Representatives from OpenAI, Anthropic, Google and Meta were pressed by New York City council members over the risks of advanced artificial intelligence systems during a hearing held on Monday (5). Under oath, the four companies avoided presenting a numerical estimate for the possibility of a catastrophic event caused by the technology.
The hearing brought together the 51 members of the New York City Council and discussed safety, legal liability and proposals that could impose new requirements on companies that develop or offer AI models in the city.
The companies did not send their CEOs. OpenAI, Anthropic, Google and Meta were represented by executives from the areas of public policy, safety and legislation, all by videoconference.
One of the main moments occurred when the Council president, Julie Menin, asked the companies to quantify the risk of a catastrophic failure. None presented a percentage. The OpenAI representative said she could not estimate the probability and argued that even low risks would be unacceptable given the severity of certain scenarios.
The executives also avoided absolute commitments on legal liability in the event of serious harm and on the automatic suspension of models that failed independent safety evaluations.
Former industry employees also participated. Jacob Coxon, who left Anthropic, said that companies are taking excessive risks in developing more capable systems. Daniel Kokotajlo, a former OpenAI researcher, drew attention to AI advances in cybersecurity.
New York discusses independent validation and fines
The Council is reviewing a package of ten proposals related to artificial intelligence. One of them would require independent validation before certain models are commercialized or deployed in the city and would require the existence of a mechanism that allows interruption by a human operator.
Violations could lead to fines of US$ 25,000 per occurrence. Another proposal would allow lawsuits against companies when foreseeable harms were not accompanied by safeguards considered reasonable.
OpenAI, Google and Anthropic confirmed their participation after the Council threatened to use its subpoena power. Meta had already agreed previously. SpaceXAI did not appear despite having been subpoenaed, and Menin said the Council intends to seek judicial enforcement of the order.
The proposals are still in progress and there is still no final decision on the measures.
