At NYC's AI hearing, OpenAI wouldn't put a number on the worst case, and SpaceXAI didn't show

New York City Council held its hearing on the risks of advanced AI at City Hall on Monday, Oct. 5, 2026. Former Anthropic engineer Jacob Coxon and other recently departed lab employees warned that the companies are racing to build systems they cannot control. Representatives of OpenAI, Anthropic, Google, and Meta testified remotely, and Speaker Julie Menin said Elon Musk's SpaceXAI skipped the hearing in violation of the Council's subpoena.

A city council just did what Congress hasn't: it put the biggest AI labs on the record about how bad things could get. The most telling moment wasn't the doom warning from a former insider. It was OpenAI saying the odds of catastrophe don't matter, which is a strange answer from a company asking to be trusted with them. And the one lab that didn't show up is now headed to court, which will test whether a city can actually make these companies answer questions.

On Monday, 5 October 2026, the New York City Council held a hearing on the risks of advanced AI at City Hall in Manhattan. Speaker Julie Menin led it. The Council had scheduled it as a Committee of the Whole, a hearing of the full Council rather than a smaller committee. The Associated Press, CNBC, CBS New York, and Bloomberg all reported the hearing that day.

Jacob Coxon testified. He is a former Anthropic engineer who quit last month and used that exit to warn, in public, about the labs. The Associated Press quoted him at the hearing: “The companies run on a startup mindset: Move fast, break things, fix them later. That works for a photo sharing app. It does not work for building the most powerful technology ever built.” He also said: “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.” Bloomberg quoted him saying, “We do not know how to control any AI system yet.” Those lines are his, as those newsrooms reported them.

The Associated Press reported that recently departed employees of Anthropic, OpenAI, and Google DeepMind described a reckless culture at the firms. The story says the biggest companies are racing to build technology they do not understand how to control. CNBC named two of the other witnesses. Alex Turner, a former Google DeepMind researcher, said China is not the only adversary, and that “with reasonably high chance, we are racing to build and grow our own adversary here at home, which is misaligned AI.” Misaligned, in that sentence, means the system is not doing what people asked it to do. Daniel Kokotajlo, a former OpenAI researcher, warned about the same move-fast culture and called on the industry to be more open and to slow the leading models. Those lines are CNBC’s account of their testimony.

Four companies sent people, and they testified by video rather than from the room. Associated Press photo captions name Morgan Dwyer of OpenAI, Shane Cahill of Meta, Alice Friend of Google, and Logan Graham of Anthropic, and say they testified via Zoom. CNBC gives the jobs. Dwyer is OpenAI’s head of policy development and operations. Cahill is Meta’s director of legislation and policy development. Friend is Google’s director of global AI public policy. Graham heads Anthropic’s Frontier Red Team, the group that tries to break a system on purpose so the lab can find the danger first. They told the Council their systems already improve daily life and that they are focused on safety. Dwyer said: “I share your commitment to ensuring that technology is developed safely and that its benefits are widely shared.”

Menin asked each company to “quantify the risk posed by AI in the worst case catastrophic scenario.” Quantify, here, means put a number on it. The Associated Press reports Dwyer’s answer: “I don’t know — I also don’t think it matters whether it’s 1% or 10% or 20% chance that something catastrophic will go wrong. None of these levels is remotely acceptable. We should not train models that we cannot make an extremely strong case that we can keep under human control.” Menin called that answer “flippant at best.” Dwyer then said: “I would say that the percentage chance doesn’t matter. What matters is the commitment to safety, and that is what OpenAI is committed to doing.” Those sentences are the Associated Press account of the exchange.

SpaceXAI was the only company the Council had called that did not send anyone. CNBC reported that Menin called the absence a “direct violation of the subpoena” the Council issued last week, and that she said the Council will “pursue this matter in court.” A subpoena is a legal order to show up and answer. CBS New York reported that the company answered the subpoena with a letter saying it wants to work with the city, and that it still did not attend. CBS also reported that the city is pursuing legal action.

Who agreed to come, as Menin told it. CNBC reported that Meta confirmed it would take part on its own, while Anthropic, OpenAI, and Google agreed only after the Council threatened to subpoena them. That matches the Council’s 28 September release, which said Meta agreed during the first round of letters and the other three agreed only after the threat. SpaceXAI is the company that received the subpoena and then did not appear.

The Council is weighing a package of bills at the same hearing. The Associated Press says the proposals include requiring companies to disclose information about AI tools, requiring third-party approval for AI systems used in the city, and letting whistleblowers receive a portion of fines paid by AI companies that break the law. A whistleblower is a person who reports a problem from inside. Third-party approval means an outside checker, not the company grading itself. CBS New York says the Council is considering 10 bills. Before the hearing, Menin told CBS that a kill switch is in the legislation, and that New York State and California have both talked about one. A kill switch, here, is a human override that can shut a system down.

The Council’s 25 September release is the list behind those headlines. Speaker Menin’s Introduction 2602 would make it unlawful to market, sell, or deploy an AI system in the city without that outside check. The checker would have to look at data quality, bias, the decisions the system produces, privacy, and security, and would have to say if it has a conflict of interest. The same bill would require a kill switch, and the checker would have to confirm it is there. The release names a $25,000 penalty each time a system is marketed, sold, or deployed without the check, or if the check was faked. It says both the business and the checker can be fined. Her Introduction 2605 would let an individual whistleblower receive part of the fines recovered from an AI company that breaks the law. The release calls that a first-in-the-nation approach.

The same release lists the rest of the opening package. Council Member Virginia Maloney’s Introduction 2600 would let a person sue an AI company for foreseeable harm from a jailbreak. A jailbreak, in the bill, is someone getting around the safety controls. The person would have to show the harm was foreseeable, the company did not use reasonable safeguards, and a third party caused the harm by exploiting that failure. Majority Whip Kamilah Hanks’s Introduction 2601 would require a city contractor, or the agency that hired them, to tell the Office of Cyber Command in writing within 24 hours about an AI safety incident on a covered contract, and would require that office to tell the public within 24 hours. Council Member Chi Ossé’s Introduction 2606 would require a plan, written with Emergency Management, for an AI event that hits city systems or public safety. Council Member Kevin Riley’s Introduction 2604 would extend the city’s whistleblower law to employees and covered contractors who report AI use they know, or reasonably believe, is a public safety threat.

Four more bills are on that same list. Council Member Carl Wilson’s Introduction 2603 would require certain disclosures about AI tools and would bar false or misleading claims about their safety. Council Member Frank Morano’s Introduction 2599 would set privacy, security, and transparency rules for chatbot providers, the companies that offer a chat program people talk to. Council Member Carmen De La Rosa’s Introduction 161 would add job impacts to the city’s yearly report on algorithmic tools. An algorithmic tool, here, is a program an agency uses to score, sort, or decide. The report would count positions cut, people moved, pay changed because the work changed, and employees sent to new training. Deputy Speaker Nantasha Williams’s Introduction 504 would let an elected official or a candidate tell the owner of a generative AI system not to make fake audio, photos, or video of them. Generative, here, means the system makes new media. After that written notice, the owner would have to block those uses. The release says a violation is a misdemeanor and a fine of up to $2,500 per depiction, and that it does not apply when a third party the owner does not control makes the media. The release also says further bills would cover deceptive deepfakes, fake media meant to mislead, and a clearer picture of how these tools change job duties.

The federal backdrop, as the Associated Press frames it. The story says the Trump administration has taken a passive approach, that Congress has been inactive, and that President Trump has dismissed AI fears as a “hoax” and signed an agreement late last month that largely lets the companies police themselves. CNBC reported that Menin opened the hearing by pointing at that light federal touch, after the voluntary accord tech leaders signed at the White House, and quoted her: “The idea that artificial intelligence is going to self-regulate defies all reason.” The Associated Press also says Gov. Kathy Hochul has moved to impose reporting and transparency rules for AI in New York State, as have governors elsewhere, and that a state’s reach is limited because the technology crosses state lines. Those lines are the newsrooms’.

The photograph is an official picture from the hearing. Speaker Julie Menin sits at the dais with a gavel in front of her and a nameplate that reads Julie Menin, Speaker. She is questioning witnesses. Gerardo Romo took it for the NYC Council Media Unit. It is the 5 October hearing. It is not a picture of the September subpoena, and it is not a portrait of Coxon.

In plain terms, New York’s city legislature put four of the biggest AI labs on the record on Monday, by video, and heard a former Anthropic engineer say that on the current path humanity is more likely than not to lose control. OpenAI’s witness would not put a number on the worst case, and said the number does not matter. SpaceXAI did not show up. Speaker Menin said that breaks the Council’s subpoena, and that the Council will take it to court. The bills on the table are still proposals.