OpenAI, Anthropic face new oversight as New York’s AI safety law takes effect

Until now, frontier AI safety commitments in the US have mostly been voluntary or untested. New York opened a portal Tuesday where large frontier developers such as OpenAI and Anthropic must register by January 1, 2027. The RAISE Act, which applies to companies with more than $500 million in annual revenue, then requires them to publish safety frameworks, report critical safety incidents within 72 hours, and issue public model transparency reports.

A new state office, the Office of Digital Innovation, Governance, Integrity, and Trust, or DIGIT, will enforce it, with penalties up to $1 million for a first violation and $3 million for subsequent ones. Paired with California's SB 53 and Illinois' SB 315, this is starting to look like a de facto national floor written by states.

We think this is a reasonable law, and the most important line in it is the clock. California's SB 53 gives labs 15 days to report a critical safety incident. New York gives them 72 hours. That's a real difference. Three days mirrors the breach-notification window Europe uses for personal data, and it forces labs to have incident response that actually works, not just a policy document. If a model does something genuinely dangerous, regulators and the public hear about it while it still matters.

We're also encouraged that New York isn't only collecting paperwork. Governor Kathy Hochul launched the New York Center for AI Responsibility and Research at SUNY Binghamton, which will independently evaluate advanced systems and help DIGIT review companies' safety frameworks. She first announced it in January with $55 million in donations and capital. A regulator with its own technical evaluators is much harder to snow than one reading self-reported PDFs.

Now the honest limits. The fines are small next to the companies they cover. A $1 million penalty barely registers for labs raising and spending at the scale OpenAI and Anthropic do. The real leverage is reputational and procedural: public transparency reports, mandatory disclosure, and the threat of enforcement for not registering, though a spokesperson declined to say what that would look like. We'll know the law's strength by its first enforcement action, not its press briefing.

The politics are worth naming too. Hochul framed this as standing against tech moguls who treat collateral damage as acceptable in a race toward what she called "the AI cliff." That's strong rhetoric, and it signals New York intends to be a counterweight, not a cheerleader. We'd push back gently on the framing. Plenty of AI work is genuinely useful, and at least one of the labs this law covers has built its brand on safety research. Good regulation should make that work more trustworthy, not treat it as the enemy.

Where this leads if it scales: a patchwork. New York, California and Illinois now have overlapping but different rules, with different deadlines and, in Illinois' case, independent compliance audits. Large labs can absorb that. The risk is that each new state adds its own clock and form, and compliance becomes the product. The better outcome is that the strictest reasonable standard, probably something like New York's 72-hour rule plus Illinois-style audits, becomes the norm everyone builds to.

New York already enforces a law requiring AI companion operators to remind users they're talking to a bot and to make reasonable efforts to detect suicidal ideation and self-harm. In October, Hochul signed a separate chatbot safety law for minors restricting deceptive, manipulative or sycophantic behavior, which takes effect June 1, 2027. California's SB 53 took effect in January and also requires whistleblower protections; Illinois' SB 315 takes effect at the start of 2027.