NYC Council Puts OpenAI, Anthropic, Google and Meta Under Oath -- SpaceXAI Skips Its Subpoena

Coverage of the hearing split along a predictable line. CNBC and UPI reported the event straight -- who attended, who didn't, what was said -- without characterizing how the companies came across. The City Reporter and amNY, by contrast, led with the companies' evasiveness, framing the entire session around "few clear answers" to the hearing's central safety questions. Both framings describe the same transcript; the disagreement is about emphasis, not fact, and neither outlet disputes any of the other's quotes. This piece follows the straight account for what happened and uses the pointed framing only where a named council member, not a reporter, is the one drawing the conclusion -- Krishnan's on-record rebuke, not an outlet's adjective.

On Oct. 5, New York City's Council did something it says it had never done before: it convened all 51 of its members at once, under its rarely-used Committee of the Whole, to put the country's four largest AI companies under oath. OpenAI, Anthropic, Google and Meta each sent a named policy executive, appearing by video link, to answer questions about whether the systems they build could cause harm the companies themselves cannot fully predict.

The Committee of the Whole is the Council's full-body format, normally reserved for matters of significant citywide importance and rarely convened for oversight hearings at all. (Routine legislative business goes to a smaller standing committee; pulling all 51 members into one room is itself a statement about how seriously the Council wanted this treated.) The Council had announced the hearing on Sept. 25, alongside the 10-bill legislative package, giving the companies ten days' notice. OpenAI sent Morgan Dwyer, its head of policy development and operations; Anthropic sent Logan Graham, head of its Frontier Red Team; Google sent Alice Friend, its director of AI and emerging-tech policy; and Meta sent Shane Cahill, its AI policy director for legislation. All four appeared remotely.

The hearing's sharpest moment didn't come from a company at all. Jacob Coxon, the researcher who resigned from Anthropic on Sept. 8 warning that frontier labs were "racing straight to self-improving superintelligence and gambling with our lives," testified again in person. "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," he told the Council -- a claim that is his own forecast, stated under oath, not a measured probability any instrument has produced. Alex Turner, a former Google DeepMind researcher, followed with a related warning: that misaligned AI "is everyone's adversary, including our own."

“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.” — Jacob Coxon, former Anthropic researcher, testifying Oct. 5, 2026

What the companies themselves offered was thinner. Councilmember Shekar Krishnan pressed the panel on whether AI-driven losses could expose the city's own pension systems to liability; when representatives answered with general statements about corporate performance rather than the specific risk he'd asked about, Krishnan pushed back on the record: "The fact that either you don't have answers to this ... does not answer the question." No company present offered a direct answer on pension liability during the hearing.

SpaceXAI -- the Musk-controlled entity formed when xAI merged into SpaceX earlier this year -- simply didn't appear, despite being subpoenaed. Council Speaker Julie Menin said afterward: "They are opting not to do that, so we are pursuing legal action," naming a planned filing in state court to compel testimony. It is the one concrete escalation to come out of the hearing; the four companies that did appear face no comparable follow-up so far.

None of this happened in a vacuum. The hearing was the direct continuation of a sequence that started a month earlier, when a researcher's resignation letter became the loudest public argument the industry had had about its own risk in months.

Even if every one of the Council's 10 bills passes, the limits of the exercise are built in. City lawmakers have no authority to write the federal rules that executives themselves say are the only ones that would actually govern frontier model training and deployment nationally -- the bills on the table target how the city defends its own systems and what AI companies must disclose to it, not how any company builds its models. That gap is the real subtext of Friday's hearing: a city government using the loudest tool it has -- subpoena power and a full-body public hearing -- against a technology whose most consequential decisions are made nowhere near City Hall.

That subtext connects directly to ground this Council had already staked out. The private right of action and whistleblower-bounty bills introduced two weeks before the hearing were written with Coxon's kind of departure in mind -- a way to formalize what, right now, only happens when a researcher quits and goes to the press. Whether city law can actually compel what a subpoena apparently couldn't is the question the next vote, not the next hearing, will answer. A private right of action would let a resident sue directly over an AI-caused harm rather than waiting on a regulator to act; a whistleblower bounty would pay a researcher like Coxon for coming forward through a formal channel instead of a resignation post on X. Neither exists yet. Both were still bills, not law, on the day four companies testified about risks the Council itself has no power to regulate at the source.