New York bill would make the largest AI developers hire state-licensed 'frontier examiners'

New York lawmakers want to go further than the AI safety law the state passed last year. On October 29, State Sen. Andrew Gounardes and Assemblymember Jordan Wright presented S10731, a bill that amends the Responsible AI Safety and Education (RAISE) Act. That law, which takes effect in January 2027, mainly requires large developers to publish safety frameworks and report critical safety incidents. The new text has only been introduced: the Senate site lists it in the Rules Committee, and no vote has taken place.

Outside examiners and faster reporting

The core of the proposal is a third-party oversight regime. Large developers would have to retain an independent "frontier examiner" licensed by the state. According to the Senate press release, the examiner would audit each year whether a company follows its own safety framework, run broader safety evaluations and issue catastrophic risk assessments every 90 days. Those reports would go to the state and be made public, and examiners would have to alert officials within 72 hours of an emergency.

Developers would also report each quarter on how much of their AI research is automated, which the release calls recursive self-improvement, and on how they use compute. Any substantial jump in risk or in a model's ability to automate AI research would have to be disclosed within 72 hours.

New powers for the state

The bill creates the Office of Digital Innovation, Governance, Integrity and Trust (DIGIT). It could limit the computing power a large developer uses to train new models or to automate AI research, and adopt 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.

Coverage would extend to companies spending more than $1 billion on AI development over three years, with compute thresholds aligned on the European Union's. A legal analysis by Elena Gurevich says this means lowering the frontier model threshold from 10^26 to 10^25 operations. The definition of a critical safety incident would widen to include models evading developer controls and near-misses. Penalties would reach $1 million per day for a first violation and $3 million per day afterwards, and whistleblower protections would extend to employees of large developers.

The sponsors frame the bill as a response to federal inaction; RAISE Act author Alex Bores called the existing law "the floor, not the ceiling." PYMNTS notes that, if enacted, it would widen the gap with the Trump administration, which has sought to limit regulation of AI developers. No reaction from AI companies was reported in the sources reviewed.