Since 2017, a bank or insurer licensed in New York has had to tell the Department of Financial Services within 72 hours when it determines that a reportable cybersecurity event has occurred. The rule sits in section 500.17 of the state's cybersecurity regulation, 23 NYCRR Part 500, and a whole compliance profession has grown around that clock. From 1 January 2027, the same department will receive a second kind of 72-hour report, this time from the companies that train the largest artificial intelligence models.
New York's Responsible AI Safety and Education Act, the RAISE Act, reached that point through five steps over thirteen months. Read in order, they show how a state frontier AI statute moved from a contested bill to a supervisory regime with a registration portal, a fee structure and a home in a financial regulator.
Practical takeaway. A large frontier developer has until 1 January 2027 to register through the RAISE Portal. From that date it must keep a published frontier AI framework and publish transparency reports for new frontier models, each on its own schedule, and it must report a critical safety incident to DIGIT within 72 hours after determining that one occurred or learning facts that give a reasonable belief that one did. DIGIT sits inside the Department of Financial Services; the Attorney General enforces.
19 December 2025: Governor Hochul signs the original RAISE Act
Hochul signed the RAISE Act on 19 December 2025, according to the Wiley alert of 3 April 2026. The signature came with an agreement to amend the text in the following legislative session, a common New York practice called a chapter amendment. The original law required a published "safety and security protocol" and was drafted before California had settled its own frontier statute, so the two states' first texts used different thresholds and vocabulary.
27 March 2026: the chapter amendment aligns New York with California's SB 53
The chapter amendment, bill S8828, was signed on 27 March 2026. It produced the final text that takes effect on 1 January 2027. As Cooley described it on 1 April 2026, the amendment gave the law "a California makeover" with some key differences. A frontier model is now a foundation model trained with more than 10^26 integer or floating-point operations, the same threshold as California's Transparency in Frontier Artificial Intelligence Act, SB 53, in force since 1 January 2026. A large frontier developer is one whose annual gross revenue, together with its affiliates, exceeded $500 million in the preceding calendar year, again the California figure.
The obligations follow the same pattern. Large frontier developers publish a frontier AI framework on their websites, review it and, as appropriate, update it at least once a year, and publish any material modification with a justification within 30 days. Before or at the time it deploys a new or substantially modified frontier model, a frontier developer publishes a transparency report on the release date and intended uses; large developers add summaries of their catastrophic-risk assessments. A system or model card that contains that information counts as compliance. The main difference lies in incident reporting. California allows 15 days. New York requires a report to the new oversight office within 72 hours, and within 24 hours to a law enforcement or public safety agency when an incident poses an imminent risk of death or serious physical injury.
Enforcement belongs to the Attorney General, who may seek civil penalties against a large frontier developer for specified violations of up to $1 million for a first violation and up to $3 million for later ones. The law creates no private right of action. Separately, the oversight office may impose $1,000 a day, after notice and a hearing, on a large developer that fails to file its disclosure statement or to correct false information in it.
21 September 2026: DFS receives the DIGIT Office and its duties are set out
On 21 September 2026 the Governor's office announced the next steps. The oversight office would be the Office of Digital Innovation, Governance, Integrity and Trust, known as DIGIT, established within the Department of Financial Services. The release said that registration would begin in November and compliance in January 2027, and it listed what large developers must do: register with DIGIT, file a disclosure statement at least every other year from 1 January 2027, pay assessments, report critical safety incidents within 72 hours and file quarterly assessments of catastrophic risk. DIGIT will also publish an annual public report summarizing the incident reports it receives.
The disclosure statement is a short, practical document. It identifies the developer's business names, its New York office addresses and its beneficial owners. Where the developer or its ultimate parent is privately or closely held, it lists everyone holding a beneficial interest of 5% or more and those who held such an interest in the preceding five years; where the owner or ultimate parent is publicly traded, it lists beneficial owners of 50% or more of the developer. It is renewed every two years and must also be updated after a material change in its information or a transfer of ownership of a frontier model. That is familiar territory for DFS, which already asks licensed institutions who owns and controls them.
10 November 2026: the RAISE Portal opens
On 10 November 2026 Hochul announced that large frontier developers must register through the newly released RAISE Portal by 1 January 2027, and that DIGIT has published a website setting out the requirements for frontier developers operating in New York. In the same announcement she launched the New York Center for AI Responsibility and Research at Binghamton University, funded by a $30 million philanthropic gift and an initial $25 million SUNY research investment. The AI Insider reported on 12 November that the center will independently evaluate advanced AI systems and work with DIGIT in reviewing companies' safety frameworks, and that a state spokesperson declined to describe the enforcement approach for companies that fail to register.
1 January 2027: what changes on the first day
On 1 January the registration deadline and the substantive duties arrive together. A large developer that has registered and published its framework must then be ready to recognize a critical safety incident, decide within hours whether it carries an imminent risk to life, and file with DIGIT inside 72 hours. The duty lands on engineering and security teams as much as on lawyers. Companies already subject to California's 15-day rule will need a faster internal escalation path for New York, and in practice that faster path will tend to become the default for both states.
How Quentir Reads It
The most consequential decision in this chronology was the choice of supervisor. New York could have placed frontier AI oversight in a technology office or left it to the Attorney General alone. It chose the Department of Financial Services, an agency whose method is registration, periodic disclosure, assessments that fund the supervisor and incident clocks backed by examination. The 72-hour report echoes the DFS cyber rule that banks and insurers have lived with since 2017; whether the drafters had that rule in mind is our inference, but the agency receiving the reports knows how to run one. Frontier developers should expect supervision that asks for documents on a schedule and follows up on gaps.
Three further consequences follow. For other states, New York and California now share thresholds and definitions while differing on speed, which makes a common national baseline easier to write and a race to shorter clocks more likely. For federal policy, the state regime sits alongside the federal pre-release review framework discussed in our post on Executive Order 14409 and its statement that it creates no AI licensing requirement; registration with DIGIT gives a state supervisor its own list of large frontier developers and their owners. For the public, DIGIT's annual public summary of the incident reports it receives will give a regular official account of what goes wrong with the most capable models. New York has also legislated on chatbots used by minors, according to the same AI Insider report, and our earlier reading of Florida's Safe Screens Act and its age-check duties for AI chatbots shows how differently states approach the consumer end of the same technology.
The open question is how DIGIT will use its rulemaking power and its authority to request more data, which Assemblymember Alex Bores highlighted in the September announcement. The first DIGIT rules and the first annual incident summary will show whether 72 hours becomes a national norm. Quentir follows this thread across state, federal and EU instruments, and the All-access membership brings that coverage together with every Signature Brief and Signature Report as it publishes and the full archive, under one organization-wide license.
Sources: Office of Governor Kathy Hochul, "Governor Hochul Declares NY Will Lead the Nation in AI Science and Safety", 10 November 2026; Office of Governor Kathy Hochul, "AI Safety: Governor Hochul Announces Next Steps to Regulate Major AI Developers", 21 September 2026; Wiley Rein, "New York Finalizes RAISE Act for Frontier AI Models; Law Takes Effect January 1, 2027", 3 April 2026; Cooley, "New York's Frontier AI Law Gets a California Makeover – With Some Key Differences", 1 April 2026; New York State Senate, bill S8828 (RAISE Act chapter amendment), signed 27 March 2026; The AI Insider, "New York Opens Registration for Frontier AI Labs Ahead of RAISE Act", 12 November 2026; New York Department of Financial Services, 23 NYCRR 500.17, notices to the superintendent, and DFS cybersecurity resource center. Links checked 14 November 2026.
Published intelligence, built to inform your own decisions. Published: November 14, 2026.