New York's RAISE Act gives big AI developers 72 hours to report serious incidents
New York's AI safety law sets a three-day reporting clock for critical incidents at large AI developers, with fines up to $3 million for repeat violations.
The short version
- Critical safety incidents must be reported within 72 hours through a dedicated state portal.
- The law covers developers with over $500 million in annual revenue whose models took over 10^26 operations to train.
- Fines run up to $1 million per violation, and up to $3 million for repeat violations.
- The agency DIGIT opened the RAISE portal on 10 November 2026; the law takes effect 1 January 2027.
- Large developers must also publish safety frameworks and transparency reports.
Why it matters
If you use AI tools at work or at home, serious problems with the biggest models may reach regulators within days instead of staying inside the company. It also shows that states are writing their own AI rules, and these rules come with real fines. Details such as the exact definition of a critical incident are not in our source, so treat this as a direction of travel.
What the law requires
New York's Responsible AI Safety and Education Act, known as the RAISE Act, was passed by the state Assembly and Senate on 12 June 2025. The bill then went to Governor Kathy Hochul for her signature. As BornCity reported, the law becomes binding on 1 January 2027.
Its central demand is speed. Covered companies must tell the state about critical safety incidents within 72 hours, which is three days. The report has to go through a dedicated state portal. Filing a note through the general complaint form of the Department of Financial Services does not satisfy the law.
The Senate's own summary says the act also imposes broad transparency duties on how so-called frontier models are trained and used. That includes writing and submitting binding safety plans.
Who counts as a frontier developer
A frontier model, in the law's terms, is a general-purpose foundation model. That means one large model built to handle many tasks, rather than a single narrow tool. To qualify, its training must have used more than 10^26 computing operations, a measure of how much raw calculation went into building it.
A company is a large frontier developer if its revenue in the previous fiscal year, counting affiliated subsidiaries, passed $500 million. The report names OpenAI and Anthropic among the large providers that must register by 1 January 2027. They must also make their safety frameworks and transparency reports public.
Consumers are exempt from the registration duty. The rules target the companies that build the models, not the people who chat with them.
Who enforces it and what it costs
Oversight sits with the Office of Digital Innovation, Governance, Integrity, and Trust, called DIGIT. It opened the RAISE portal on 10 November 2026. DIGIT can take enforcement action against companies that fail to report or never register.
The penalties are up to $1 million per violation. Repeat violations can bring that to as much as $3 million. For a company with more than $500 million in revenue, those sums are modest, but they come with public scrutiny of safety plans and filings.
New York's wider push
The state is also building its own technical capacity. A network of more than 80 scientists and researchers was presented on 10 November 2026, centred on the New York Center for AI Safety and Responsible Research at Binghamton University. It draws on computing resources from the Empire AI initiative. Greg Morrisett of Cornell Tech and Jeannette Wing of Columbia University are its joint acting leaders.
Its jobs include safety testing of advanced AI systems, basic research, and helping DIGIT analyse the safety plans companies submit. Binghamton received $30 million in private donations for this. Hochul's budget includes $250 million for the testing and research consortium. That is more than half of a $400 million public-private total announced the year before.
Separately, Hochul signed a law on 28 October 2026 aimed at chatbots and young people. From 1 January 2027 it bans twelve companion features deemed unsafe for users under 18, where the operator knows the user is a minor. Examples include faking human emotions, excessive flattery and encouraging self-harm. Only the state attorney general can enforce it, with fines up to $25,000. Another law bans emotional manipulation of teen users from 1 June 2027.
Pushback and open questions
Industry did not welcome the rules. Trade outlet insideaipolicy.com reported clear resistance from the technology sector in an article dated 15 November 2026.
Our source leaves some gaps. It does not define what makes an incident critical, so we cannot say what would trigger the 72-hour clock. It also does not say whether the reports will be public or stay with regulators.
For now, the practical picture is simple. Large AI developers have until 1 January 2027 to register and have a state portal ready to receive incident reports.
What to watch
- How the law defines a critical safety incident, and whether reports will be made public.
- Whether large developers register with the RAISE portal by 1 January 2027.
- How industry pushback shapes enforcement once DIGIT starts acting.
Try it
If your work depends on a large AI vendor, ask them in writing how they will handle incident notices under rules like this, and whether you would hear from them first.
Sources and links
- ReportingBornCity: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by BornCity · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
