New York lawmakers propose a second RAISE Act with compute caps
Bill S10731 would lower the covered-model threshold to 10²⁵ operations, require state-licensed auditors from 2029 and let a state office limit training compute; it is in Senate Rules.
- Government & policy
- Safety & alignment
- Minor
New York State Senator Andrew Gounardes introduced bill S10731 at the end of October, amending the state’s RAISE Act to widen oversight of frontier AI developers. The Senate’s bill page lists introduction on 28 October; StateScoop reported 29 October and named Assemblymember Jordan Wright as a sponsor alongside Gounardes. The bill was referred to the Senate Rules Committee, where it remained when StateScoop wrote on 6 November.
According to the bill’s listing on the Senate website, it would lower the frontier-model training threshold from 10²⁶ to 10²⁵ operations and replace the existing $500 million revenue test for covered developers with one based on more than $1 billion spent on frontier models over 36 months. It would add new categories of “critical safety incident”, including a model circumventing developer controls or impairing third-party computer systems without consent, and require large developers to retain independent “frontier examiners” for annual audits and 90-day risk assessments, licensed by the state from 1 January 2029.
Developers would file quarterly reports on automated AI research and compute use, and report within 72 hours of a substantial increase in risk. The central new power is over compute: the state oversight office, which StateScoop identified as the Office of Digital Innovation, Governance, Integrity and Trust, could under pre-adopted rules order limits on training compute for all large developers at once, but only to reduce catastrophic risk or avert emergencies. The attorney general could seek injunctions, including restrictions on a model’s deployment in New York, with penalties rising to $1 million per day for a first violation and $3 million per day after that. The bill also extends whistleblower protections to developer employees and contractors. The bill page notes a conflict on timing: the sponsor memo says the act takes effect 30 days after becoming law, while the bill text ties it to the same date as the original act.
Ben Snyder, director of state policy at Encode, a safety advocacy group, told StateScoop that industry self-regulation is not keeping pace with AI risks and that the state needs independent expertise to evaluate company claims. StateScoop set the bill against a federal posture of deregulation, including a Justice Department task force to challenge state AI laws that it said was later succeeded by a “Super Intelligence Task Force”, and against state moves elsewhere: California’s September law on third-party verification and Illinois’s requirement for independent evaluations.
Nothing about the bill’s prospects was reported beyond its committee referral. The original RAISE Act was enacted in December 2025, with its start date pending agreed amendments to align its thresholds with California’s law; New York’s companion-chatbot law had been reported the same week.