If Washington will not set the test, New York should
Under oath, four leading AI companies would not put a number on the worst case, and none said it was insured against one. A city of eight million is entitled to ask for proof.
On 5 October New York's City Council put OpenAI, Anthropic, Google and Meta under oath and asked them, one by one, to put a number on the worst case. The council sat as a Committee of the Whole, all 51 members, for the first time since 2022. When the speaker, Julie Menin, asked which of the companies carried insurance against catastrophic risk, none of the four raised a hand, TNW reported.
Her lead bill is modest in concept. It would bar selling or deploying an AI system in the city unless it has been checked by an independent validator and can be shut down by a human operator, with a $25,000 penalty per instance. That is the logic of the elevator inspection and the restaurant grade: if a product can hurt the public, someone other than the seller signs off on it.
The companies' own words make the case. They have signed voluntary pledges and promised outside evaluators, and OpenAI's witness told the council that voluntary commitments and self-regulation are not enough, as PPC Land reported. If that is true, someone has to make the checks mandatory, and no federal law yet does.
The objection is the patchwork, and it is a real cost. But a patchwork is how much of American consumer protection began: states and cities moved first, industry asked Congress for one national rule, and Congress eventually wrote one. Menin put it in a sentence: "When the federal government fails to act, cities take the lead."
The bill has gaps. It does not define what counts as an instance, and who may act as a validator is left to the city's Cyber Command to decide. Those should be settled before any vote. The principle is still sound. A city of more than eight million people is entitled to know that a product sold to them has been tested by someone who does not sell it.