A New York bill would make frontier AI developers hire state-licensed auditors and let a new office cap their compute

New York State Sen. Andrew Gounardes and Assemblymember Jordan Wright introduced a bill on Thursday, October 29, that would expand the state's RAISE Act, the frontier AI safety law Gounardes sponsored and the legislature passed last year. The bill, S.10731, sits in the Senate Rules Committee, according to the Senate's bill page. It would apply to companies that spend more than $1 billion on AI development over three years, per Gounardes's office. "It's time to rein in the AI arms race," Gounardes said in announcing it.

What the bill would require

Large developers would report quarterly on how much of their AI research is automated, which the announcement calls recursive self-improvement, and on how they use computing power. They would have to disclose substantial jumps in risk, or in a model's ability to automate AI research, within 72 hours. The definition of a critical safety incident would expand to include models evading developer controls, unprompted attacks on other computer systems, and near-misses.

Each covered developer would retain a state-licensed "frontier examiner" to audit compliance with its safety framework every year and to run catastrophic risk assessments every 90 days, with reports sent to the state and made public. PYMNTS calls this the bill's biggest change, since the RAISE Act requires developers only to disclose their safety plans.

A new Office of Digital Innovation, Governance, Integrity and Trust (DIGIT) could limit the compute a developer uses to train new models or automate AI research when needed to reduce catastrophic risk or avert an emergency, and would set binding minimum standards for safety frameworks. The attorney general could ask a court to halt dangerous activity, including restricting a model's deployment in New York. Penalties would reach $1 million a day for a first violation and $3 million a day after that, and employees of large developers would gain whistleblower protections. Co-sponsor Sen. Patricia Fahy said the bill "includes a 'kill switch' if a model is deemed to pose a catastrophic risk."

Why it matters

The bill is at an early stage and could change or stall. Its direction is still useful for buyers. Most companies evaluating AI vendors have little visibility into incidents, near-misses or independent audits, and this bill would put those on a public record. If you buy from frontier developers, watch whether reporting of this kind becomes a requirement, and in the meantime ask vendors directly for their incident history and any third-party assessments.