On November 10, 2026, New York opened registration for frontier AI developers, and Governor Kathy Hochul announced the next steps in implementing the Responsible AI Safety and Education (RAISE) Act the same day. Under the law, large frontier model developers with more than $500 million in annual revenue must register by January 1, 2027, and then disclose their safety practices and report major safety incidents as required. TechCrunch reports that OpenAI and Anthropic are both in scope. It is the first state-level system in the US to turn frontier-model safety disclosure into a legal duty: registration comes first, and the substantive compliance duties begin in January next year.
Who must register, and what they must file
The RAISE Act targets developers of "frontier models," defined by training compute: a model counts as frontier if it was trained with more than 10^26 operations. A developer qualifies as "large" — and must register with the state's new Office of Digital Innovation, Governance, Integrity, and Trust (DIGIT) — when it and its affiliates together exceeded $500 million in gross revenue in the previous calendar year. From January 2027, covered developers must do three things: publish a safety framework on their website, file regular model transparency reports, and report critical safety incidents within 72 hours. They must also file a disclosure statement at least once every two years and pay assessments. DIGIT sits inside the Department of Financial Services, and the governor has already appointed Marc Gilman as Deputy Director for the RAISE Act to lead implementation.
The penalties are steep, but the list is not public yet
The cost of breaking the rules is spelled out precisely: the attorney general can bring civil actions, with fines of up to $1 million for a first violation and up to $3 million for subsequent ones. What is striking is that New York has not yet said publicly which companies DIGIT will treat as covered, and the details — how the incident-reporting clock starts, how submissions are verified — are still unwritten. The registration portal is open, in other words, but "who is on the list" remains an open question. For leading companies like OpenAI and Anthropic, the real test is not the registration form itself but the safety-framework publication and incident reporting that start in January: what they write, and how fast they report, will be read under a magnifying glass by peers and the public alike.
The act was signed by Hochul in December 2025, and when the implementation timetable was published in September, the rhythm was already set — registration from November, compliance from January — as New York's RAISE Act timeline: frontier AI firms must register from November, 72-hour incident reporting effective next January explained at the time. This step is simply the timetable playing out. The real dividing line comes in January: only when the first registration list and the first safety frameworks appear will it be clear how sharp the act's teeth really are.
