New York Becomes the First State to Mandate Frontier AI Safety Reporting

New York has moved ahead of Washington on frontier AI regulation. Governor Kathy Hochul signed the Responsible AI Safety and Education Act — the RAISE Act — on March 27, 2026, making New York the first US state to impose mandatory transparency and safety-reporting obligations on developers of the most powerful AI models.

The law takes effect January 1, 2027, giving affected companies roughly nine months to prepare.

Who It Covers

The RAISE Act is narrow by design. It targets developers with annual revenues above $500 million that train frontier models requiring more than 10²? floating-point operations at a compute cost above $100 million. That definition captures a small number of companies — OpenAI, Google DeepMind, Anthropic, Meta, and perhaps a handful of others — while leaving the vast majority of the AI industry untouched.

Covered developers must publish a "frontier AI framework" describing their development and safety practices, file transparency reports before each new model deployment, and report safety incidents to regulators within 72 hours of discovery. Repeat violations carry civil penalties of up to $3 million.

A New Office Within DFS

Oversight sits inside New York's Department of Financial Services, which already regulates banks, insurers, and fintech companies operating in the state. The choice of DFS is somewhat unusual — AI development is not a financial activity — but the agency has established infrastructure for handling sensitive disclosures and has experience applying systemic-risk thinking to large institutions.

Whether DFS has the technical capacity to meaningfully evaluate frontier AI safety reports is a fair question. The 72-hour incident-reporting window, in particular, assumes regulators can assess what they receive in a timely way. Critics of the law have noted the gap.

Context: Federal Action Has Been Limited

The Biden administration's 2023 AI Executive Order required safety testing and government reporting for certain frontier models, but subsequent federal action has been slow. The current White House AI policy framework — published earlier this year — emphasizes voluntary commitments and sector-specific guidance rather than statutory requirements.

That gap has pushed states to act on their own. New York's RAISE Act is among the most substantive state-level interventions so far, though it covers a narrower slice of the industry than earlier drafts proposed. A full analysis of the law's requirements is available from Davis Wright Tremaine.