TL;DR: New York has opened the registration portal that frontier AI developers like OpenAI and Anthropic must use before January 1, 2027, when the state’s RAISE Act takes effect. Large developers face fines up to 3 million dollars for repeat violations, and the law requires safety frameworks, incident reporting within 72 hours, and quarterly risk summaries.

Governments move slowly, except when they are regulating AI, where apparently they move at a reasonable clip.
New York’s Department of Financial Services opened a registration portal this month for what the state is calling the RAISE Act, short for Responsible AI Safety and Education. Starting January 1, 2027, any company that qualifies as a large frontier AI developer has to maintain an active disclosure statement with a newly created oversight office, the Office of Digital Innovation, Governance, Integrity, and Trust, known inside state government simply as DIGIT. OpenAI and Anthropic are both explicitly named as companies on the hook, alongside any other lab that clears the law’s spending threshold.
Governor Kathy Hochul signed the final version of the RAISE Act on December 19, 2025, after a round of amendments meant to align it more closely with California’s own AI transparency law, the Transparency in Frontier Artificial Intelligence Act. That alignment matters practically, because it means a company already building compliance infrastructure for California’s law is not starting from zero when New York’s requirements kick in a year later.
The law splits obligations into two tiers. Every frontier developer operating in New York has to report what the law calls critical safety incidents through a dedicated RAISE portal, regardless of company size. Large frontier developers, defined as companies that have trained at least one frontier model and spent more than 100 million dollars in aggregate compute costs doing it, face the heavier lift: filing a detailed disclosure statement, paying assessments tied to that registration, and submitting quarterly summaries of catastrophic risk once they are registered and approved.
“We’re building a national model for smart, sensible AI regulation, a model that puts communities above corporations and people above powerful technology,” Hochul said when the center overseeing the law was first announced, with 55 million dollars in donations and capital behind it.
The penalties are real money even for a well-funded AI lab. A first violation can cost a company up to 1 million dollars, and repeat violations climb to 3 million dollars each. That is not enough to bankrupt a company the size of OpenAI or Anthropic, but it is enough to show up on a quarterly earnings call, and more importantly it creates a paper trail of noncompliance that plaintiffs’ lawyers in other states will absolutely cite later.
New York is explicitly not acting alone here, and the state’s own materials frame the RAISE Act as part of a coordinated, if uncoordinated-looking, wave of state-level AI rules. California’s SB 53 already took effect in January and requires frontier labs to publish safety frameworks, put whistleblower protections in place, and report critical safety incidents within 15 days, a tighter window than the federal government has ever proposed. New York also passed a separate chatbot safety law in October 2026 focused specifically on protecting minors, which takes effect June 1, 2027, layering a second compliance deadline onto the same companies within the same calendar year.
Both OpenAI and Anthropic have publicly supported versions of the RAISE Act, which is a notably different posture than the tech industry’s usual reflexive opposition to new regulation. OpenAI’s chief global affairs officer Chris Lehane told the New York Times that having similar legislation across two large state economies is “a big step in the right direction” for AI policy generally, essentially arguing that one coherent multi-state standard beats fifty different ones. That is a defensible position for a company that would otherwise be compliance-testing against a patchwork, though it is also the position any company facing inevitable regulation would prefer: shape the rules early rather than fight them later.
Not every industry group agrees. Lobbying organizations representing tech companies broadly resisted the RAISE Act as it moved through the legislature and have signaled they intend to keep pushing back on similar state laws going forward, arguing that a state-by-state approach to AI safety creates exactly the fragmented compliance burden Lehane claims to want to avoid.
The registration portal itself, run through DIGIT, asks for more than a one-time filing. Large frontier developers have to designate a point of contact responsible for incident reporting, describe the testing and evaluation process used before a model ships, and commit to updating that disclosure statement whenever a material change occurs, not just on some fixed annual schedule. That ongoing-maintenance requirement is a meaningfully heavier lift than a simple registry entry, since it effectively asks a compliance team to treat the state filing as a living document rather than paperwork filed once and forgotten.
Why this lands differently than it would have a year ago
This registration requirement arrives in the same stretch of weeks as Anthropic’s own Nasdaq debut, which briefly pushed the company’s valuation past 2 trillion dollars. A newly public company with that kind of market scrutiny has a very different relationship to regulatory risk than a private lab burning venture capital. Public disclosure obligations under securities law and public disclosure obligations under a state AI safety law start to rhyme uncomfortably, and a critical safety incident that has to be reported to New York within 72 hours is also, almost certainly, a material event that needs disclosing to shareholders. The RAISE Act did not anticipate that overlap when it was drafted in 2025, but it is going to live inside it starting next year regardless.
The bottom line
January 1, 2027 is still more than a month out, but the portal opening now is the real start of the clock, not a formality. New York has built a law with actual enforcement teeth, a dedicated oversight office, and penalties that scale with repeat offenses, and it did so with the support, rather than the opposition, of the two companies it most directly targets. Whether that cooperative posture survives contact with the first real enforcement action is the question worth watching into 2027.



