The amendments update the requirements for reporting, incident disclosure, and safety documentation for developers of large AI systems operating in New York State.

New York Governor Kathy Hochul signed NY Senate bill 8858 that amended the Responsible AI Safety and Education Act (RAISE Act), expanding disclosure and safety requirements for companies developing or deploying large-scale AI systems.

Under the revised law, “frontier” developers must now submit more detailed public disclosures about their models. Specifically – 

1. Detailed product documentation – This includes training methods, system capabilities, known limitations, and material updates to the product. 
2. Pre-deployment safety testing – Companies must test their models to ensure they’re safe enough for internal use at scale or for public release. Results must be reported.
3. Incident reporting – Developers must report “safety incidents” to the state within 72 hours. These include cases where an AI system causes or materially contributes to unauthorized system access, data exposure, or other harmful outcomes. Companies must provide details on what occurred, how it was detected, and what corrective actions occurred.
4. Internal safety and risk management documentation – Companies must document how they identify, evaluate, and mitigate risks throughout the AI system’s lifecycle. 
5. Testing documentation – The law requires companies to describe how they test systems for harmful or unintended behavior and what safeguards are in place to reduce those risks.
6. Show compliance – Retain records demonstrating compliance with the law’s requirements.

The amendments clarify that these obligations apply to developers of “frontier” AI systems, defined based on computational scale and advanced capabilities. Covered entities must ensure that disclosures remain current when material changes are made to the system.

The amendments go into effect January 1, 2027.