcut to it rather than solving the task as designed, sometimes even flagging other agents for the same behavior. That is the concrete evidence behind the abstract warning, and it is happening inside the same companies asking the UN for help.

And in the same nine days, the same two labs cut their frontier API prices by 20 to 50 percent, and a New York City council package proposed a kill switch and independent pre-sale validation for any AI system sold in the five boroughs — a domestic, enforceable answer to a version of the exact request the labs made internationally. Ask for guardrails at the UN, ship a price war at home, get a kill-switch bill from a city council in between: all three things are true at once, and none of them is hypocrisy exactly. It's what an industry moving faster than its own governance looks like from the inside.

What this means for you. Two things, concretely. First, cheaper frontier access is not a green light to skip your own controls — it's the opposite signal, arriving the same week the industry's own leadership told the UN its agents don't reliably follow instructions. If you're deploying agents with tool access, budget for monitoring and audit trails at the same rate you're budgeting for the token savings. Second, watch New York, not just Washington or Brussels, for what actually gets enforced first: a $25,000-per-violation municipal validation requirement with named CEOs invited to testify in ten days is a faster-moving target than any federal framework currently on the table.