New York opens RAISE Act registration for large AI developers

New York State opened an online registration portal for large frontier AI developers on November 10. Companies that develop, deploy or operate covered models in the state have until January 1, 2027 to register if they meet the law’s definition of a large developer, the governor’s office said. That is the day New York’s RAISE Act takes effect. DIGIT, a unit of the state Department of Financial Services, handles registration. New York had set out the November start in September. It has now opened the portal and released the filing instructions.

New York uses both computing and revenue thresholds. A frontier model is a foundation model trained with more than 10²⁶ integer or floating-point operations. The count includes the original training run and work the developer subsequently does to fine-tune the model, train it with reinforcement learning or materially modify it in other ways. Developers that have used that much computing or begun training with plans to do so count as large if they and their affiliates had more than $500 million in combined gross revenue in the previous calendar year. The law covers models developed, deployed or operating wholly or partly in New York.

What goes to the state and what goes online

The portal requires an NY.gov ID. State guidance says the person submitting information must first establish that they can act for the company. Once the account is authorized, the developer files a disclosure with its main business address, any New York offices, applicable beneficial ownership information and three points of contact. Once registration is approved, the same account handles incident reports, quarterly risk summaries and assessment payments.

The filing comes up for renewal every two years, or sooner if the model changes hands or previously reported information materially changes. The law directs DIGIT to publish a list of large developers that have filed. Contact details for the named representatives are withheld from that list.

Large developers must also post an AI safety framework on their own websites. It explains how they identify and reduce risks and respond to serious incidents. They have to review it at least annually and publish any material revision, with their reasons, within 30 days.

Developers of any size covered by the law must post a transparency report before or when they deploy a new or substantially modified frontier model. It covers the release date, supported languages, output types and use restrictions. Large developers also summarize their catastrophic-risk assessments, what they found and the extent of outside evaluators’ involvement. A model card can carry that information if it covers the statutory requirements.

A 72-hour deadline for critical incidents

Once the law takes effect, frontier developers have 72 hours to notify DIGIT from the point at which they establish a critical safety incident or have reasonable grounds to believe one occurred. If they discover an imminent risk of death or serious physical injury, they must notify an appropriate authority within 24 hours. The public can report suspected critical safety incidents starting January 1, the state says.

Large developers separately send DIGIT confidential summaries of assessments of catastrophic risk arising from their internal use of frontier models. The statute calls for a summary every three months and allows a different schedule agreed with the agency in writing. The portal guidance puts the default filing deadline at 30 days after the end of the calendar quarter.

A research network based at Binghamton

Governor Kathy Hochul also announced a statewide safety research effort based at the New York Center for AI Responsibility and Research at Binghamton University. The center, first announced in January, is to draw more than 80 scientists from New York institutions to test advanced systems independently. Cornell Tech’s Greg Morrisett and Columbia University’s Jeannette Wing will serve as interim co-directors.

Researchers will remain at their home institutions. The center will use Empire AI’s shared computing infrastructure for universities and research institutions. Binghamton University says the consortium’s initial computing infrastructure is housed at the University at Buffalo. Alongside testing current models, the center plans to develop measurement methods and standards for evaluating future systems.

The governor’s office said the center can provide technical assessments at the state’s request. One task it identified is working with DIGIT to review developers’ published safety frameworks and propose ways to strengthen the requirements.