New York's RAISE Act Takes Effect: Frontier AI Developers Must Now Register with the State

Frontier AI developers are required to begin registering with New York State this month under the RAISE Act — making New York the most aggressive US state regulator of foundation models, a position it is building with support from more than 1,100 frontier AI employees who signed a public letter backing the law. The registration requirement is the first compliance milestone, with broader safety obligations — including a 72-hour incident reporting mandate — taking effect January 2027.

What the RAISE Act Requires

The Responsible AI Safety and Enforcement (RAISE) Act creates a formal registration regime for developers of frontier AI systems — defined as models trained beyond specified compute thresholds that pose potential risks if misused or misaligned. Key requirements:

• Registration with the state of New York — effective November 2026; covered developers must submit identifying information, model descriptions, and safety documentation
• 72-hour safety incident reporting — takes effect January 1, 2027; any incident where a frontier model causes or nearly causes significant harm must be reported to state authorities within 72 hours
• Ongoing transparency obligations — documentation of training data provenance, capability evaluations, and red-team findings
• No carve-out for research — academic institutions deploying frontier-class models for research are covered unless exempted by rulemaking

Senator Gounardes, the law's primary sponsor, has simultaneously proposed an additional layer of regulations covering AI safety audits, public transparency reports, and whistleblower protections for AI employees who raise safety concerns through state channels.

The Letter Behind the Law

The RAISE Act is unusual in the breadth of industry support it has attracted. A public letter backing Senator Gounardes's broader AI safety framework was signed by more than 1,100 employees from frontier AI companies, including:

• Dario Amodei, CEO of Anthropic
• Chief scientists at OpenAI and Meta

That list — executives and researchers from the very companies being regulated signing a letter asking for more regulation — reflects the split inside the AI industry over whether voluntary commitments are sufficient. The signatories argue that binding safety requirements create a level playing field that prevents companies from racing to skip safety steps to stay competitive.

Why New York and Why Now

The United States has no federal AI safety law for foundation models. The Biden-era executive order created voluntary reporting channels; the Trump administration rescinded most of it. That vacuum has pushed individual states to act, and New York — as a global financial and media hub — has more leverage than most. Companies that cannot operate in New York are companies that struggle to serve Wall Street, the US media industry, and the country's largest school system simultaneously.

The RAISE Act enters force as compliance obligations are already stacking up. According to Hinshaw Law's 2026 compliance tracker, frontier AI developers now face active regulatory obligations across New York, Colorado, California, Connecticut, and the European Union — each with different definitions of "high-risk" and different reporting timelines. New York's combination of registration plus incident reporting is the most operationally demanding of any US state framework.

What "Frontier" Means in Practice

The RAISE Act defines frontier AI by training compute threshold — a proxy for capability that regulators have borrowed from the EU AI Act and the Biden-era executive order. The threshold covers models in the same class as GPT-6, Claude, Gemini 4, and Llama's largest variants. It does not currently cover smaller fine-tuned models, application-layer wrappers, or retrieval-augmented generation systems built on top of frontier APIs.

That definition matters because it shapes who must register. A startup building a legal AI tool on top of the Claude API is not a covered developer. Anthropic, OpenAI, Google DeepMind, Meta, and xAI are.

Impact on AI Companies

Registration itself is largely administrative, but it creates a compliance paper trail that state attorneys general can use in enforcement actions. New York's AG office has broad consumer protection authority and has already issued civil subpoenas in AI-related matters. The 72-hour incident reporting requirement — which kicks in January — is where companies will feel operational pressure. It means they need monitoring systems capable of detecting and triaging safety events on a timeline tighter than most current enterprise incident-response playbooks are designed for.

Legal counsel at covered companies should begin now. The January 2027 deadline for incident reporting is 60 days away. Firms that have not identified their reportable-incident criteria, internal escalation paths, or state notification contacts are behind.

What to Watch

Senator Gounardes's additional proposal — which goes beyond the RAISE Act's current scope — will move through the New York legislature in 2027. If it passes, it would create the country's most comprehensive state-level AI safety regime, with mandatory third-party audits and whistleblower protections that mirror what some EU AI Act provisions require at the federal level in Europe. The outcome will depend partly on how smoothly the RAISE Act registration process goes in November and December.

By Hector Herrera