New York Opens RAISE Act Registration, Putting OpenAI and Anthropic Under State Supervision for the First Time
The week of November 10 marked a quiet but structural turning point in artificial intelligence regulation in the United States. New York state officially opened the registration portal for the RAISE Act — the Responsible AI Safety and Education Act — meaning that OpenAI, Anthropic, and other large frontier model developers now have a formal deadline of January 1, 2027, to enroll in the state's new oversight system. Until then, the New York Center for AI Responsibility and Research, based at Binghamton University, is already operational, and the DIGIT (Digital Innovation, Governance, Integrity and Trust) office within the state's Department of Financial Services already has legal authority to take action against companies that refuse to register.
What makes this moment historically significant is that, for the first time in history, a U.S. jurisdiction has transformed voluntary safety promises into binding legal obligations for the most powerful companies in the technology sector. This is not a guideline or a directive — it is a law with penalties of up to $1 million per initial violation and $3 million for repeat offenses, in addition to civil action capability by Attorney General Letitia James.
What the RAISE Act requires from large developers
The RAISE Act applies to "frontier developers," defined as those that "trained or initiated the training" of high-compute models or foundation models trained with more than 10²⁶ computational operations. For "large frontier developers" — those with gross annual revenue above $500 million — the law imposes substantially additional requirements.
First, each large frontier developer must publish on its website a "Frontier AI Framework" — an annual governance framework describing risk structures, assessment thresholds, mitigation strategies, cybersecurity practices, and alignment with international standards. The law specifically requires the framework to address catastrophic risks arising from internal model use, an requirement that goes beyond what most voluntary industry standards cover.
Second, before deploying any frontier model, all developers must publish detailed transparency reports detailing the model — intended uses, modalities, restrictions — as well as summaries of catastrophic risk assessments, their results, and the role of any third-party evaluators. This represents a paradigm shift: what were previously internal AI governance documents are now public information under law.
The incident reporting requirement is arguably the most impactful. Developers must notify DIGIT about critical safety incidents within 72 hours from the moment they form a "reasonable belief" that an incident occurred — and only 24 hours in case of imminent danger of death or serious injury. The definition of critical incident is broad: contributes to death or serious injury of 50 or more people, causes at least $1 billion in damages, offers expert-level assistance in creating chemical, biological, or radiological weapons, involves criminal conduct or cyberattack without meaningful human intervention, or allows the model to escape its developer's control.
How the RAISE Act compares to California's SB 53
To understand the depth of what is happening in New York, it is useful to compare it with the equivalent legislation in California — SB 53, in effect since January 2026. The RAISE Act and SB 53 share broadly similar core requirements, but have significant structural differences that may shape how companies adapt.
According to the Foundation for Progress and Freedom (FPF), the RAISE Act establishes a specific frontier developer disclosure program — requiring information such as ownership structure, business addresses, and designated points of contact — that SB 53 does not require. In contrast, SB 53 includes whistleblower protections, absent from the RAISE Act.
The most notable difference lies in incident reporting timeframes: the RAISE Act requires 72 hours (24 hours in cases of imminent danger), compared to 15 days for SB 53. The RAISE Act uses stricter qualifiers, establishing that notification begins when there is a "reasonable belief" of an incident, while SB 53 uses a more open standard.
The RAISE Act also includes a territorial limitation — it applies only to models developed or operated in whole or in part in New York — which is not present in SB 53. FPF observes that this limitation may make the RAISE Act more likely to survive a challenge under the Dormant Commerce Clause, if contested judicially.
Additionally, unlike SB 53, the RAISE Act explicitly permits a large frontier developer to assert that alleged harm or damage was caused by another person, entity, or contributing factor — a legal safeguard absent from California legislation.
What comes next
Governor Kathy Hochul announced she is "exploring ways to build on the RAISE Act" in the coming weeks. This suggests the legislation may be expanded before even the January 1 deadline, possibly with chapter amendments already approved and expected to be formally adopted by the Assembly in January.
Other states are already following the model: Michigan has already introduced similar legislation (HB-4668), and Illinois already has SB 315 in effect from January 2027, which adds independent safety compliance audits to the transparency and reporting requirements.
OpenAI, in internal communications, has already described these three states (California, New York, and Illinois) as "beginning to create a de facto national framework" — which implicitly recognizes that federal regulation remains nonexistent while states advance in a coordinated manner.
The question remains: when New York's registration portal receives the first round of enrollments from OpenAI and Anthropic in January, what will they reveal — and what will they try to hide under the justification of trade secret, cybersecurity, or national security?
Sources: TechCrunch, Governor Kathy Hochul, FPF
✓ Independent sources cross-checked and verified before publishing
