Skip to content
Komposite News
Markets · Technology · Capital
Thursday, September 10, 2026Global EditionGot a tip?
AI Safety

New York opens its register for frontier AI developers

Developers of the largest AI models have 52 days to register with New York before the RAISE Act takes effect on 1 January. Our side-by-side with California's law shows where New York is stricter: a 72-hour clock and a $3m penalty.

Jonathan Bright · House byline
November 12, 20264 min read
The New York State Capitol in Albany, a granite building with red-roofed towers, seen across a park with an equestrian statueShaunfrombrooklyn, CC BY-SA 4.0, via Wikimedia Commons
The New York State Capitol in Albany, seen across East Capitol Park, in a file photograph from 2017.Photograph via Wikimedia Commons

New York opened a registration portal on Tuesday for the largest developers of frontier artificial-intelligence models, which must sign up by 1 January 2027, when the state's Responsible AI Safety and Education (RAISE) Act takes effect, TechCrunch reported. OpenAI and Anthropic are among the companies expected to register.

The law applies to developers with more than $500m of annual revenue whose models were trained with more than 1026 computing operations. From 1 January they must publish a safety framework, report a critical safety incident to the state within 72 hours, and pay fines of up to $1m for a first violation and $3m for later ones. Registration opened 52 days before the deadline, by our count.

The registrations go to the Office of Digital Innovation, Governance, Integrity and Trust, or DIGIT, a unit the state placed inside the Department of Financial Services, the regulator of New York's banks and insurers. DFS said in September that registration would start in November and named Marc Gilman, DIGIT's first full-time hire, as deputy director for the act. "DFS is proud to lead the implementation of the RAISE Act for New York," Kaitlin Asrow, the acting superintendent, said then. On Tuesday the state also launched the New York Center for AI Responsibility and Research at SUNY Binghamton, which TechCrunch said will evaluate advanced systems and help DIGIT review the companies' frameworks.

Who has to register?

Companies with more than $500m of revenue in the previous calendar year that train or operate a frontier model. The threshold counts the compute used in original training plus later fine-tuning, according to an analysis of the amended act by the law firm Davis Polk. A large developer files a disclosure statement with DFS before developing, deploying or operating such a model, renews it every two years and pays a share of the office's running costs. The revenue line is not close for the best-known labs. Anthropic's listing documents showed first-half revenue of $16.3bn.

A state spokesperson declined to say what the state would do to a covered company that did not register, according to TechCrunch. The act gives DFS its own penalty of $1,000 a day for failures to file disclosures or pay fees, separate from the Attorney General's power to seek the larger fines, and there is no private right of action.

How does it compare with California's law?

New York copied California's thresholds and went further on speed and penalties. California's Transparency in Frontier Artificial Intelligence Act, SB 53, signed by Governor Gavin Newsom in September 2025, uses the same compute and revenue tests. Its incident clock runs for 15 days. New York's runs for 72 hours, five times faster by our arithmetic, and 24 hours where there is an imminent risk of death or serious injury. New York's cap for a repeat violation is three times California's.

Komposite's side-by-side from Davis Polk's analysis of the amended act, the Department of Financial Services and the California governor's office.
New York RAISE Act (as amended)California SB 53
Signed19 Dec 2025; amended version 27 Mar 202629 Sep 2025
Frontier modelTrained with more than 1026 operations, counting fine-tuningSame compute threshold
Large developerMore than $500m of revenue in the prior yearSame revenue threshold
Critical incident report72 hours; 24 hours if there is an imminent risk of death or serious injury15 days
Who receives itDIGIT, a new office in the Department of Financial ServicesCalifornia's Office of Emergency Services
Maximum penalty$1m for a first violation, $3m after that (Attorney General); $1,000 a day for disclosure or fee failures (DFS)Up to $1m a violation (Attorney General)
Registration and feesDisclosure statement before operating, renewed every two years; share of the office's costsNone
Whistleblower protectionsNot includedIncluded
RulemakingDFS may write rules after notice and commentNot comparable

The comparison runs the other way on whistleblowers. California's law protects employees who disclose significant health and safety risks; New York's amended act does not, the Davis Polk analysis notes. California also has no registration step. Its law sends incident reports to the Office of Emergency Services and leaves enforcement to the Attorney General.

Why did it take a year?

Because the law was signed before it was finished. Governor Kathy Hochul signed the original bill on 19 December 2025 on condition that the legislature amend it, and signed the amended version 98 days later. The register opened 228 days after that.

Komposite's day count from the dates in the sources linked in this article.
DateStepDays
Fri 19 Dec 2025Governor Hochul signs the RAISE Act, on condition that it is amended0
Fri 27 Mar 2026Amended act signed98
Mon 21 Sep 2026DFS names DIGIT to run it and says registration starts in November276
Tue 10 Nov 2026Registration portal opens326
Fri 1 Jan 2027Act takes effect; registration deadline378

Hochul has presented the act as a check on the industry rather than a brake on it. "Tech moguls tell us that their vision for AI is inevitable," she said at a press briefing on Tuesday, TechCrunch reported. The state has gone further than New York City, whose council in October heard a proposal for a validator for every model, and it already has an operating office, while Australia said this week only that it plans a frontier AI regulator.

The next date is 1 January, when the act takes effect and the first disclosure statements are due. DFS can also write rules under the act, after notice and comment, which would fill in what the statute and the portal leave open.

MORE ON THIS STORY