A working reference to the RAISE Act’s defined terms for frontier-model developers and their counsel. Figures are drawn from the enacted text of the RAISE Act (S.8828).
New York's Responsible AI Safety and Education Act — the law on safety and transparency for developers of the most capable AI models. It creates Article 44-B.
The RAISE Act adds a new Article 44-B to New York's General Business Law. It requires large frontier developers to publish a frontier AI framework, requires a transparency report before deploying a frontier model, requires reporting of critical safety incidents to a new DFS office, and authorizes Attorney General enforcement. Governor Hochul first signed the Act on December 19, 2025; a chapter amendment passed both chambers and was signed in final form on March 27, 2026. The regime takes effect January 1, 2027.
The chapter of New York's General Business Law that the RAISE Act creates, running from § 1420 through § 1429. It is the operative text of the regime.
Article 44-B sits in the General Business Law as §§ 1420–1429: Definitions (§ 1420), Transparency requirements (§ 1421), Reporting (§ 1422), Loss of equity (§ 1423), Duties and obligations (§ 1424), Scope (§ 1425), Exceptions (§ 1426), Violations (§ 1427), Large frontier developer disclosure (§ 1428), and Rulemaking authority (§ 1429). The chapter amendment repealed the originally-passed Article 44-B and replaced it with this text.
The follow-on bill that repealed the originally-passed Article 44-B and replaced it with the final version — the one in force from January 1, 2027.
After the RAISE Act first passed (S.6953-B / A.6453-B) and was signed on December 19, 2025, the Governor negotiated a chapter amendment (Senate bill S.8828, introduced January 6, 2026). It passed the second chamber on March 11, 2026 and was signed March 27, 2026. The amendment repealed and re-enacted Article 44-B, adding the explicit compute (10²⁶ operations) and revenue ($500M) thresholds and bringing several definitions into line with California's TFAIA. Treat the chapter-amendment text as controlling.
Article 44-B takes effect January 1, 2027. Before that date there is no live filing obligation and no official state portal.
The regime's obligations begin on January 1, 2027 — a date set in § 3 of the act, not inside Article 44-B itself (the chapter amendment changed the original "ninetieth day after it shall become a law" to a fixed January 1, 2027). Because the start date is in the future, there is no official New York filing system yet; Frontier models what one should be, using voluntarily-published lab frameworks as illustrative reference entries — not actual 44-B filings.
A foundation model trained using more than 10²⁶ integer or floating-point operations — the class of models Article 44-B is written to govern.
Under § 1420, a frontier model is a foundation model trained using a quantity of computing power greater than 10²⁶ integer or floating-point operations. The count includes the original training run plus any subsequent fine-tuning, reinforcement learning, or other modification a developer applies to a preceding foundation model. The 10²⁶ threshold matches California's TFAIA and is what keeps the regime focused on the most resource-intensive models rather than ordinary or research-scale systems.
An AI model that is trained on a broad data set, designed for generality of output, and adaptable to a wide range of distinct tasks.
Section 1420 defines a foundation model as an artificial intelligence model that is all of the following: (a) trained on a broad data set; (b) designed for generality of output; and (c) adaptable to a wide range of distinct tasks. A frontier model is a foundation model that also crosses the 10²⁶-operation compute threshold — so 'foundation model' is the broader category and 'frontier model' the regulated subset.
A person who has trained — or initiated the training of — a frontier model. The transparency-report and incident-reporting duties attach at this level.
Under § 1420, a frontier developer is a person who has trained, or initiated the training of, a frontier model — including a person who used (or intends to use) at least as much computing power to train it as would meet the frontier-model specification. 'Person' is defined broadly to cover individuals and essentially any business entity. Transparency reports (§ 1421(3)) and critical-safety-incident reporting (§ 1422) apply to every frontier developer; the heavier framework and disclosure duties apply only to large frontier developers.
A frontier developer that, with its affiliates, had annual gross revenues over $500 million in the preceding calendar year. The heaviest duties attach here.
Section 1420 defines a large frontier developer as 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. This $500M revenue bar — matching California's TFAIA — is what triggers the core obligations: publishing a frontier AI framework (§ 1421), filing a biennial disclosure statement (§ 1428), and transmitting quarterly internal-use catastrophic-risk assessments (§ 1422).
A foreseeable, material risk that a frontier model materially contributes to the death of or serious injury to more than 50 people, or more than $1 billion in property damage, from a single incident.
Section 1420 defines catastrophic risk as a foreseeable and material risk that a frontier developer's development, storage, use, or deployment of a frontier model will materially contribute to the death of, or serious injury to, more than fifty people — or more than one billion dollars in damage to or loss of property — arising from a single incident, where the model does any of: (i) provides expert-level assistance in creating a chemical, biological, radiological, or nuclear weapon; (ii) engages in conduct with no meaningful human oversight that is a cyberattack, or that would be murder, assault, extortion, or theft if done by a human; or (iii) evades the control of its developer or user. It excludes risk from otherwise-public information, lawful federal-government activity, or harm where the model did not materially contribute. (Per § 1423, a loss of equity value does not count as property damage.)
The two numeric bars that decide who is covered: > 10²⁶ training operations marks a frontier model; > $500M annual revenue marks a large frontier developer.
Article 44-B draws its scope with numbers, not names. A model is a frontier model when trained with more than 10²⁶ integer or floating-point operations (§ 1420(9)); a developer is a large frontier developer when it and its affiliates exceed $500 million in annual gross revenue (§ 1420(10)). Both thresholds were added by the chapter amendment and match California's TFAIA. FilingsGPT keeps these config-driven because DFS rulemaking — and future amendments — could move them.
The documented protocols a large frontier developer must write, follow, and conspicuously publish, describing how it manages, assesses, and mitigates catastrophic risk.
Section 1421 requires every large frontier developer to publish a frontier AI framework on its website describing how it: incorporates national/international standards; defines thresholds to identify catastrophic-risk capability; applies and reviews mitigations; uses third parties to assess risk; updates the framework; secures unreleased model weights; institutes internal governance; identifies and responds to critical safety incidents; and assesses catastrophic risk from internal use. It must be reviewed at least annually, and a material modification must be published with a justification within thirty days. The published lab frameworks on /frameworks are illustrative examples of what this looks like.
A report a frontier developer must publish before deploying a new or substantially modified frontier model — covering its release, intended use, and catastrophic-risk assessment.
Under § 1421(3), before (or concurrently with) deploying a new frontier model or a substantially modified version, a frontier developer must conspicuously publish a transparency report: developer contact information and website; the model's release date, supported languages, and output modalities; intended uses and any use restrictions; and — for large frontier developers — summaries of the catastrophic-risk assessments, their results, and the extent of third-party evaluator involvement. A developer that publishes this within a larger document such as a system card or model card is deemed in compliance. Information may be redacted to protect trade secrets, cybersecurity, public safety, or national security, with the unredacted version retained for five years.
A defined serious-failure event involving a frontier model — the kind a frontier developer must report to the DFS office within 72 hours.
Section 1420(4) defines a critical safety incident as any of: (a) unauthorized access to, modification of, or exfiltration of model weights that results in death or bodily injury; (b) harm resulting from the materialization of a catastrophic risk; (c) loss of control of a frontier model causing death or bodily injury; or (d) a frontier model using deceptive techniques against its developer to subvert controls or monitoring, outside an evaluation and in a manner that demonstrates materially increased catastrophic risk. Section 1422 sets the reporting clocks for these incidents.
A critical safety incident must be reported to the DFS office within 72 hours — and within 24 hours to law enforcement if it poses an imminent risk of death or serious physical injury.
Section 1422 requires a frontier developer to report a critical safety incident to the Office within seventy-two hours of determining one occurred, or of learning facts sufficient to establish a reasonable belief that one occurred. If the developer discovers the incident poses an imminent risk of death or serious physical injury, it must also disclose to an appropriate law-enforcement or public-safety agency within twenty-four hours (and still report to the Office within 72 hours). New York's 72-hour clock is notably stricter than California TFAIA's 15-day timeline.
A large frontier developer must send the DFS office a summary of its catastrophic-risk assessment from internal use of its frontier models every three months.
Under § 1422(2), a large frontier developer must transmit to the Office a summary of any assessment of catastrophic risk resulting from internal use of its frontier models every three months (or on another reasonable schedule agreed with the Office), with written updates as appropriate. The Office must establish a mechanism for confidential submission and take all reasonable precautions to limit access to authorized personnel. This is a distinct, recurring obligation separate from incident reporting.
A new office within New York's Department of Financial Services, reporting to the Superintendent, tasked with implementing Article 44-B and publishing the list of large developers who have filed.
Section 1420(16) defines 'Office' as an office within the Department of Financial Services that reports to the Superintendent and is tasked with implementing the article. It receives transparency information, critical-safety-incident reports, and quarterly internal-use assessments; maintains and publishes the list of large frontier developers who have filed disclosure statements (§ 1428(6)); transmits an annual anonymized report to the Governor and Legislature beginning January 1, 2028; may share reports with the Attorney General and other agencies; and holds rulemaking authority (§ 1429). No such public portal is in operation yet — FilingsGPT is an independent reference implementation of what one should look like, NOT the Office or any state system.
A registration a large frontier developer must file with the Office — renewed every two years — before it may develop, deploy, or operate a frontier model in New York.
Section 1428 bars a large frontier developer from developing, deploying, or operating a frontier model in New York without a current disclosure statement on file and its required assessment paid. The statement is renewed every two years (or sooner on a transfer or material change) and identifies the developer and all business names; the principal place of business and each New York office address; beneficial owners of a 5%-or-greater interest (or 50%+ if publicly traded), including prior-five-year owners if privately held; and points of contact. The Office maintains and publishes the list of filers (minus contact details). Operating without a filing draws a $1,000-per-day civil penalty plus the assessments owed.
There is no private lawsuit under Article 44-B. The New York Attorney General enforces it, with civil penalties up to $1 million for a first violation and up to $3 million for subsequent violations.
Section 1427 vests enforcement in the New York Attorney General, who may bring a civil action for penalties not exceeding one million dollars for a first violation and not exceeding three million dollars per subsequent violation, scaled to severity. Violations include a large frontier developer's failure to publish a compliant framework or transparency report, a materially false statement in violation of § 1421(4), failure to report an incident under § 1422, or failure to comply with its own frontier AI framework. The article creates no private right of action. (Separately, the Office can levy a $1,000-per-day penalty under § 1428 for a missing disclosure statement.)
The Office may adopt rules to implement Article 44-B, including additional reporting or publication requirements that facilitate safety and transparency.
Section 1429 authorizes the Office to adopt rules and regulations to implement the article as needed. Where it determines doing so will facilitate safety and transparency consistent with the article's purpose, the Office may consider additional reporting or publication requirements — including post-critical-safety-incident information, sharing plans and protocols, and the transmission of frontier AI frameworks to the Office. Because much operational detail is delegated to this rulemaking, FilingsGPT keeps its disclosure model loosely coupled / config-driven rather than hardcoding a filing format DFS has not yet defined.
The voluntarily-published safety frameworks from the major frontier labs. FilingsGPT uses them as illustrative reference entries that map onto Article 44-B's disclosure duties — they are NOT actual 44-B filings.
Several frontier developers have published their own safety frameworks (Anthropic's Responsible Scaling Policy, OpenAI's Preparedness Framework, Google DeepMind's Frontier Safety Framework, Meta's Frontier AI Framework, Microsoft's Frontier Governance Framework, and xAI's Risk Management Framework). The RAISE Act's own findings note that major developers 'have already voluntarily established' such frameworks as industry best practice. FilingsGPT seeds its registry from these documents to show what a § 1421 frontier-AI-framework disclosure could look like. They are clearly labeled as illustrative — the regime does not begin until 2027, so no 44-B filings exist yet. Aggregators such as the METR FSP index and the Frontier Model Forum catalogue are used for comparison.
The enacted text of the RAISE Act / Article 44-B in New York's General Business Law (Senate bill S.8828). It is the primary source the chat assistant quotes from.
FilingsGPT's section summaries and the grounded assistant rest on the enacted text of Article 44-B (§§ 1420–1429) as chaptered by S.8828, the final chapter amendment. Definitions, thresholds (10²⁶ operations, $500M revenue), the catastrophic-risk figures (>50 people, >$1B), the 72h/24h reporting clocks, and the $1M/$3M penalties are taken directly from that text. Where DFS rulemaking will fill in operational detail, the site says so.
Definitions drawn from New York’s Article 44-B (General Business Law §§ 1420–1429), enacted by the RAISE Act and revised by the S8828 chapter amendment, effective Jan 1, 2027. Dollar amounts, deadlines, and the frontier threshold are drawn from the enacted text (S.8828). Reference entries on Frameworks are voluntarily-published lab safety frameworks, not official 44-B filings. Provided for orientation, not legal advice.
FilingsGPT is an independent reference implementation of New York's Article 44-B (the RAISE Act). It is not affiliated with or operated by the NY Department of Financial Services or the State of New York, and is not an official filing system.