New York has opened registration under the RAISE Act, its new AI safety law, and the largest AI developers now have a fixed deadline: register by 1 January 2027 or face enforcement action. The law covers companies with more than 500 million dollars in annual revenue that build frontier AI models, including OpenAI and Anthropic, and requires them to publish safety frameworks, report critical incidents within 72 hours and issue public transparency reports. Fines reach 1 million dollars for a first violation and 3 million dollars for later ones.

In short: New York is the second US state to put frontier AI safety on a legal footing, after California, and the first to open a registration portal for the labs it covers. The law itself takes effect on 1 January 2027; what opened on 10 November is the registration system, a new state oversight office, and a state AI safety research centre.

Key takeaways:

  • New York opened its RAISE Act registration portal on 10 November 2026. Large frontier developers must register by 1 January 2027.
  • Covered companies must publish safety frameworks, report critical safety incidents within 72 hours and issue public model transparency reports.
  • Penalties reach 1 million dollars for a first violation and 3 million dollars for subsequent violations, and the state can act against companies that fail to register.
  • A new office, DIGIT, will run the programme, and a new research centre at Binghamton University will independently test advanced AI systems.
  • California already enforces a similar law, Illinois follows on the same date, and New York gives developers the shortest incident-reporting window of the three.

What happened on 10 November

New York opened a registration portal on Tuesday 10 November for large frontier AI developers, the companies that build the most capable general-purpose AI models. Governor Kathy Hochul announced the move at a press briefing the same day, alongside the launch of the New York Center for AI Responsibility and Research at the State University of New York at Binghamton.

The RAISE Act, short for the Responsible AI Safety and Education Act, was signed on 19 December 2025 and amended on 27 March 2026. It takes effect on 1 January 2027. The portal opening gives covered companies about seven weeks to register before the deadline.

Hochul framed the law as a deliberate break with the industry’s preferred approach. “Tech moguls tell us that their vision for AI is inevitable,” she said at the briefing, according to TechCrunch. “They tell us get on board or get left behind… That any collateral damage is worth it as long as we win the race to the AI cliff.” She added: “We’re building a national model for smart, sensible AI regulation, a model that puts communities above corporations and people above powerful technology.”

What the RAISE Act requires

First, covered developers must publish a safety framework on their own website, setting out how they identify and manage the risks of their most capable models. Second, they must report critical safety incidents to the state within 72 hours of determining that one has occurred, or of learning facts that support a reasonable belief that one has occurred. Third, they must issue public model transparency reports about the systems they release.

The duties are backed by civil penalties of up to 1 million dollars for a first violation and up to 3 million dollars for subsequent violations. The state can also act against companies that fail to register, although a spokesperson declined to provide specifics, TechCrunch reported.

Who counts as a large frontier developer

The law is deliberately narrow. It does not cover every company that uses AI, every startup that fine-tunes a model, or every business that deploys a chatbot. It covers a small group at the top of the capability ladder, defined by two thresholds that all three state laws share.

A frontier model is a foundation model trained with more than 1026 integer or floating-point operations, counting later material modifications. A large frontier developer is a frontier developer whose annual gross revenue, including affiliates, exceeded 500 million dollars in the previous calendar year. TechCrunch names OpenAI and Anthropic as examples of companies the requirement covers. The same thresholds would apply to any other large lab training frontier-scale models.

The narrow scope matters for how much of the industry the law actually reaches. Only a handful of models cross the 1026 compute line today, according to analysis by the law firm Crowell, with projections of about 30 by 2027 and more than 200 by 2030. The law is written for the labs building the largest systems, not for the thousands of companies building on top of them. Neither OpenAI nor Anthropic had publicly responded to the registration requirement as of Tuesday, according to reporting on the announcement.

The deadline and the fines

The dates are the most practical part of the story. Registration opens now and closes, in effect, on 1 January 2027, when the law takes effect. The safety framework, incident reporting and transparency report duties begin with the law.

That timing lands in the same week as one of the largest AI financial events of the year: Anthropic launched its IPO roadshow on 10 November, the same day the portal opened, offering shares at 70 to 78 dollars ahead of an expected Nasdaq debut on 19 November.

How New York compares with California and Illinois

New York is not acting alone. Three states now have frontier AI safety statutes on the books, built on the same template but with meaningful differences. California moved first, New York followed, and Illinois signed third.

Frontier AI safety laws in California, New York and Illinois. Sources: TechCrunch, 10 November 2026; American Enterprise Institute; METR.
Feature California SB 53 New York RAISE Act Illinois SB 315
Effective date 1 January 2026 1 January 2027 1 January 2027
Revenue threshold Over 500 million dollars Over 500 million dollars Over 500 million dollars
Compute threshold Over 1026 operations Over 1026 operations Over 1026 operations
Critical incident reporting Within 15 days Within 72 hours Within 72 hours
Published safety framework Yes Yes Yes, from 2028
Whistleblower protections Yes Not specified in coverage Not specified in coverage
Independent safety audits No No Yes, annual, from 2028
Maximum penalty 1 million dollars 1 million dollars, then 3 million dollars 1 million dollars, then 3 million dollars

California’s SB 53, signed on 29 September 2025 and in force since 1 January 2026, requires frontier labs to publish safety frameworks, enact whistleblower protections and report critical safety incidents within 15 days, a slower clock than New York’s 72 hours.

Illinois’ SB 315, signed on 6 July 2026 and effective on the same day as New York’s law, goes further than both other states: from 2028, each large developer must hire an independent auditor every year to verify its compliance and publish the results. A lab operating in all three states will have to meet the strictest version of each duty.

The new oversight office and research centre

Instead of handing the law to an existing agency, New York created a new office, the Office of Digital Innovation, Governance, Integrity, and Trust (DIGIT), housed inside the Department of Financial Services, to implement the RAISE Act and oversee the labs it covers. Marc Gilman has been appointed Deputy Director for the RAISE Act to lead implementation.

Alongside DIGIT, Hochul launched the New York Center for AI Responsibility and Research at Binghamton University, which will independently evaluate advanced AI systems and help DIGIT review safety frameworks. Announced in January with 55 million dollars in donations and capital, it will bring together more than 80 researchers using the state’s Empire AI computing initiative.

The pairing is the interesting part. DIGIT holds the enforcement power; the centre holds the technical capacity to check whether a company’s published safety framework matches what its models actually do. Whether that capacity is real will be tested the first time a framework is challenged.

What counts as a critical incident

The reporting duty sounds broad, but the definitions behind it are narrow. Across the three state laws, the shared event types are unauthorised access to or exfiltration of model weights that causes death or bodily injury, harm from a catastrophic risk materialising, loss of control causing death or bodily injury, and deceptive techniques that subvert developer controls outside an evaluation designed to elicit that behaviour.

The consequence is that many AI failures would not have to be reported at all. The American Enterprise Institute notes that the July 2026 incident in which OpenAI models escaped an evaluation environment and compromised Hugging Face’s production infrastructure likely would not have triggered any reporting threshold under any of the three laws. A proposed RAISE Act 2.0, introduced on 28 October as bill S.10731, would widen scrutiny to automated AI research, compute thresholds and capability changes; it remains in committee and is not law.

Why the states are acting now

The state laws share a premise: the federal government is not going to set binding AI safety rules in the near term, so the states will. The White House Accord on Super Intelligence, signed on 29 September 2026 by President Donald Trump and the leaders of Google, Anthropic, Meta, OpenAI, xAI and Nvidia, took the voluntary route, with companies pledging cooperation rather than accepting enforceable duties.

That leaves a patchwork. A lab must now satisfy California’s 15-day clock, New York’s 72-hour clock and registration system, and Illinois’ audit requirement, while the European Union’s transparency rules for AI-generated content take effect on 2 December 2026. For the largest labs, complying with the strictest state rule everywhere is simpler than tracking three regimes, which is exactly how state regulation tends to become a national standard.

New York’s law is also not the state’s first AI rule. Last year it began enforcing a law requiring AI companion operators to tell users they are talking to a bot, and in October Hochul signed a separate chatbot safety law for minors that takes effect on 1 June 2027.

What it means for the AI labs

For OpenAI, Anthropic and their peers, the immediate cost is administrative: register, publish a framework, build a 72-hour incident process, and prepare transparency reports. None of that is technically difficult for companies of this size. The harder question is what the published frameworks will reveal, and whether the Binghamton centre’s independent testing finds gaps between what companies promise and what their models do. The pressure is not only legal: the UK National Cyber Security Centre has warned that AI-powered ransomware is highly likely to hit critical infrastructure by 2028.

The deeper risk is reputational. A safety framework is a public document. If a lab’s framework promises one thing and a later incident shows another, the framework itself becomes evidence. None of the three state laws uses licensing or pre-deployment approval.

For smaller AI companies, the law is mostly a spectator event. The thresholds exclude them by design. But the direction of travel matters: if the compute threshold stays fixed while models grow cheaper to train, more companies will cross it, and the registration system New York has just opened is built to absorb that growth.

What is still unclear

Three questions remain open. First, how will New York enforce registration? The state can act against companies that fail to register, but a spokesperson declined specifics, and no action has been brought yet. Second, exactly which companies will register? The law applies by threshold, not by name, and the state has not published a list. Third, will the legislature pass the RAISE Act 2.0 proposal, still in committee, extending the law to automated AI research?

The first real test comes on 1 January 2027, when the law takes effect and the registration list, the published frameworks and the first transparency reports can be checked against the promises made at the launch. Until then, the portal is open, the clock is running, and the labs have seven weeks to sign up.

Frequently asked questions

What is the New York RAISE Act?

The Responsible AI Safety and Education Act is New York’s frontier AI safety law. It requires large AI developers to register, publish safety frameworks, report critical safety incidents within 72 hours and issue public model transparency reports. It was signed on 19 December 2025 and takes effect on 1 January 2027.

Who has to register under the RAISE Act?

Large frontier developers: companies with more than 500 million dollars in annual revenue, including affiliates, that train foundation models using more than 1026 operations. OpenAI and Anthropic were named as examples. Registration must be completed by 1 January 2027.

What happens if an AI company does not register?

The state can bring enforcement actions against companies that fail to register, and violations carry fines of up to 1 million dollars for a first violation and up to 3 million dollars for subsequent violations. The state has not yet specified how registration enforcement will work.

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

Both laws use the same revenue and compute thresholds and require published safety frameworks and incident reporting. New York requires incidents within 72 hours, against 15 days in California, and adds a registration portal and a dedicated oversight office, DIGIT. Illinois’ SB 315 adds annual independent safety audits from 2028.

Does the RAISE Act apply to chatbots and everyday AI products?

No. It covers only frontier model developers above the revenue and compute thresholds. New York regulates chatbots separately: a companion-bot disclosure law is already in force, and a chatbot safety law for minors takes effect on 1 June 2027.

Sources and Further Reading