Analysis
Three former AI lab researchers testified before the New York City Council on October 5 that humanity faces a real risk of losing control over advanced AI systems, while Google, OpenAI, Anthropic and Meta sent policy representatives after the council threatened subpoenas, Fortune reported. Jacob Coxon, who previously worked at both OpenAI and Anthropic, testified voluntarily: "it is more likely than not that humanity loses control to these AIs."
Subpoenas, Not Invitations
Daniel Kokotajlo, now director of the AI Futures Project after leaving OpenAI, and Alex Turner, a former Google DeepMind researcher, were both subpoenaed rather than appearing voluntarily. Kokotajlo warned that current safety measures may be "duct tape that will fall off later," while Turner put the odds of an AI takeover scenario at roughly one in three. SpaceX-AI was also subpoenaed and did not appear -- the one notable absence among the major labs. Council member Julie Menin, who organized the hearing, has proposed legislation requiring independent validation of advanced AI systems, a mandatory human shutdown capability, fines of $25,000 per violation, whistleblower protections, and the ability to sue over foreseeable harms from jailbroken AI tools.
“Kokotajlo warned that current safety measures may be "duct tape that will fall off later," while Turner put the odds of an AI takeover scenario at roughly one in three.”
The hearing lands days before Anthropic's own IPO prospectus reportedly devoted roughly a third of its risk-factor section to existential AI risk, and as OpenAI pushes toward its own public listing despite pausing parts of its offering over safety concerns. The researchers' core complaint isn't abstract: they argued AI systems are increasingly bad -- and getting worse -- at flagging their own misalignment, that most code inside frontier labs is now AI-written with limited human review, and that none of the major labs carry meaningful insurance against catastrophic AI risk.
None of this testimony is independently verifiable by a city council -- Coxon, Kokotajlo and Turner are all former employees with no current access to the systems they're describing, and their probability estimates are personal judgments, not measured outcomes. The labs that did show up sent policy staff, not their CEOs, and offered no public rebuttal to the specific numbers cited.
The skepticism isn't limited to critics of AI risk messaging -- it's bipartisan in its own way: a Republican senator sent Anthropic CEO Dario Amodei a letter the same week accusing him of an "alarmist" approach that undercuts public trust in AI investment (Axios), while city council members argue the labs aren't taking the risk seriously enough. Both critiques can't be right at the same time, and that contradiction is exactly why this debate keeps resurfacing without resolution.
Menin's bills would be among the first binding AI-safety rules passed by a US city rather than a state or federal body, and their $25,000-per-violation fines and whistleblower provisions would give city regulators tools federal lawmakers still don't have. Whether New York can enforce rules against companies headquartered elsewhere -- or whether other cities copy this once Washington keeps stalling -- is the open question.
