Never lose a customer to a missed message
An AI agent trained on your own business, replying in seconds, in any language, on every channel your customers already use.
Executives from OpenAI, Google, Meta, and Anthropic testified under oath this week at a NYC Council AI hearing convened before all 51 council members, and when pressed on the most basic safety question — what is the numerical probability your technology causes catastrophic harm — none of them gave a number. Three former employees of those same companies, testifying as whistleblowers, were far more direct: one estimated roughly a one-in-three chance of an eventual AI takeover, and another said humanity “more likely than not” will eventually lose control of advanced systems.
The hearing was called by Council Speaker Julie Menin to inform a legislative package that would require third-party safety validation, mandatory human “kill switches,” and 24-hour incident reporting for AI systems deployed in the city — rules that would put New York ahead of federal regulators, who have moved far more slowly on binding AI safety requirements.
What the NYC Council AI hearing revealed
Speaker Menin asked each company representative three direct questions: what is the numerical probability of a catastrophic AI event, would the company commit to halting a model’s release if it failed internal or independent safety testing, and would the company accept legal liability if its system caused serious harm, data exposure, injury, or death. OpenAI’s Morgan Dwyer responded, “I don’t know. I also don’t think it matters whether it’s 1% or 10%,” and declined to commit to blocking releases over failed safety tests. Anthropic’s Logan Graham discussed the company’s risk assessment work without offering a figure. Meta’s Shane Cahill deferred, promising written follow-up answers. Google’s Alice Friend said no rigorous scientific method currently exists for assigning probabilities to catastrophic AI events at all.
The contrast with the whistleblower testimony was stark. Jacob Coxon, who previously worked at both OpenAI and Anthropic, told the council: “We don’t fully control it. We don’t understand its drives or why it does the things it does.” Alex Turner, a former Google DeepMind researcher, put the odds of an eventual AI takeover at roughly one in three. Daniel Kokotajlo, formerly of OpenAI, echoed concerns about increasingly autonomous systems operating beyond meaningful human oversight.
The legislation behind the NYC Council AI hearing
The council’s proposed package goes well beyond disclosure requirements. It would mandate independent third-party validation for AI systems marketed or deployed within city limits, require human override mechanisms — kill switches — for qualifying systems, create financial incentives for employees who report safety concerns, give people harmed by AI systems a legal right to sue, and impose 24-hour incident reporting on any city agency or contractor using AI. If enacted, it would be among the most aggressive municipal AI safety regimes in the country, arriving well ahead of comparable federal action.
The hearing also touched on New York State’s RAISE Act, a frontier AI safety law, with Assembly Member Alex Bores accusing OpenAI of perjury over the company’s stated position on an earlier version of that bill — a charge that, if pursued, would add legal exposure to what was already a reputationally difficult hearing for the companies involved.
Who didn’t show up
Elon Musk’s xAI was subpoenaed to appear and did not, a notable absence given that xAI’s models have drawn their own separate safety scrutiny. The council did not immediately say what enforcement options it has against a company that ignores a subpoena of this kind.
What happens next for New York’s AI rules
The council’s proposed legislation now moves toward committee consideration, where the specific thresholds for “qualifying” AI systems, the mechanics of a mandated kill switch, and the scope of 24-hour reporting requirements will likely be negotiated in more detail. Given the scale of the companies involved and the precedent a binding New York City law could set for other municipalities, expect heavy lobbying from the AI industry as the bill moves through committee, alongside continued pressure from the whistleblowers and safety advocates who testified this week.
Questions people are asking after the hearing
Which companies testified at the NYC Council AI hearing?
Representatives from OpenAI, Google, Meta, and Anthropic testified under oath, along with three former employees of AI companies appearing as whistleblowers.
Did any company give a risk estimate for catastrophic AI harm?
No. All four companies declined to provide a specific numerical probability when asked directly by Council Speaker Julie Menin.
What did the whistleblowers say?
Former researchers estimated risks ranging from a one-in-three chance of an eventual AI takeover to a “more likely than not” chance that humanity eventually loses meaningful control of advanced systems.
What would the proposed NYC legislation require?
Mandatory third-party safety validation, human kill switches, whistleblower incentives, a legal right to sue over AI-caused harm, and 24-hour incident reporting for city agencies and contractors.
Did xAI testify?
No. Elon Musk’s xAI was subpoenaed but failed to appear.
What is the RAISE Act?
It is a New York State frontier AI safety law referenced during the hearing, with an assembly member accusing OpenAI of misrepresenting its position on an earlier version of the bill.
Keep reading on this topic
For more on AI regulatory pressure, see our coverage of the FTC’s probe into AI agents and OpenAI’s warnings on rogue AI agents. We also previously reported on an OpenAI safety employee’s resignation over internal culture concerns.
Sources
- UPI — New York City Council to hold AI hearing with tech leaders. https://www.upi.com/Top_News/US/2026/10/05/new-york-city-council-ai-hearing-openai-meta-anthropic-google/3621791217612
- amNewYork — AI giants give few clear answers to key safety questions at NYC Council hearing amid whistleblower warnings. https://www.amny.com/news/ai-giants-nyc-council-whistleblower-warnings/
Verification your users actually receive.
Send one-time passcodes over WhatsApp with a single API call. Replio can generate, hash and verify the code for you.
