Today in New York: the AI labs testify under oath. The bills are the story.

Today, the New York City Council's Committee of the Whole convenes all 51 members to put the AI industry under oath. Anthropic, OpenAI, Google, and Meta will testify, per the Council's September 28 announcement, and Elon Musk's SpaceXAI has been subpoenaed after failing to respond at all. The Council says this is the first public testimony under oath from the major AI firms, and the trigger is explicit: "weeks of serious warnings from whistleblowers and AI executives, and reports that rogue AI agents have already inflicted damage to government databases."

Note the choreography. Letters went out September 15 to 17. Meta agreed. Google and Anthropic declined. Speaker Julie Menin then authorized subpoenas effective 9:00 a.m. on September 28. Over the weekend, OpenAI and Google agreed to appear; Anthropic confirmed late Sunday night, "just hours before the subpoena was due to be served." Only SpaceXAI is still holding out. If it does not comply, the Council says it will seek enforcement in New York State Supreme Court.

The hearing is theater. The legislation is the story. On September 25, the Council published a package the press has called one of the most ambitious local AI frameworks in the country. Its provisions, as summarized in reporting on the package (the bill text itself was not independently readable at press time, so treat specifics as reported):

Independent third-party validation for any AI system sold or deployed in the city, covering data quality, bias, decision outputs, privacy, and security, under standards set by the city's Office of Cyber Command. Every AI system must have a human-operated kill switch. Non-compliance carries a $25,000 fine per instance, and in agent-swarm deployments the fine applies per individual agent. A first-in-the-nation whistleblower incentive program, paying tipsters a share of recovered fines, plus expanded protections for city employees and contractors who report AI safety threats. A private right of action letting New Yorkers harmed by AI tools sue companies directly, where the harm came from circumvention of safety controls and the company failed to implement reasonable safeguards. And 24-hour AI safety incident reporting for city contractors.

Three of those provisions should change what you build this month, even if the bills never pass.

The kill switch is the operational one. A human-operated manual override on every deployed AI system is a design requirement, not a policy sentence. If your agent fleet cannot be stopped by a human who is not the agent, you are already non-compliant with the shape of regulation that is coming. Build the stop button now. The per-agent fine structure, if it holds, turns every additional agent instance into a liability unit, which changes the economics of sprawling multi-agent deployments.

The whistleblower provisions are the cultural one. Whistleblower rewards aimed at AI safety mean your employees' silence is no longer priced at zero. If your incident disclosures are slow, the Council is betting someone inside will sell the timeline. The DIVD disclosure last week, public within days, is the model; the months-long corporate reviews are the thing this bill is priced against.

The private right of action is the legal one. Legal experts quoted in the coverage call it potentially the most consequential liability provision in the country. If harm from "jailbroken" agents becomes a private lawsuit rather than a regulator's enforcement choice, your safety claims become exhibits. Last week this publication covered the FTC's probe under unfair-or-deceptive-practices authority, where the same logic applies: your marketing about guardrails is a statement a court can hold you to.

Two cautions. First, the hearing's outcome is not knowable at press time, and nothing here is law yet. Do not plan around the bills passing as written. Second, this is a city council, not Congress. It cannot set national AI policy. But city councils do something Congress cannot: move fast and create a template. Whatever NYC passes, the next three cities will copy with local edits. The compliance shape, validation evidence, human kill switches, fast incident reporting, defensible safety claims, is the same shape federal and state regulators keep circling. Build to it once and you are built for all of them.

Watch the testimony for one thing: what the companies commit to under oath. Sworn statements create a record. The voluntary era, in which the labs signed a one-page "morally binding" accord at the White House while a federal probe opened the next day, is colliding with a sworn one. Today's hearing is where you see which executives are willing to say what they believe on the record. Plan your vendor relationships accordingly.