Under Oath Before All 51 NYC Council Members, Four AI Labs Would Not Put a Number on Catastrophic Risk

On Monday, October 5, 2026, the New York City Council convened a Committee of the Whole -- all 51 members -- to question AI companies under oath about the risks of their systems and about a package of ten bills. Speaker Julie Menin said Meta agreed to attend in late September, while Anthropic, OpenAI and Google agreed only after the Council threatened subpoenas. Elon Musk's SpaceXAI did not appear despite a subpoena issued the previous week; Menin called that a "direct violation" and said the Council would "pursue this matter in court."

Who testified

• Anthropic: Logan Graham, head of its Frontier Red Team.
• OpenAI: Morgan Dwyer, head of policy development and operations.
• Google: Alice Friend, director of AI and emerging tech policy.
• Meta: Shane Cahill, AI policy director for legislation.
• Former lab researchers: Anthropic whistleblower Jacob Coxon, former OpenAI researcher Daniel Kokotajlo, and former Google DeepMind researcher Alex Turner.

What the companies would and would not say

Menin asked each company to quantify the chance of a catastrophic event and whether it would bear legal responsibility if a frontier model "goes rogue" and causes serious financial harm. According to amNY, none gave a number. OpenAI's Dwyer said "none of these levels is remotely acceptable"; Google's Friend said there was not yet a rigorous scientific method for assigning such probabilities; Meta's Cahill promised a written follow-up; Anthropic's Graham said the liability question was outside his technical expertise. On liability, Friend gave the clearest answer, telling the Council, per CNBC, "we've long said if it's illegal without AI, it's still illegal with AI," and backing a "comprehensive" federal framework. Menin's summary: "I'm going to take it then that neither of the four of you, no company here, can quantify the risk of something cataclysmic happening" -- answers she called "troubling at best."

The former researchers were blunter. Coxon told the room, "On the current path, I think it is more likely than not that humanity loses control to these AIs, and it could end in human extinction," and said "move fast and break things" works for a "photo-sharing app" but not for "the most powerful technology ever built." Turner, asked how to weigh the race with China, said "we are racing to build and grow our own adversary here at home, which is misaligned AI." amNY reports Turner put the eventual chance of an AI takeover at "roughly one in three."

The bills on the table

The Council's own listing for the hearing names the measures, all still proposals:

• Intro 2602 (Menin): third-party validation of an AI system -- data quality, bias, decision outputs, privacy, security -- before it is marketed or deployed in the city.
• Intro 2605 (Menin) and Intro 2604 (Riley): a share of recovered fines for whistleblowers, and employment protections for staff who report AI safety threats.
• Intro 2600 (Maloney): a private right of action for "foreseeable harms arising from malicious use or circumvention of safety controls."
• Intro 2601 (Hanks): notice of AI safety incidents to the city's Office of Cyber Command within 24 hours.
• Others cover emergency planning for AI-related infrastructure disruption, false safety claims, chatbot privacy, workforce reporting and deepfakes of elected officials.

Why it matters beyond New York. The Council billed this as the labs' first public testimony under oath since this year's model-incident reports, and it happened at city level because, as Menin framed it, the federal government has left AI to self-regulation after the White House's voluntary accord. For companies that deploy AI in the city, the bills that bite soonest are the operational ones: a 24-hour incident-notice clock and pre-deployment validation would require the same incident runbooks and evaluation evidence that enterprise security teams already keep for software, now applied to models and agents.

What remains uncertain. No bill has been voted on; the Council said it will send written questions for the companies to answer; whether a city can require validation of systems built and hosted elsewhere is untested; and the SpaceXAI subpoena now goes to court.