TL;DR
New York's RAISE Act formally launched registration for large frontier AI developers on September 21, making New York the first major U.S. state to impose mandatory safety reporting, risk assessments, and civil penalties on AI companies while the federal government accelerates development.
What happened
- Gov. Kathy Hochul, AG Letitia James, and legislators gathered in New York City on Sept. 21 to announce the RAISE Act's implementation timeline and appoint Marc Gilman as deputy director for the act.
- Registration for large frontier AI developers begins in November 2026, ahead of full enforcement starting January 1, 2027.
- As of Jan. 1, 2027, covered developers must: post safety and transparency frameworks publicly, report critical safety incidents to the state's DIGIT Office within 72 hours, and file quarterly catastrophic-risk assessments.
- Penalties: first violations up to $1 million, subsequent violations up to $3 million, enforced by the AG's office through civil action.
- Hochul signaled she may pursue AI kill-switch legislation if deemed feasible, and Sen. Gounardes introduced two additional bills requiring third-party audits and stronger safety frameworks.
Why it matters
- New York is moving in direct opposition to the Trump administration, which announced plans for an "AI Force" and a new AI czar to accelerate federal AI development the same weekend.
- The RAISE Act creates a state-level compliance layer that frontier AI developers, many headquartered or operating in New York, cannot ignore regardless of federal posture.
- Whistleblower infrastructure is live: AG James is actively soliciting reports of unsafe or illegal AI conduct through a secure portal, raising litigation risk for developers operating in the state.
- A one-year moratorium on hyperscale data centers (50-plus megawatts) already in place compounds pressure on AI infrastructure investment in New York.
- Gounardes framed the state's role explicitly as a national backstop: "until the federal government acts, we bear responsibility to protect all Americans."
What to watch next
- November 2026 registration deadline: which frontier AI developers comply, which resist, and whether the DIGIT Office has capacity to enforce.
- Kill-switch legislation: Hochul's exploratory language is a leading indicator of how far state-level AI governance could go in 2027 legislative sessions.
- Federal-state conflict: whether the Trump administration's AI expansion agenda triggers preemption battles with New York and other states pursuing independent regulatory frameworks.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.