New York's AI Safety Law Starts Jan. 1, and Hochul Wants Congress to Copy It. Does the RAISE Act Apply to Your Startup?

NEW YORK — In seven weeks, New York’s law on the most powerful AI models takes effect. On Tuesday, Gov. Kathy Hochul said she wants it to go national.

“When they start looking for ways to regulate AI, the first place they should look is New York,” Hochul said at a Midtown press conference, Spectrum News NY1 reported. She said the incoming Democratic-controlled Congress should adopt New York’s Responsible AI Safety and Education Act, known as the RAISE Act, which starts Jan. 1, 2027. Her administration said at least two companies have agreed to sign on, but declined to name them, according to Spectrum.

For the thousands of New York startups that build products with AI, the practical question is simpler: does this law apply to me? For most, the answer is no. But the details are worth knowing.

How the Law Got Here

Hochul first signed the RAISE Act in December 2025, then negotiated a set of changes with lawmakers. That amended version was signed on March 27, 2026, according to a summary by the law firm Wiley. The changes brought New York’s definitions closer to California’s frontier AI transparency law, though Wiley notes New York kept some stricter terms, including a 72-hour deadline to report safety incidents, compared with 15 days in California.

The law is overseen by a new office inside the state Department of Financial Services (DFS), which has posted guidance for frontier developers and opened a reporting portal.

Who’s Covered

The law turns on three definitions, as summarized by DFS:

• Frontier model: a foundation model trained with more than 10^26 integer or floating-point operations of computing power, counting the original training run plus fine-tuning, reinforcement learning or other material changes.
• Frontier developer: anyone who has trained, or started training, a frontier model.
• Large frontier developer: a frontier developer that, with its affiliates, had more than $500 million in annual gross revenue the previous year.

That compute threshold is enormous, the scale of the largest models built by the biggest AI labs. A New York startup that calls a model through an API, fine-tunes an open-source model, or builds an app on top of someone else’s model is not, by these definitions, training a frontier model of its own.

What Covered Companies Must Do

According to DFS, all frontier developers must:

• Publish a transparency report before or when they deploy a new or substantially modified frontier model, covering things like release date, supported languages and output types, intended uses and restrictions.
• Report qualifying “critical safety incidents” to DFS within 72 hours, and within 24 hours to law enforcement or a public safety agency if there’s an imminent risk of death or serious injury.

Large frontier developers must also register with DFS starting Jan. 1, publish and update a frontier AI safety framework at least once a year, and send DFS quarterly summaries of catastrophic-risk assessments tied to internal use of their models.

Enforcement runs through the state attorney general, with civil penalties of up to $1 million for a first violation and up to $3 million for later ones, according to Wiley. The law does not create a private right of action.

What It Means for New York Startups

Most of you are out of scope, but your vendors may be in it. If your product runs on a frontier model from a major lab, that lab is the one with reporting duties. Its published transparency reports and safety frameworks may become useful reading when you evaluate vendors or answer customer questions about the models you use.

Watch your own AI rules, not just this one. The RAISE Act is narrow. Other state rules are more likely to touch everyday products, such as New York’s new law on AI chatbots and minors, which we covered in our founder’s guide.

If you’re training very large models, prepare now. DFS “strongly encouraged” all frontier developers, not just large ones, to set up a RAISE Portal account in advance so they can report incidents quickly. Accounts require an NY.gov ID.

Expect the debate to grow. Wiley notes the law sits in tension with the Trump administration’s push for one federal standard instead of a patchwork of state laws. Hochul is now pushing the opposite approach, a federal law modeled on New York’s. If Congress takes up her suggestion, the definitions here could become the national baseline.

Keep an eye on rulemaking. The new DFS office can issue regulations, including added reporting requirements, according to Wiley. Founders in AI should follow DFS announcements in early 2027 to see whether the scope changes.

The Bottom Line

The RAISE Act is aimed at the handful of companies training the biggest AI models, not at the startups building on top of them. But with Jan. 1 approaching and the governor pitching it to Washington, New York founders building with AI should understand where the lines are, and which of their suppliers sit on the other side of them.