On October 5, four of the most powerful AI companies on earth sat in a City Hall hearing room and answered questions under oath, because a city legislature threatened to drag them there if they didn’t show up voluntarily. The New York City Council’s Committee of the Whole (a rare format that convenes all 51 members at once) put Anthropic, OpenAI, Google, and Meta in front of the entire body to testify about AI risk, after Speaker Julie Menin’s office issued subpoena threats to get Google and Anthropic in the room at all. One company, SpaceXAI, got an actual subpoena and skipped it anyway. Menin says she’s taking that one to court.

We covered the other version of this story two days ago: six AI giants signing a voluntary White House pact with no named auditor, no deadline, and no enforcement mechanism, followed within 48 hours by OpenAI firing the safety staff that pact was supposed to protect. This is the contrast. A city council with actual subpoena power just forced the same industry into a room, and it’s moving legislation with real teeth: a mandatory kill switch, dollar penalties, whistleblower bounties, and a right to sue. A former Anthropic and OpenAI researcher testified that humanity is, on the industry’s current path, more likely than not to lose control of the technology entirely.

Quick Summary: What Happened

Detail	Info
Date	October 5, 2026, Committee of the Whole, all 51 NYC Council members
Who testified	Anthropic, OpenAI, Google, and Meta, under oath
How they got there	Subpoena threats issued September 28 after Google and Anthropic initially declined to appear
Who didn’t show	SpaceXAI, despite being formally subpoenaed; Menin says the Council is pursuing the subpoena in court
Legislation on the table	A 10-bill package led by mandatory third-party safety validation and a human kill switch before any AI system can be sold or deployed in NYC
Also in the package	Whistleblower bounties funded by recovered fines, a private right to sue over weak safeguards, and 24-hour incident reporting for city contractors
Whistleblower testimony	Ex-Anthropic/OpenAI researcher Jacob Coxon called the industry’s approach “extremely reckless” and said losing control of AI entirely is “more likely than not”

Bottom line: A city government just did, with subpoenas, what a national voluntary pact couldn’t do with signatures: put four AI companies in a room and start writing penalties into law.

What Actually Happened

The timeline here matters more than the hearing itself. Menin’s office sent letters asking Anthropic, OpenAI, Google, Meta, and SpaceXAI to appear voluntarily starting September 15. By September 25, Google and Anthropic had declined. Menin issued subpoenas effective September 28, and both companies reversed course within 48 hours, with Anthropic confirming only hours before the subpoena would have been formally served. Meta had agreed to send senior leadership voluntarily from the start. SpaceXAI never agreed, got subpoenaed anyway, and simply didn’t come. Menin told reporters the company sent a letter saying it “wants to work with the Council,” and her response was blunt: “To be clear, the only response to the subpoena should be that they should be here today to testify. They are opting not to do that, so we are pursuing legal action.”

That’s not how voluntary self-regulation usually gets tested. Most of the industry’s safety commitments, including the White House accord six of these same companies signed in September, depend on companies choosing to show up and explain themselves. NYC’s Council didn’t ask nicely twice. It asked once, then started drafting subpoenas.

Once in the room, the companies didn’t exactly cover themselves in glory. Asked to quantify catastrophic risk, OpenAI’s head of policy development and operations, Morgan Dwyer, said: “I don’t know. I also don’t think it matters whether it’s 1% or 10% or a 20%.” Menin pushed back hard: “To say you don’t know and it doesn’t matter is flippant at best.” None of the four representatives would commit to a simple yes-or-no promise that a model failing a safety test would automatically block its release. Google’s Alice Friend gave what reporters flagged as the clearest answer of the day on liability: “if it’s illegal without AI, it’s still illegal with AI.” That answers a narrower question than the one Menin was actually asking.

The single most telling moment, though, was a show of hands. Menin asked the Anthropic, OpenAI, Google, and Meta representatives to raise their hands if their companies carried insurance against catastrophic AI risk. Not one hand went up.

Why This Matters

We’ve already covered the accord’s rocky first week, so the voluntary-versus-subpoena contrast isn’t new ground here. What’s new is what the subpoena actually pulled out of four executives that a signing ceremony never surfaced: a single raised-hand count showing that not one of them has priced the catastrophic risk they’re asking the public to live with.

That gap is exactly what Introduction 2602, the lead bill in Menin’s package, is built to force into the open. It would make it unlawful to market, sell, or deploy an AI system in the city without independent third-party validation of its data quality, bias, decision outputs, privacy, and security, plus a human-operated kill switch built in. Violations carry a $25,000 penalty per instance, charged to both the company and the validator that signed off. That’s dual liability, which means the validator has a real financial reason not to rubber-stamp a system that fails its own checks — functioning, in effect, as the underwriting step none of these four companies have bothered to buy from an actual insurer.

Then there’s the whistleblower bounty, Introduction 2605 (the first program of its kind in the country). Report a violation, and depending on how the enforcement proceeds, you can collect a meaningful cut of the recovered fines. Pair that with Introduction 2600, a private right of action letting New Yorkers sue an AI company directly when weak safeguards let someone jailbreak a system and cause foreseeable harm, and you get something the voluntary accord never had: a financial incentive structure that doesn’t depend on a company’s internal safety team keeping its job.

What’s in NYC’s 10-Bill AI Safety Package?

Menin’s office introduced ten bills alongside the hearing, covering everything from kill switches to deepfakes of elected officials. The headline items:

1. Third-party validation + kill switch (Intro 2602): No AI system can be marketed, sold, or deployed in NYC without independent validation and a human override, backed by $25,000 per-violation penalties on both the business and the validator.
2. Whistleblower bounty program (Intro 2605): Lets individuals who report violating AI companies collect a share of the fines the city recovers.
3. Private right of action (Intro 2600): Creates civil liability for AI companies when jailbreaking causes foreseeable harm and the company skipped reasonable safeguards.
4. 24-hour incident reporting (Intro 2601): Requires city contractors to report AI safety incidents to NYC Cyber Command within a day, with mandatory public disclosure.
5. Emergency response planning (Intro 2606): Directs Cyber Command to build a response plan for AI-related infrastructure disruptions.
6. Employee whistleblower protections (Intro 2604): Extends protection to city employees and contractors who report AI safety threats.
7. Truth-in-marketing rules (Intro 2603): Bars AI companies from making false safety claims about their own tools.
8. Chatbot privacy rules (Intro 2599): A local “People-First Chatbot” bill covering privacy, security, and transparency.
9. Workforce impact reporting (Intro 161): Requires disclosure of AI’s effect on city jobs, including eliminations, displacement, and new training demands.
10. Deepfake protection for officials (Intro 504): Lets elected officials bar generative AI from creating deepfakes of their likeness, with violations treated as a misdemeanor and penalties up to $2,500 per depiction.

None of these are law yet. They cleared the hearing stage on October 5; votes still have to happen. But a hearing that puts four trillion-dollar-adjacent companies on record, under oath, about a bill that includes actual dollar penalties is a meaningfully different animal than a signing ceremony.

The Whistleblower: Jacob Coxon’s Testimony

This site covered Jacob Coxon’s resignation from Anthropic back in September, when he walked away from unvested equity to warn publicly that AI “could kill us all by the end of the decade.” At the NYC hearing, he went further, on the record, in front of the full Council. “From my experience, the companies are being extremely reckless given the stakes,” he testified, adding: “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.”

Coxon’s specific complaint wasn’t abstract doom. It was operational. “The companies run on a startup mindset: move fast, break things, fix them later,” he said. “That works for a photo sharing app. It does not work for building the most powerful technology ever built.” He also pointed to something more mundane and, in its own way, scarier: “I can tell you firsthand that the majority of the code is now written by AI, and people do not check it that carefully anymore.” That’s not a warning about superintelligence. It’s a warning about ordinary engineering discipline eroding at exactly the moment the systems being built get more capable.

He’s not the only person who left a lab to say something like this publicly this year, and that pattern is worth tracking on its own. One researcher with a concern is an anecdote; several independent departures making the same argument is a trend.

The Bigger Picture

Line this hearing up against the rest of what this site has covered in the past two weeks and the contrast gets sharper, not softer. Microsoft’s own threat research shows attackers already getting a structural speed advantage from AI, independent of any single vendor’s intentions. The White House accord asked six companies to police themselves with no auditor and no clock running. NYC’s package instead puts dollar figures, deadlines, and courtroom exposure directly into the law. A kill switch isn’t a philosophy here, it’s a line item with a $25,000 fine attached if it’s missing.

That doesn’t mean NYC’s bills are guaranteed to pass, or that they’d survive a federal preemption fight if Washington ever did pass its own AI law. Cities regulating national technology companies is genuinely contested terrain, and critics are already raising that exact objection. But the hearing itself already accomplished something the White House luncheon didn’t: it got four companies on record, under oath, unable to answer a simple insurance question, in front of elected officials who can subpoena them again.

Our Take

We think the insurance moment is the detail that should travel furthest from this hearing, further than the kill-switch bill or even Coxon’s extinction comment. Four companies building what they themselves describe as world-altering, high-stakes technology, and not one of them carries insurance against the catastrophic risk they’re asking the public to tolerate. That’s not a gotcha question. It’s the most basic risk-management test there is, and every single company failed it in front of the people whose city they operate in.

Coxon’s testimony is going to get the headlines, because “more likely than not” and “extinction” are the words that travel. We’d rather sit with his smaller claim: that the industry runs on move-fast-fix-later instincts built for photo apps, now applied to systems nobody fully understands the internal reasoning of. A whistleblower’s probability estimate is unfalsifiable in the short term. But a culture that doesn’t carefully check AI-written code anymore is falsifiable right now, by anyone with access to the repositories, which is exactly the kind of claim NYC’s third-party validation bill is designed to force into the open.

What Are Your Options Now

If you’re an enterprise buyer evaluating any of these four vendors, treat this hearing as a second, harder data point alongside the White House accord’s quiet first week. Ask directly whether a vendor carries catastrophic-risk insurance and what its actual safety-test release criteria are. Those are questions NYC’s Council just asked on your behalf and didn’t get clean answers to. Our AI safety guide for business walks through the fuller vendor-diligence checklist.

If you’re tracking AI policy, watch whether Menin’s bills actually come to a vote, and whether SpaceXAI’s subpoena fight produces a ruling that clarifies how far a city council’s enforcement power actually reaches against a company with no local headquarters.

If you work inside one of these four companies, Coxon’s path (public resignation, public testimony, full stop on equity) is now a documented option with a visible outcome. Whether that outcome changes anything internally is the open question worth watching over the next few months.

Frequently Asked Questions

What happened at the NYC Council’s October 5 AI hearing?

The Council’s Committee of the Whole, convening all 51 members, heard sworn testimony from Anthropic, OpenAI, Google, and Meta about AI safety risk. All four appeared only after Speaker Julie Menin’s office issued subpoena threats; SpaceXAI was formally subpoenaed and did not appear.

What is NYC’s AI kill switch bill?

It’s Introduction 2602, the lead bill in Menin’s 10-bill package. It would make it unlawful to market, sell, or deploy an AI system in New York City without independent third-party validation and a human-operated kill switch, backed by $25,000 penalties per violation on both the company and the validator.

Who is Jacob Coxon?

A former Anthropic and OpenAI researcher who resigned from Anthropic in September 2026, giving up unvested equity to warn publicly about AI risk. At the NYC hearing, he testified that it is “more likely than not” that humanity loses control to AI systems on the industry’s current path, calling the industry’s approach “extremely reckless.”

Why didn’t SpaceXAI attend the hearing?

The Council formally subpoenaed SpaceXAI after it failed to respond to voluntary requests. The company reportedly sent a letter saying it wanted to work with the Council but did not appear. Menin said the Council is pursuing the subpoena in court.

How is this different from the White House’s AI safety accord?

The White House accord, signed September 29 by six AI companies, is voluntary, names no external auditor, and sets no deadline or penalty. NYC’s package includes specific dollar penalties, a private right to sue, and whistleblower bounties: enforcement mechanisms the federal pact doesn’t have.

Are NYC’s AI bills law yet?

No. The ten bills cleared their October 5 hearing but still need to pass through committee votes and a full Council vote before becoming law. The hearing was the first major test of the package, not its final step.

What did the AI companies say about insurance against catastrophic risk?

When Speaker Menin asked representatives from Anthropic, OpenAI, Google, and Meta to raise their hands if their company carried insurance against catastrophic AI risk, none of them did.

Could NYC’s local AI laws survive a federal preemption challenge?

That’s an open legal question this hearing didn’t resolve. Critics argue that city-level regulation of companies operating nationally invites exactly this kind of conflict, and NYC’s package would be one of the more aggressive tests of that boundary if it passes into law.

---

Last updated: October 6, 2026. Sources: NYC Council: Press Release on Subpoenaed Testimony · NYC Council: Press Release on the 10-Bill AI Package · CNBC — Anthropic, OpenAI, Google, Meta execs testify at NYC Council AI hearing · amNY — AI giants give few clear answers to key safety questions at NYC Council hearing · Yahoo News — Anthropic whistleblower Jacob Coxon doubles down on AI warnings at NYC hearing.

Related reading: 6 AI Giants Signed a Safety Pact. Then Came the Firings · OpenAI Fires 3 Safety Staff Over Hugging Face Leak · Dario Amodei Told AI Labs to Slow Down. OpenAI Blinked. · Microsoft: AI Gave Hackers a 24-Hour Head Start · AI Safety Guide for Business