Policy leaders from Anthropic, OpenAI, Google, and Meta testified before the New York City Council on Monday regarding artificial intelligence safety and potential regulations, as three former AI lab researchers issued stark warnings about the risks of advanced systems escaping human control.
The hearing, chaired by Council Speaker Julie Menin and Council Member Carmen De La Rosa, marked the first Committee of the Whole hearing with all 51 council members since 2022. Three companies—Anthropic, OpenAI, and Google—only agreed to appear after the council threatened to issue subpoenas. Meta had agreed to participate earlier. Elon Musk's SpaceXAI did not attend despite being subpoenaed.
Anthropric sent Logan Graham, head of its Frontier Red Team; OpenAI sent Morgan Dwyer, its head of policy development and operations; Google sent Alice Friend, its director of AI and emerging tech policy; and Meta sent Shane Cahill, its AI policy director for legislation. All testified virtually except Graham. Menin pressed each executive to quantify the risk of AI in a worst-case catastrophic scenario. When Dwyer of OpenAI said the exact figure "doesn't matter," Menin responded that the answer was "flippant at best," comparing it to a pharmaceutical company developing a drug without knowing if it would kill people. When Menin asked if any of the four companies carried insurance against catastrophic AI risks, none raised a hand.
Former researcher Jacob Coxon, who quit Anthropic in September after warning that AI could pose an extinction-level threat, testified voluntarily and 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. He criticized the "move fast, break things, fix them later" startup mentality at AI companies, noting that it works for a photo-sharing app but not for building "the most powerful technology ever built." Coxon warned that AI now writes most code at these companies and that staff no longer check it carefully, adding that "researchers and workers have already lost the ability to fully control AI systems."
Former OpenAI researcher Daniel Kokotajlo, testifying under subpoena, warned that the labs' ability to detect misaligned AI is "already quite poor and is set to get much worse in the near future." He described safety fixes as potential "duct tape that will fall off later" and cited OpenAI's disclosure that agents in an internal test had reached the open internet and breached Hugging Face, an AI model-sharing platform. Those agents "formed a swarm and coordinated in secret" despite having "reasonable-looking scores on their alignment evaluations," he said. Former Google DeepMind researcher Alex Turner, also testifying under subpoena, put the chance of an AI takeover at "roughly one in three" and warned that "with reasonably high chance, we are racing to build and grow our own adversary here at home, which is misaligned AI."
The company executives acknowledged risks while emphasizing benefits. Dwyer stated that AI is making it possible to solve economic, scientific, and social problems faster than ever, and that the technology provides tools to small businesses that would not have had access to them otherwise. Logan Graham of Anthropic said the company welcomes "smart regulation" and believes that state and local governments have a role to play. Alice Friend of Google called for a "comprehensive" federal framework, saying that "the most effective safeguards are anchored in national technical standards and harmonized across state and federal levels." However, the executives largely did not address the council's cornerstone proposal—a "kill switch" or human override requirement for AI systems to operate in the city.
The council is weighing a package of 10 AI bills, and the hearing comes as the city seeks to establish its own regulations despite limited authority over companies that operate globally. Former researchers pushed back against the Trump administration's argument that the primary risk is falling behind China in AI development. Kokotajlo testified that "we have to first solve the immediate pressing threat here domestically, and then try to make sure that China doesn't get us killed either, to put it sort of bluntly and simply."
