The New York RAISE Act is a state law that makes the companies training the most powerful AI models publish safety plans, post transparency reports for new models and report serious safety incidents to the state within 72 hours. It takes effect on January 1, 2027. The heaviest duties fall on "large frontier developers," meaning companies that train models above 10^26 operations and had more than $500 million in revenue the year before. On November 10, 2026, New York opened the online RAISE Portal where those companies must register before the law starts.

Below: what the final text requires, how it changed in 2026, how it compares to California's SB 53 and what is proposed next. Facts come from the bill text, the Department of Financial Services (DFS) and the Governor's office, checked against legal and news coverage.

New York RAISE Act at a glance

QuestionShort answer
Full nameResponsible AI Safety and Education (RAISE) Act, Article 44-B of the General Business Law
Original billS6953B / A6453B, passed both chambers June 12, 2025
SignedDecember 19, 2025 (Chapter 699)
Final amendmentS8828 / A9449, signed March 27, 2026 (Chapter 96)
Effective dateJanuary 1, 2027
RegulatorOffice of Digital Innovation, Governance, Integrity, and Trust (DIGIT), inside DFS
EnforcerNew York Attorney General (civil penalties); DIGIT (registration fines)
Who is coveredDevelopers of models trained above 10^26 operations; extra rules above $500 million in revenue
Incident deadline72 hours, or 24 hours if there is imminent risk of death or serious injury
Top penaltiesUp to $1 million for a first violation, up to $3 million for each later one
Private lawsuitsNot allowed under the Act

The bill history comes from the New York State Assembly record for S6953B and the record for S8828. The rules come from the amended bill text (A9449) and the DFS page Information for Frontier Developers.

Who the RAISE Act covers

The law sorts companies with two tests that match California's.

  • Frontier model: a foundation model trained with more than 10^26 integer or floating point operations. The count includes the original training run plus later fine tuning, reinforcement learning or other material changes.
  • Frontier developer: anyone who has trained, or started training, a model at that level.
  • Large frontier developer: a frontier developer that, together with its affiliates, had annual gross revenue above $500 million in the previous calendar year.

The Act applies only to frontier models "developed, deployed, or operating in whole or in part in New York state." It does not apply to accredited New York colleges and universities doing academic AI research, or to the state's Empire AI consortium.

No official list says which labs cross the compute line. TechCrunch reported that companies "like Anthropic and OpenAI" will have to register. The law says DIGIT will publish the list of large frontier developers that file, without their contact details. For background on the compute test itself, see what is a frontier model.

What every frontier developer must do

These duties apply to any company above the compute line, whatever its revenue.

Transparency reports. Before or at the same time as deploying a new or substantially modified frontier model, the developer must publish a report on its website. It lists the website, a way for a person to contact the company, the release date, supported languages, output modalities, intended uses and any general restrictions on use. Publishing this inside a system card or model card counts.

Incident reports within 72 hours. A "critical safety incident" means stolen or tampered model weights that cause death or injury, harm from a catastrophic risk, loss of control causing death or injury, or a model deceiving its developer to get around controls outside a test, in a way that shows materially higher risk. The clock starts when the developer decides an incident happened or reasonably believes one did. If there is imminent risk of death or serious injury, it must also tell an authority such as police within 24 hours.

No misleading statements. A developer may not make materially false or misleading statements about catastrophic risk from its models. Good faith statements that were reasonable at the time are protected.

DFS says the public will also be able to report incidents once the law is in effect.

Extra rules for large frontier developers

Large frontier developers carry the main load.

  1. Publish a frontier AI framework. This written safety plan must explain the company's capability thresholds for catastrophic risk, its mitigations and predeployment reviews, its use of outside assessors, security for unreleased model weights, incident response, internal governance and risks from internal use, including a model getting around oversight.
  2. Review it every year. A material change must be published with a justification within 30 days.
  3. Add risk summaries to transparency reports. Reports must summarize the catastrophic risk assessments, their results, how much third party evaluators were involved and other steps taken under the framework.
  4. Send quarterly summaries. Every three months the company must give DIGIT a confidential summary of catastrophic risk from internal use of its models. DFS says these are due within 30 days of the end of each calendar quarter.
  5. Register and pay. From January 1, 2027, no large frontier developer may develop, deploy or operate a frontier model in New York without a current disclosure statement on file and its share of the state's costs paid.

These frameworks resemble the voluntary policies labs already publish, compared in our Preparedness Framework vs Responsible Scaling Policy guide. The difference: under the RAISE Act, not following your own framework becomes a violation.

How RAISE Act registration works

The RAISE Portal opened on November 10, 2026, the same day Governor Kathy Hochul launched the New York Center for AI Responsibility and Research at Binghamton University, according to the Governor's announcement. Large frontier developers must register by January 1, 2027.

The steps DFS describes:

  1. Create a free NY.gov ID account.
  2. Request account authorization by giving the company's legal name, country, street address and the details of the person filing.
  3. After DIGIT approves the account, file the disclosure statement: principal place of business, every New York office, beneficial ownership and three named contacts (primary, secondary and tertiary).

Private companies list owners of 5% or more (and former 5% owners from the past five years); public companies list holders of 50% or more. Statements are renewed every two years, or sooner after an ownership transfer or material change.

Missing the filing has its own penalty. After notice and a hearing, DIGIT can charge $1,000 for each day a company fails to file or fails to correct false information, plus any assessments owed.

RAISE Act penalties and enforcement

The New York Attorney General can sue a large frontier developer for up to $1 million for a first violation and up to $3 million for each later violation, scaled to severity. Covered violations are failing to publish or send a required document, making a false statement about its framework, failing to report an incident, or not following its own framework. The Act creates no private right of action, so individuals cannot sue under it.

These caps are a tenth of the original. The June 2025 version allowed up to $10 million and $30 million, as Cooley and the IAPP both note.

How the RAISE Act changed from 2025 to 2026

The legislature passed the original bill on June 12, 2025, and Governor Hochul signed it on December 19, 2025 with agreed changes to follow. Those came as S8828, signed March 27, 2026, which replaced the 2025 article.

According to Morrison Foerster and Cooley, the amendment did three main things: it swapped the earlier definition of a large developer for California's $500 million revenue test, it cut penalties to $1 million and $3 million, and it pushed the start date to January 1, 2027. The 72 hour incident window survived.

New York RAISE Act vs California SB 53

California's Transparency in Frontier Artificial Intelligence Act, SB 53, was signed on September 29, 2025 and took effect in January 2026. New York copied its definitions, then went further in places.

FeatureNew York RAISE ActCalifornia SB 53
Compute and revenue tests10^26 operations; $500 million10^26 operations; $500 million
Incident reporting window72 hours15 days
Penalty caps$1 million first, $3 million after$1 million per violation
Dedicated regulatorDIGIT inside DFS, funded by developer assessmentsNo new office; incidents go to the Office of Emergency Services
Registration and feesYesNo
Quarterly internal use summariesYesNot in this form
Whistleblower protectionsNot in the ActYes
EffectiveJanuary 1, 2027January 2026

The California comparison draws on Wiley, Cooley and TechCrunch.

RAISE Act 2.0: what is being proposed next

On October 28, 2026, Senator Andrew Gounardes, one of the original sponsors, filed S10731, with Senator Patricia Fahy as cosponsor. The Assembly record shows it was referred to the Senate Rules Committee and nothing more. Reporting by StateScoop and the Transparency Coalition says it would:

  • cover developers that spent more than $1 billion on frontier work over three years, instead of using the revenue test
  • require independent, state licensed examiners to audit large developers' safety practices
  • let DIGIT limit the compute large developers use for training and automated AI research to reduce catastrophic risk
  • let the Attorney General seek court orders, including limits on deploying a model in New York

None of that is law. The bill is aimed at the 2027 legislative session, and the RAISE Act that starts on January 1 is the March 2026 version described above.

Pros and cons of the RAISE Act

Pros

  • Turns voluntary safety frameworks into published, enforceable commitments.
  • The 72 hour window is the fastest frontier incident rule in the US.
  • A dedicated office funded by the developers themselves, with an annual public report from January 1, 2028.

Cons

  • Most duties rest on documents the labs write themselves; the Act sets no safety standard a model must meet.
  • Incident reports and quarterly summaries are exempt from New York's public records law, so the public sees only aggregated data.
  • No whistleblower protections, unlike California.
  • Its future depends partly on federal policy. Executive Order 14365 set up a Justice Department task force to challenge state AI laws, and the White House National AI Policy Framework asks Congress to preempt them.

What it means if you use ChatGPT, Claude, Gemini or Grok

The RAISE Act puts no duties on people who use AI. It regulates the companies behind assistants such as ChatGPT, Claude, Gemini and Grok, if they meet the tests. From 2027 users get more published safety information with each new model and a state office to report a serious incident to. Researchers at groups such as METR and the Center for AI Safety gain new documents to examine. For a far stricter federal proposal, see the Ban Artificial Superintelligence Act, and for basics, what AI safety means.

FAQ

When does the New York RAISE Act take effect?

On January 1, 2027. The March 2026 amendment changed the start date from 90 days after signing to that fixed date. Large frontier developers must register through the RAISE Portal by the same day.

Does the RAISE Act apply to small AI startups?

Only if they trained a model above 10^26 operations, which is a very high compute bar. A frontier developer with less than $500 million in revenue must still post transparency reports and report incidents, but it does not need a safety framework, quarterly summaries or registration.

What counts as a catastrophic risk under the RAISE Act?

A foreseeable, material risk that a frontier model contributes to more than 50 deaths or serious injuries, or over $1 billion in property damage, in one incident, by giving expert help with chemical, biological, radiological or nuclear weapons, acting without meaningful human oversight in a cyberattack or serious crime, or evading control.

Can I sue an AI company under the RAISE Act?

No. The Act creates no private right of action. The Attorney General seeks the main penalties and DIGIT charges the daily registration fine. The public can report incidents to DIGIT.

Is the RAISE Act the same as California SB 53?

They share definitions, but New York is stricter on timing and penalties: 72 hours to report instead of 15 days, and up to $3 million for repeat violations instead of $1 million. New York also adds a regulator, registration, fees and quarterly internal use summaries, while California has whistleblower protections that New York's Act lacks.