The biggest AI companies now have a deadline in New York. Any frontier AI developer with more than $500 million in yearly revenue must register with the state by 1 January 2027, and from then on it has 72 hours to report a serious safety incident.
The registration portal opened on Tuesday 10 November. The announcement came from the governor’s office, alongside a new state-backed centre for testing AI systems. The rules come from the RAISE Act, a state law on AI safety and transparency. A new office called DIGIT, which sits inside the state’s financial regulator, will run it.
Who has to sign up
The law covers “frontier models”, meaning the largest AI systems. The regulator’s guidance defines them by training compute. A model counts if it was trained with more than 10^26 operations, a level of computing power that only a handful of labs reach today. A developer becomes a “large frontier developer” if it and its affiliates earned more than $500 million in the previous calendar year.
Large developers must do four main things:
- Register with the state and renew the filing at least every two years.
- Publish a safety framework explaining how they assess and reduce catastrophic risks and protect unreleased model weights.
- Send the regulator a summary of catastrophic-risk assessments from their own internal use of the models every three months.
- Publish a transparency report each time they release a new or substantially changed model.
Smaller frontier developers still have to report critical safety incidents and publish transparency reports. If an incident poses an imminent risk of death or serious injury, the guidance says it must be reported to a relevant authority within 24 hours.
What breaking the rules could cost
Penalties reach up to $1 million for a first violation and $3 million for later ones, according to reporting on the law. The state has not said what happens to a company that simply fails to register.
New York is also setting up the New York Center for AI Responsibility and Research at Binghamton University. It will bring together more than 80 researchers to test advanced AI systems independently and review the safety frameworks companies publish. The centre is funded by a $30 million philanthropic gift and an initial $25 million from the state university system.
“The stakes are too high to simply take AI companies at their word,” Governor Kathy Hochul said.
The move follows an earlier state law on companion chatbots and arrives in the same week that Australia announced its own plans for an AI safety regulator.
Why it matters: For the first time, the largest AI labs face a dated, enforceable obligation in a major US state to tell regulators when something goes badly wrong. The 72-hour clock gives the public a way to learn about serious AI failures that companies might otherwise keep quiet.
