New York Put a Clock on Frontier AI, and DIGIT Is the Regulator Behind It
The RAISE Act's chapter amendment covers labs past 10^26 operations and US$500 million in revenue, gives them 72 hours to report incidents and lets DFS fine non-filers US$1,000 a day. Registration is due in November.
Updated Oct. 19, 2026: This story has been rewritten against the statute itself. The chapter amendment settles the coverage question our Oct. 1 version left open (it is compute plus revenue, not a US$100 million compute bill), adds a 24-hour rule for imminent danger and a US$1,000-a-day filing penalty, and confirms that as of Oct. 19 New York has named only “November” for registration, with no published portal or form.
New York’s frontier AI law now has an office, a staffer and a calendar. On Sept. 21, Gov. Kathy Hochul announced that the Department of Financial Services will run the RAISE Act through a new Office of Digital Innovation, Governance, Integrity and Trust, or DIGIT, with Marc Gilman as deputy director for the act. The state will direct the largest AI developers to register in November 2026. Compliance starts in January 2027, and Attorney General Letitia James enforces it.
The release is the plan. The law is in S8828, the chapter amendment that rewrote the original act and became Article 44-B of the General Business Law, sections 1420 to 1429. We read it line by line, because the press release leaves out three things a compliance team needs: who is covered, how much a violation costs and what happens if you simply don’t file.
Who the RAISE Act covers: 10^26 operations and US$500 million
When we first wrote about the office on Oct. 1, published legal analyses disagreed on the threshold. Some described a test based on spending more than US$100 million on training compute. Others described a US$500 million revenue test. Both were describing a real text. They were describing different versions.
The bill Hochul signed in December 2025, S6953-B, keyed coverage to models whose training compute cost exceeded US$100 million. The chapter amendment replaced that. Under section 1420(9) of S8828, a frontier model is a foundation model trained with more than 10^26 integer or floating-point operations, counting the original run plus later fine-tuning and reinforcement learning. There is no dollar figure for compute anywhere in the definition.
The heavier duties then fall on a “large frontier developer,” defined in section 1420(10) as a frontier developer whose group, affiliates included, had “annual gross revenues in excess of five hundred million dollars in the preceding calendar year.” So the answer is both, stacked: the compute threshold decides whether you build a frontier model, and the revenue threshold decides whether most of the act applies to you. A well-funded startup that crosses 10^26 operations but books less than US$500 million is a frontier developer with lighter obligations.
In practice that list is short. Our read: it means the handful of US labs and big-tech parents already covered by California’s equivalent law, which is the point. The sponsor’s memo says the rewrite aligns New York more closely with California.
What large frontier developers must do, and by when
The DFS release lists four duties. The statute supplies the detail behind each.
- Publish a framework. Developers must set out safety and transparency frameworks and post them on their websites.
- Report incidents fast. Section 1422(3)(a) gives developers “seventy-two hours from a determination that a critical safety incident has occurred” to tell the office. Section 1422(3)(b) adds a tighter rule the release doesn’t mention: where an incident poses an imminent risk of death or serious physical injury, the developer must alert an appropriate authority, such as law enforcement, “within twenty-four hours.”
- Send risk summaries quarterly. Section 1422(2)(a) requires a summary of internal catastrophic-risk assessments “every three months,” or on another reasonable schedule the developer requests and the office accepts.
- File, renew and pay. Under section 1428, no large frontier developer may operate a frontier model in New York without a current disclosure statement on file and its “required share” of costs paid. Statements are renewed every two years, starting Jan. 1, 2027, or sooner after a change of ownership or a material change.
The fee is not a flat number. Section 1428(4) has DFS assess covered developers pro rata for the office’s administrative costs, which means the bill depends on how big DIGIT gets and how many companies share it.
RAISE Act penalties: US$1 million, then US$3 million
Section 1427 lets the attorney general bring a civil action for up to US$1 million “for a first violation” and up to US$3 million “per subsequent violation,” scaled to severity. There is no private right of action. The December 2025 text set those caps at US$10 million and US$30 million; the amendment cut them by 90%.
A second, quieter penalty sits in section 1428(5). After notice and a hearing, the office itself can fine a developer US$1,000 a day for failing to file a disclosure statement or for failing to correct false information in one, plus the assessments it owes. That is the provision that gives the November registration call real force. It doesn’t need the attorney general.
The compute threshold decides whether you build a frontier model. The revenue threshold decides whether most of the law applies to you.
RAISE Act vs California SB 53, side by side
California’s SB 53, signed Sept. 29, 2025, has applied since Jan. 1, 2026. New York borrowed its definitions and then went further on speed and on who’s watching.
| Provision | New York RAISE Act (S8828) | California SB 53 |
|---|---|---|
| Regulator | DIGIT, inside the Department of Financial Services | No new office; incidents go to the Office of Emergency Services |
| Enforcer | Attorney General Letitia James | California attorney general |
| Coverage | Over 10^26 operations; most duties at over US$500M revenue | Same two thresholds |
| Registration | Directed to register in November 2026 | None |
| Compliance starts | Jan. 1, 2027 | Jan. 1, 2026 |
| Incident reports | 72 hours; 24 hours if imminent danger | 15 days; 24 hours if imminent danger |
| Risk-assessment summaries | Every three months | Quarterly |
| Disclosure statement | Every two years, from Jan. 1, 2027 | None |
| Fees | Pro rata assessment of DFS costs | None |
| Civil penalties | Up to US$1M first, US$3M subsequent; US$1,000/day for not filing | Up to US$1M per violation |
Two rows matter most. New York’s 72-hour clock is far shorter than California’s 15 days, and New York’s registration and fee rows have no California counterpart at all. A lab compliant in Sacramento still has a filing to make in Albany. We track both alongside Colorado’s rewrite on our state AI law map.
Inside DIGIT: who runs it and what’s still missing
Gilman comes from compliance technology. The DFS release says he was general counsel and vice-president of compliance at Theta Lake and an adjunct professor at Fordham University School of Law. His stated aim, per the release, is to “create more transparency in this industry and support responsible AI development.” Acting Superintendent Kaitlin Asrow is the department’s head. Hochul’s own framing was that the companies building these systems owe the public an account: people “deserve to know what they’re doing to keep it safe.”
The office’s remit is broader than a mailbox for filings. Per the release, members of the public can report suspected critical safety incidents to DIGIT, and the office will publish an annual public report. Bloomberg reported on Sept. 21 that Hochul said the office could explore a kill switch, an idea that appears nowhere in the statute.
What we could not find, as of Oct. 19: a registration portal, a published form, a fee schedule or a specific November date. The DFS release says only that the state “will direct” developers to register in November. Companies that think they’re covered should not wait for a portal to start drafting the disclosure statement, because the statute already says what it must contain.
What it means
For readers, little changes on screen. The law’s audience is a few labs, and its main output is paperwork that, for the first time, a New York regulator can demand on deadline. The test comes on the first critical incident after Jan. 1: whether a developer files within 72 hours, and whether DIGIT says anything public about it.
For the labs, the questions are cost and overlap. The pro rata fee has no number yet. The quarterly summaries duplicate what California already gets, on a slightly different form. More on how labs decide what counts as catastrophic in our explainer on AI safety cases, and more policy coverage here.
The US$1,000-a-day clause is the one to watch. It starts running on paperwork, not harm, and in November it will be the first part of New York’s AI law with a date on it.
Researched and drafted with AI tools, fact-checked against primary sources and approved by a human editor before publishing. Prices and dates are as of publication. Our standards →
Sources- dfs.ny.gov — announced
- legislation.nysenate.gov — S8828
- nysenate.gov — S6953-B
- nysenate.gov — sponsor's memo
- legiscan.com — SB 53
- bloomberg.com — Bloomberg reported
Latest from prompt/power
- IBM, NVIDIA and Fermilab plan a quantum test bed. Here’s what each brings, and what’s still missingNov 19
- OpenAI’s free CSAM classifier misses half the infant images in its own test set. What the model card showsNov 18
- Anthropic’s last filing before pricing: Nasdaq approval, a US$6.5-billion stock-pay bill and a redrawn lock-upNov 18
- Celestica, the Toronto company building AI’s racks, expects US$40 billion in revenue next yearNov 18
- ChatGPT now lets brands claim a profile and run their own agent. What creators and publishers should knowNov 18