New York Reins In OpenAI and Anthropic: a Critical Incident Must Be Reported Within 72 Hours, and Fines Go Up to 3 Million Dollars

New York has opened a registration portal and given the biggest players in artificial intelligence a deadline: OpenAI, Anthropic and everyone like them must register by 1 January 2027. That is when the RAISE Act takes effect, one of the most serious AI safety laws in the US - and a law that asks the labs for more than goodwill.

The rules apply to so-called large frontier model developers, companies with more than 500 million dollars in annual revenue. They will have to publish their safety framework, report every critical safety incident within 72 hours and regularly publish transparency reports on their models. The half-billion threshold is no accident: small startups stay out, and the ones in the crosshairs are exactly those leading the race.

Oversight goes to the new Office of Digital Innovation, Governance, Integrity and Trust (DIGIT). Fines go up to 1 million dollars for a first violation and up to 3 million dollars for each subsequent one. The state can also go after companies that fail to register, but a spokesperson declined to say how. For firms that raise billions in a single funding round, three million dollars is pocket change. Will it be the fine that makes them careful - or the shame of a publicly reported incident?

Governor Kathy Hochul didn't mince words. "Tech titans tell us their vision for AI is inevitable," she said at a press conference. "They tell us: get on board or get left behind... That any collateral damage is worth it, as long as we win the race to the AI abyss." It is rare to hear a politician publicly call the AI race an abyss.

The same day, Hochul also opened the New York Center for AI Accountability and Research at Binghamton University. The centre will evaluate advanced systems independently and help DIGIT check companies' safety frameworks. It was announced back in January, with 55 million dollars in donations and capital. "We're building a national model for smart, sensible AI regulation, one that puts communities over corporations and people over powerful technology," Hochul said.

RAISE isn't New York's only law of this kind. Since last year a rule has required operators of AI companions to regularly remind users they are talking to a bot and to try to recognise signs of suicidal thoughts and self-harm. In October Hochul also signed a separate chatbot safety law for minors, which bans deceptively posing as a human, emotional manipulation and sycophancy, as well as encouraging suicide or eating disorders. It takes effect on 1 June 2027.

California, with its SB 53 law, has since January required safety frameworks, whistleblower protection and reporting of a critical incident within 15 days. Illinois, with SB 315, goes a step further from early 2027, with independent audits. The rules for the most powerful technology of the decade are being written by the states, one at a time, while the European Union has one law for all its members. Will America end up with fifty different versions? And are 72 hours short enough when the model is already in the hands of millions of people?