New York Enforces Landmark AI Safety Law on OpenAI and Anthropic

The New York State Department of Financial Services (DFS) opened its registration portal on Tuesday, November 10, marking the practical start of implementation for one of the most comprehensive artificial intelligence regulatory regimes ever adopted anywhere in the world. Frontier AI developers including OpenAI and Anthropic — the companies behind ChatGPT and Claude — now have until January 1, 2027 to register with the state and comply with a detailed set of obligations ranging from safety framework publication to mandatory incident notification within 72 hours.

What the RAISE Act requires

The RAISE Act (Responsible AI Safety and Education Act), signed by Governor Kathy Hochul on March 27, 2026, following a protracted legislative process, applies to models classified as "frontier" — namely, foundation models trained with computing power exceeding 10²⁶ integer or floating-point operations — and to developers with more than US$ 500 million in annual revenue. For these companies, the law establishes three main pillars.

The first pillar is the mandatory publication of safety and transparency frameworks on the companies' own websites. These frameworks must include risk evaluation practices, harm mitigation procedures, internal governance for implementation, and a critical incident identification and response process. They must be updated annually or whenever there is a material modification to the model.

The second pillar is transparency reports. Before launching any new model or modified version of an existing model, developers must publish a report on their website containing contact information, release date, supported languages, output modalities, intended use, use restrictions, summaries of catastrophic risk assessments, and the degree of involvement of independent evaluators. This requirement draws inspiration from California's SB 53, but with significantly tighter timelines.

The third pillar — and the one generating the most practical impact — involves reporting requirements. The law requires developers to transmit quarterly summaries of any catastrophic risk assessment resulting from internal use of their models, to report critical safety incidents to the new DIGIT office within 72 hours, and, in cases of imminent risk of death or serious injury, to notify public safety agencies within 24 hours. Penalties reach US$ 1 million per initial violation and up to US$ 3 million for subsequent violations.

A national laboratory

Governor Hochul went beyond regulation alone. During the same November 10 event, she announced the launch of the New York Center for AI Responsibility and Research, housed at the State University of New York in Binghamton. The center received US$ 55 million in donations and capital and will independently evaluate the safety, cybersecurity, and operation of advanced AI systems, working alongside the DIGIT office to review companies' safety frameworks. Marc Gilman, the first full-time DIGIT hire, was appointed deputy director of the RAISE program.

The law also builds on two other New York initiatives already in force or in implementation. California's SB 53, effective since January 2026, requires safety frameworks and whistleblower protections. And in October 2026, Hochul signed a separate law requiring chatbots targeting minors to disclose they are AI and actively monitor for risky behaviors such as self-harm and eating disorders — that law takes effect June 1, 2027.

Federal context and comparison

AI regulation in the United States is advancing on fragmented terrain. While the Trump administration promotes comprehensive federal legislation, states are creating their own regimes. California SB 53, in effect since January 2026, requires labs to publish safety frameworks and report critical incidents within 15 days — significantly longer than New York's 72-hour deadline. The RAISE, by contrast, imposes tighter timelines but does not require whistleblower protections at the same level as SB 53.

This mosaic of state laws is precisely what the RAISE Act sought to avoid in terms of fragmentation, yet it is also what makes New York a laboratory: if the law works, other states will follow; if it creates excessive friction with innovation, it serves as a warning.

What to expect

The practical effect is what matters most. OpenAI and Anthropic operate globally — New York is an important market but not the only one. Penalties of up to US$ 3 million sound significant, but are negligible for companies moving billions per year. The real pressure will come from compliance costs: hiring dedicated teams, auditing models, maintaining updated frameworks. And from reputational risk when incidents are reported publicly on the state's portal.

The RAISE is above all a signal. It shows that AI regulation is no longer a theoretical discussion at conferences — it is a legal requirement with deadlines, fines, and oversight bodies. If the federal administration fails to create a coherent framework, states like New York and California will keep stacking rules, and AI companies will have to navigate an ever-more detailed maze of requirements.

Sources: Yahoo News / TechCrunch, NY DFS, Wiley Rein LLP

✓ Independent sources cross-checked and verified before publishing