Assembly advances RAISE Act to require safety plans and reporting for the largest AI developers

The Assembly on June 13 approved the RAISE Act, legislation that requires very large artificial intelligence developers to adopt formal safety plans, disclose certain “critical safety incidents” to the state attorney general’s office and implement cybersecurity protections for frontier-scale models. Sponsor Assemblymember Jeff Boris said the measure targets a narrow set of companies and models and is meant to avoid a patchwork of state regulation while giving New Yorkers protections from the most extreme risks.

Under the bill, developers that meet the bill’s “large developer” threshold (explicitly tied to training compute spend on final model training runs) must maintain documented safety plans and provide timely notice of incidents that meet a statutory definition of critical harm. The bill focuses on the largest, most capable systems—described in the text as “frontier” models—and includes a definition that points at the highest-capability runs. The law does not create a private right of action; enforcement is carried out by the state Attorney General.

Sponsor Jeff Boris described the measure as “light touch” and said it was drafted to avoid measures that had raised opposition in other states while still addressing capabilities that experts say require urgent caution. He said the bill is intended to capture the extreme harms that experts worry about—such as models that could materially enable misuse at scale—rather than to regulate every use of AI or academic research. The bill explicitly exempts academic research and clarifies it applies only once a company crosses a high compute-spend threshold.

Floor debate reviewed several key tradeoffs. Members asked whether the bill would drive companies out of New York, whether it would inadvertently push firms to obscure spending figures or to shift research offshore, and whether the law would be future-proof as models and workloads evolve. Boris and others said the bill had been modified after stakeholder conversations and that the penalty structure and the “large developer” definition were chosen to keep the law narrowly focused on a handful of companies and the most consequential risks.

Supporters praised the bill as a state-level stopgap where federal action has not yet arrived. Critics urged caution about state-by-state regulation and pressed for clearer definitions in places.

Ending: The Assembly passed the measure; sponsors described it as a model for balanced state action to mitigate high-end risks from frontier-scale AI development while preserving innovation in New York.