OpenAI, Meta, Anthropic, Google unable to quantify catastrophic AI risk at New York City Council hearing

At an October 5 New York City Council hearing, policy heads from OpenAI, Meta, Anthropic, and Google testified under oath but could not quantify the risk of AI in the worst catastrophic scenarios. Morgan Dwyer, OpenAI's head of policy development, said in response to a question from Council Speaker Julie Menin that she did not know the specific probability and considered 1%, 10%, or 20% all unacceptable; Menin called the response flippant. All four companies emphasized their commitment to safe AI but could not guarantee that models would always follow safety guardrails, nor could they commit to legal liability for harm caused. The hearing lasted several hours, with about 30 council members questioning on issues such as AI-driven job loss, artist protections, and data centers. All 51 council members attended, and 10 pieces of legislation regulating AI development, deployment, and use were introduced, one establishing a whistleblower incentive program with rewards up to 25% of recovered funds, and another requiring AI systems deployed by the city to be equipped with a kill switch.