New York RAISE Act AI safety rules: registration from 2027

- New York has opened a registration portal for frontier AI developers ahead of the RAISE Act taking effect at the start of 2027.
- Companies with annual revenue above $500 million must register by 1 January 2027 and publish safety frameworks and transparency reports.
- Critical safety incidents must be reported within 72 hours, with penalties of up to $1 million per violation and $3 million for repeat breaches.
- California and Illinois have introduced separate AI safety measures with different reporting deadlines and audit requirements.
New York’s RAISE Act converts voluntary assurances into a filing obligation. The dispute it has opened concerns what oversight means when companies supply the evidence themselves, rather than whether frontier AI needs oversight at all.
Supporters describe the regime, which takes effect at the start of 2027, as among the strictest state-level arrangements in the United States. According to TechCrunch’s report on the registration portal and the new law, developers with annual revenue above $500 million must enrol by 1 January 2027, publish safety frameworks, report critical incidents within 72 hours and issue public transparency reports. Enforcement sits with the Office of Digital Innovation, Governance, Integrity, and Trust, with penalties reaching $1 million per violation and $3 million for subsequent breaches, alongside separate action against companies that fail to register at all.
Critics rarely argue that transparency is undesirable. Their objection is that self-authored frameworks make weak evidence. A company that writes its own safety document and then reports against it has produced a record. That is not the same thing as an accountability mechanism. Governor Kathy Hochul used the announcement to reject the claim that AI progress is inevitable and therefore beyond regulation. She paired the law with a $55 million centre at Binghamton University, funded by donations, to evaluate advanced systems and review company frameworks alongside the regulator.
The established facts are narrow, and worth separating from the claims around them. Registration, disclosure, incident reporting and financial penalties are written into statute. What remains uncertain is how demanding published frameworks will be in practice. Whether the $500 million threshold leaves a meaningful gap beneath it is a further question. So is whether a donation-funded evaluation centre can scrutinise the industry that surrounds it. California’s SB 53, effective in January, allows 15 days for incident reports and adds whistleblower protections; Illinois’ SB 315, starting in 2027, requires independent compliance audits. Different clocks and different auditors invite regulatory arbitrage, though no evidence yet shows it occurring. Enforcement capacity is the untested variable, since a register is cheap to open and expensive to police.
For universities and schools buying AI tools, the practical consequence is procurement rather than philosophy. An institution able to ask a vendor for its published framework, transparency report and incident record is better placed than one relying on assurances offered in a sales meeting. Whether those documents will contain anything inconvenient is a question the statute cannot answer for itself.