What does the New York RAISE Act require, and when does it take effect?

On this page

1. What does the New York RAISE Act require?
2. When does the RAISE Act take effect, and how did it get here?
3. Who does the RAISE Act cover?
4. What must a large frontier developer publish?
5. What must every frontier developer publish?
6. What must be reported to the state, and how fast?
7. What is the disclosure statement, and what does it cost?
8. What are the penalties?
9. How does the RAISE Act differ from California’s SB 53?
10. Can the federal government block it?

What does the New York RAISE Act require?

Starting January 1, 2027, New York’s Responsible AI Safety and Education (RAISE) Act requires developers of the largest AI models to publish a transparency report for each new frontier model and to report critical safety incidents to the state within 72 hours. Developers with more than $500 million in annual revenue must also publish how they manage catastrophic risk and keep a disclosure statement on file with a new office in the Department of Financial Services. The law does not require a license, an audit or state approval of any model.

The law applies only to frontier models “developed, deployed, or operating in whole or in part in New York state” (section 1425). Most of its text follows California’s SB 53, which has been in force since January 1, 2026. The main differences are a shorter incident deadline, a registration-style disclosure statement with a fee, a regulator with rulemaking power, and higher penalties for repeat violations.

Quotations here are from the final 2026 text. The frontier AI laws guide compares New York with the EU, California, China and South Korea.

When does the RAISE Act take effect, and how did it get here?

January 1, 2027. The date is set by section 3 of the 2025 chapter as amended in 2026. The original bill would have taken effect 90 days after becoming law; the chapter amendment replaced that with the fixed date.

• June 12, 2025: the Senate passed S. 6953-B and the Assembly substituted it for A. 6453-B and passed it the same day.
• December 19, 2025: Governor Kathy Hochul signed it as Chapter 699 of 2025, with an approval memorandum. Her office announced the signing together with the agreed chapter amendments that would rewrite the law in the next session.
• January 8, 2026: Senator Andrew Gounardes introduced the chapter amendment, S. 8828 (same as A. 9449). The Senate passed it January 28, 2026 and the Assembly on March 11, 2026.
• March 27, 2026: the Governor signed S. 8828 as Chapter 96 of 2026. It repealed the 2025 version of Article 44-B and enacted a new one, and moved the effective date to January 1, 2027.

The 2026 text is the one that counts. Summaries of the 2025 chapter describe a different statute, with a $100 million compute-cost test, a deployment ban and penalties of up to $10 million and $30 million. Some also describe a third-party audit requirement and employee protections; those were in earlier drafts of the bill and were removed before it passed in June 2025. None of it is in the law that takes effect.

Who does the RAISE Act cover?

Section 1420 sets three nested definitions. Which one a company falls under decides which duties apply.

• Frontier model: “a foundation model that was trained using a quantity of computing power greater than 10²⁶ integer or floating-point operations.” The count includes the original training run “and for any subsequent fine-tuning, reinforcement learning, or other material modifications the developer applies to a preceding foundation model.” A foundation model is one trained on a broad data set, designed for generality of output and adaptable to a wide range of tasks.
• Frontier developer: a person who has trained, or initiated the training of, a frontier model, “with respect to which the person has used, or intends to use, at least as much computing power to train the frontier model as would meet” the 10²⁶ threshold.
• Large frontier developer: a frontier developer that “together with its affiliates collectively had annual gross revenues in excess of five hundred million dollars in the preceding calendar year.”

The compute count includes post-training. Fine-tuning, reinforcement learning and other material modifications are added to the original training run, so post-training can carry a model over the 10²⁶ line. Whether a company that fine-tunes someone else’s model becomes a frontier developer is less clear, because section 1420(8) asks whether that company itself used, or intends to use, computing power at the threshold. Cooley advises companies that fine-tune, retrain or otherwise materially modify frontier models to assess whether that work brings them within the law. The compute calculator estimates where a training run lands against this and the other thresholds.

Section 1426 excludes accredited New York colleges and universities doing academic research on AI models, and the Empire AI consortium. Section 1425 limits the whole article to models developed, deployed or operating at least in part in New York.

Does the RAISE Act apply to me?

1. Is the model a foundation model trained with more than 10²⁶ operations, counting fine-tuning and other material modifications? If not, the law does not apply.
2. Is it developed, deployed or operating at least in part in New York? If not, the law does not apply.
3. Did you, with your affiliates, have more than $500 million in gross revenue in the preceding calendar year? If yes, every duty below applies. If no, the duties are the transparency report, the incident reports and the ban on false statements about catastrophic risk, plus the redaction rules in section 1421(5) if you redact a published document.

What must a large frontier developer publish?

A frontier AI framework. Section 1421(1) requires a large frontier developer to “write, implement, comply with, and clearly and conspicuously publish” on its website a framework that describes in detail how it handles ten things, including:

• incorporating national and international standards and industry best practices;
• defining and assessing the thresholds it uses to decide whether a model could pose a catastrophic risk;
• applying mitigations based on those assessments, and reviewing them before deploying a model or using it extensively internally;
• using third parties to assess catastrophic risk and the effectiveness of mitigations;
• cybersecurity practices that secure unreleased model weights;
• identifying and responding to critical safety incidents;
• internal governance, and the management of catastrophic risk from internal use, “including risks resulting from a frontier model circumventing oversight mechanisms.”

The framework must be reviewed, and updated as appropriate, at least once a year. A material modification must be published with a justification within 30 days (section 1421(2)).

“Catastrophic risk” is defined in section 1420(3) as a foreseeable and material risk that development, storage, use or deployment of a frontier model will materially contribute to the death of, or serious injury to, more than 50 people, or more than $1 billion in property damage or loss, from a single incident. The incident has to involve the model providing expert-level help creating or releasing a chemical, biological, radiological or nuclear weapon; carrying out a cyberattack, or conduct that would be murder, assault, extortion or theft, with no meaningful human oversight; or evading the control of its developer or user. Risk from information already publicly accessible elsewhere, lawful federal activity, and harm the model did not materially contribute to are excluded. Section 1423 adds that a loss of equity value does not count as property damage.

What must every frontier developer publish?

Every frontier developer, large or not, must publish a transparency report for each model. Under section 1421(3), before or at the same time as deploying a new frontier model or a substantially modified one, a frontier developer must publish on its website the developer’s website address, a way for a natural person to contact it, the model’s release date, supported languages, output modalities, intended uses and any generally applicable restrictions or conditions on use.

A large frontier developer must add summaries of its catastrophic-risk assessments for the model, the results, the extent of third-party evaluator involvement and the other steps taken under its framework. Publishing the same information inside a system card or model card “shall be deemed in compliance” (section 1421(3)(c)).

Section 1421 also governs what a developer may say and what it may withhold. Section 1421(4) bars a frontier developer from making a materially false or misleading statement about catastrophic risk from its models or its management of that risk, and bars a large frontier developer from making one about its compliance with its framework; a statement made in good faith and reasonable under the circumstances is excused. Section 1421(5) allows redactions to protect trade secrets, cybersecurity, public safety or national security, or to comply with law, but the developer must describe the character and justification of each redaction and keep the unredacted version for five years.

What must be reported to the state, and how fast?

Section 1422 sets three reporting clocks. Two run to the new office in the Department of Financial Services. The imminent-risk disclosure goes to whichever authority fits the incident.

• 72 hours for critical safety incidents. A frontier developer must report a critical safety incident involving one of its frontier models “within seventy-two hours from a determination that a critical safety incident has occurred or within seventy-two hours of the frontier developer learning facts sufficient to establish a reasonable belief that a critical safety incident has occurred.” An amended report may follow if more is learned.
• 24 hours when the risk is imminent. If the incident “poses an imminent risk of death or serious physical injury,” the developer must disclose it within 24 hours to an appropriate authority, including a law enforcement or public safety agency with jurisdiction. The 72-hour rule in section 1422(3)(a) is “subject to” the imminent-risk paragraph, and the text does not say whether the 24-hour disclosure replaces the 72-hour report to the office or is added to it.
• Every three months for internal-use risk. A large frontier developer must send the office a summary of any assessment of catastrophic risk from internal use of its frontier models every three months, or on another reasonable schedule it proposes in writing and the office agrees to.

A “critical safety incident” (section 1420(4)) is unauthorized access to, modification of or exfiltration of model weights that results in death or bodily injury; harm from a catastrophic risk materializing; loss of control of a model causing death or bodily injury; or a model using deceptive techniques against its developer to subvert controls or monitoring, outside an evaluation designed to elicit that behavior, in a way that shows materially increased catastrophic risk.

The office must build a reporting mechanism that members of the public can also use. Reports and internal-use summaries are exempt from disclosure under New York’s Freedom of Information Law, and the office may pass them to other government bodies, including the Attorney General. Starting January 1, 2028, the office must publish an annual report with anonymized, aggregated incident information and any recommended changes to the law.

There is a federal off-ramp for incident reporting only. The office may designate a federal law, regulation or guidance document whose incident-reporting standard is substantially equivalent or stricter; a developer that declares its intent to comply that way is deemed in compliance with the 72-hour rule, provided it sends the office copies of its federal reports at the same time (section 1422(8) and (9)). The Governor announced the office, called DIGIT, on September 21, 2026. As of November 2, 2026 we found no designation.

What is the disclosure statement, and what does it cost?

Section 1428 is the provision with no counterpart in California. “[N]o large frontier developer may develop, deploy, or operate a frontier model, in whole or in part in New York state, without having a current disclosure statement filed with the office and paying the required share.”

The statement must identify the developer and every name it does business under, the address of its principal place of business and of each New York office, and three named points of contact. A privately held developer must list every person or entity that beneficially owns 5% or more, and anyone who did in the preceding five years; a publicly traded one lists owners of 50% or more. It must be renewed every two years, or sooner when ownership of the model is transferred or the reported information materially changes.

The cost is not a fixed fee. Large frontier developers are “assessed in pro rata shares” by the department to cover all direct and indirect costs of administering the article. The office will publish the list of developers that have filed, without their contact details.

Because a large frontier developer may not operate a frontier model in New York without a current statement on file, the provision works like a registration. The office sets the form and the required contents of the statement (section 1428(2)), but the statute gives it no express power to refuse a filing or to approve a model, which is why this site does not count it as a license. The distinction is explained in AI licensing explained.

What are the penalties?

The Attorney General and the new DFS office enforce different parts of the law, on different scales.

• Attorney General (section 1427): a civil action for up to $1 million for a first violation and up to $3 million per subsequent violation, set by severity, where a large frontier developer fails to publish or transmit a required document, makes a statement that breaks section 1421(4), fails to report an incident, or fails to comply with its own frontier AI framework.
• The DFS office (section 1428(5)): after notice and a hearing, $1,000 per day for operating without a current disclosure statement or failing to correct false information in one, plus any unpaid assessment.

The Attorney General’s penalty sentence in section 1427 names large frontier developers and no other kind of developer. A frontier developer below the $500 million line carries the transparency-report and 72-hour duties, but as written the $1 million and $3 million penalties attach to a large frontier developer’s failures. Section 1429 gives the office a general power to adopt rules implementing the article and lets it consider additional reporting or publication requirements; it does not mention penalties. Section 1424 leaves duties and remedies under other law in place. There is no private right of action (section 1427(2)), and the law does not stop a developer from arguing that someone else caused a harm.

How does the RAISE Act differ from California’s SB 53?

The two laws share definitions, the framework, the transparency report, the 24-hour rule for imminent risk and the catastrophic-risk concept. They part ways on timing, registration, enforcement structure and scope.

New York RAISE Act and California SB 53 compared, as of November 2026

Feature	New York RAISE Act	California SB 53
In force	January 1, 2027	January 1, 2026
Frontier model	Foundation model, >10²⁶ operations including post-training	Same
Large developer	>$500M gross revenue with affiliates, preceding calendar year	Same
Critical safety incident deadline	72 hours to the DFS office; 24 hours to an appropriate authority if imminent risk	15 days to the Office of Emergency Services; 24 hours if imminent risk
Registration	Disclosure statement with beneficial owners, renewed every two years, plus a pro rata assessment	None
Regulator	New office in the Department of Financial Services, with rulemaking power	Office of Emergency Services receives reports; no general rulemaking grant
Penalties	Up to $1M first, $3M per subsequent violation (Attorney General); $1,000 per day for missing disclosure (office)	Up to $1M per violation (Attorney General)
Whistleblower protections	None in the final text	Yes, with an anonymous internal reporting channel for large frontier developers
Territorial limit	Models developed, deployed or operating at least in part in New York	No express limit
Exemptions	Accredited New York colleges and universities doing academic research; Empire AI	None comparable
Public computing cluster	None	CalCompute consortium

A company already complying with SB 53 can likely reuse most of its framework and transparency reports in New York, because the required contents match closely. What it cannot reuse is its incident clock, which drops from 15 days to 72 hours. A large frontier developer operating in the state also needs a current disclosure statement on file with the DIGIT Office once the law takes effect on January 1, 2027. The Governor said on September 21, 2026 that the state would begin directing large frontier developers to register in November 2026.

Can the federal government block it?

Not as of November 2026. No federal statute preempts state frontier-AI laws, and an executive order cannot void a state statute on its own. Executive Order 14365 of December 11, 2025 directs a Justice Department task force to challenge state AI laws in court, and we found no federal suit against the RAISE Act. The status of that effort, the FRONTIER Act in the House and the letter from 26 state attorneys general are tracked in the frontier AI laws guide.

Dates and dollar figures here are taken from the enacted text and are stated as of November 2, 2026. On that date we found no rules, filing forms or assessment amounts from the DIGIT Office. This guide is general information, not legal advice.

Last updated Nov 2, 2026 by ZSKFL Management. Time-sensitive facts on this page are stated as of that date.