Overview of the RAISE Act
The Responsible AI Safety and Education Act (RAISE Act), signed by New York Governor Kathy Hochul on December 19, 2025, establishes regulatory requirements for developers of large frontier artificial intelligence models. The law was introduced in the New York State Assembly by Assemblymember Alex Bores on March 5, 2025, and in the Senate by Senator Andrew Gounardes on March 27, 2025. It was amended in March 2026 to align more closely with California’s Transparency in Frontier Artificial Intelligence Act (TFAIA), which was signed in September 2025. The RAISE Act is set to take effect on January 1, 2027, making it the second U.S. state law to impose such regulations on frontier AI developers.
The law was enacted following negotiations between Hochul and the legislature, which modified the original bill to address concerns raised by the tech industry. Key changes included removing provisions that defined large developers based on compute costs and prohibiting deployment of models posing unreasonable risks. These revisions were formalized through chapter amendments (A9449/S8828) introduced in January 2026, though they remained in committee as of February 2026. Legal experts and the Governor’s office treat the agreed-upon amendments as the final form of the law.
Its passage coincides with federal debates over preemption, as President Donald Trump’s December 2025 executive order directed the Department of Justice to challenge state AI laws deemed conflicting with a potential federal framework. Legal challenges may arise over issues such as compelled speech, dormant Commerce Clause violations, or federal preemption.
Scope and Definitions
The RAISE Act applies to entities classified as large frontier developers, defined as companies with annual revenues exceeding $500 million that develop frontier models. A frontier model is specified as a foundation model trained using more than 10^26 floating-point operations (FLOPs). This threshold aligns with California’s TFAIA, ensuring consistency between the two state laws.
The original version of the bill, passed in June 2025, had defined large developers based on aggregate compute costs exceeding $100 million and included a prohibition on deploying models posing unreasonable risks of critical harm. These provisions were removed during negotiations with Governor Hochul, who sought to align the law more closely with California’s model. Exemptions are granted to accredited colleges and universities engaged in academic research, as well as the state’s Empire AI consortium.
The law’s definitions also include critical terms such as catastrophic risk and critical safety incidents. Catastrophic risk is defined as a foreseeable and material risk that a frontier model could contribute to the death or serious injury of more than 50 people, or property damage exceeding $1 billion. Loss of equity value is explicitly excluded from the definition of property damage.
Safety and Transparency Requirements
Large frontier developers under the RAISE Act are required to establish, implement, and publicly publish a frontier AI framework. This framework must detail how developers assess and mitigate catastrophic risks, secure unreleased model weights, engage third-party evaluators, govern internal use of frontier models, and respond to safety incidents. Unlike California’s TFAIA, which requires developers to describe their approach, the RAISE Act mandates that these measures be described in detail, including justifications for material modifications within 30 days of implementation.
Developers must also publish transparency reports before or concurrently with deploying new or substantially modified frontier models. These reports must include the model’s release date, supported languages and output modalities, intended uses, and any restrictions on use. Additionally, summaries of catastrophic risk assessments and the extent of third-party involvement must be disclosed. The framework and transparency reports must be reviewed at least annually to ensure ongoing compliance with safety standards.
The RAISE Act imposes stricter timelines for incident reporting compared to California’s law. Developers must report critical safety incidents within 72 hours, with a reduced window of 24 hours for incidents posing an imminent risk of death or serious physical injury. Critical safety incidents include unauthorized access to model weights resulting in harm, materialization of catastrophic risks, loss of control over a frontier model causing harm, and instances where a model uses deceptive techniques to subvert developer controls outside evaluation contexts, thereby increasing catastrophic risk.
Enforcement and Compliance
Enforcement of the RAISE Act is overseen by a newly established office within the New York State Department of Financial Services. This office is responsible for receiving incident reports, monitoring compliance, and publishing annual reports on AI safety beginning in 2028. Large frontier developers must file disclosure statements with this office and contribute pro rata assessments to fund its operations.
The New York Attorney General has the authority to bring civil actions against non-compliant developers. Penalties for violations start at $1 million for a first offense and increase to $3 million for subsequent violations. These penalties are lower than the original bill’s proposed $10 million and $30 million fines, reflecting the negotiated amendments. The law does not create a private right of action, meaning individuals cannot sue developers directly under its provisions.
The RAISE Act’s enforcement structure contrasts with California’s TFAIA, which routes reports through the Office of Emergency Services. New York’s dedicated office underscores its commitment to proactive oversight and accountability in AI safety. The law’s alignment with California’s framework facilitates coordination between the two states, though potential federal challenges may arise due to concerns over compelled speech, dormant Commerce Clause implications, or conflicts with emerging federal AI policies.
Legislative History and Industry Response
The RAISE Act’s legislative journey began with its introduction in March 2025 and culminated in its signing in December 2025 after a series of amendments. Governor Hochul initially proposed a near-complete rewrite of the bill to align it with California’s TFAIA, but legislators resisted significant changes, leading to a compromise that retained key provisions such as the 72-hour incident reporting timeline and the creation of a dedicated enforcement office.
Tech industry lobbying played a role in shaping the final version of the law. OpenAI and Anthropic expressed support for the RAISE Act, with Anthropic’s head of external affairs noting that the state laws could inspire federal action. However, the super PAC network Leading the Future, backed by Andreessen Horowitz and OpenAI president Greg Brockman, announced plans to challenge Assemblymember Bores in future elections, reflecting ongoing tensions between regulators and industry stakeholders.
Its alignment with California’s TFAIA sets a precedent for interstate coordination, though the potential for federal preemption remains a point of contention. The law’s enforcement mechanisms and reporting requirements signal a shift toward greater accountability in the development and deployment of frontier AI models.
Required actions
- Publish and implement a safety and security protocol addressing critical harms and unauthorised access to model weights - Report safety incidents to the state within 72 hours - Publish transparency information before deploying frontier models; do not deploy models that create unreasonable risk of critical harm - A new oversight office within the Department of Financial Services receives reports