New York's RAISE Act Puts Frontier AI Developers on Registration Clock Starting Now

New York's RAISE Act has put frontier AI developers on a registration clock that starts this month. Under the law, companies developing advanced AI systems must begin registering with New York State in November 2026, ahead of full compliance requirements — including mandatory 72-hour safety incident reporting — taking effect in January 2027. New York becomes the first [U.S. state](/government/state-ai-laws-25-enacted-june-deadline) to impose pre-deployment registration requirements on frontier AI developers, setting a precedent other states are likely to follow.

Context: The Regulatory Squeeze Is Tightening From Both Sides

The timing is not accidental. The EU AI Act's transitional period for high-risk AI systems ends December 2, 2026 — just weeks after New York's registration window opens. Global AI companies are now absorbing dual compliance pressure from opposite sides of the Atlantic simultaneously, with no federal U.S. framework to simplify the patchwork.

This follows a wave of state-level AI legislation that has accelerated through 2026. Colorado enacted consumer protection requirements. Connecticut's AI law took effect October 1. California passed a series of sector-specific bills. But the RAISE Act is structurally different from all of them: it targets the development stage, not deployment outcomes. It creates a registry of who is building frontier AI before those systems reach consumers or enterprises.

That is a meaningful shift in regulatory philosophy — from governing AI effects after the fact to tracking AI capabilities before they ship.

What the RAISE Act Requires

The registration requirements apply to developers of frontier AI — defined under the Act as AI systems meeting thresholds for compute, capability benchmarks, or both. The specific thresholds are set by regulation rather than hardcoded in statute, giving New York flexibility to adjust the definition as the technology evolves.

Key requirements:

• November 2026: Registration period opens. Covered developers must file with the designated state authority.
• January 2027: Full compliance deadline. All registered developers must have safety incident reporting infrastructure operational.
• 72-hour reporting window: Any safety incident — including jailbreaks enabling dangerous outputs, significant autonomous behavior violations, and material misuse events — must be reported to state authorities within 72 hours of detection.
• Pre-deployment documentation: Developers must maintain and submit records of safety testing conducted before deploying covered systems.

Who Is Affected

The RAISE Act's thresholds are calibrated to capture the frontier model developers — OpenAI, Anthropic, Google DeepMind, Meta AI, and companies at that capability tier — rather than the far larger population of companies fine-tuning or deploying existing models. Organizations building exclusively on top of registered models, without training their own frontier systems, have reduced obligations under current draft regulations.

That said, legal teams across the AI industry are actively analyzing where their training runs fall relative to the compute threshold. For companies near the boundary, the compliance determination is not straightforward.

Why This Matters

For AI developers: New York is not a jurisdiction they can treat as peripheral. Any company deploying AI systems accessed by New York residents — which describes virtually every major AI product in the United States — is within scope. Unlike California, which developers have occasionally gamed by limiting explicit state-level sales, New York's population density and economic activity make avoidance implausible.

For other states: The RAISE Act provides a replicable template. State legislators watching New York's implementation will use it as a model for their own pre-deployment registration programs. California, already the most active state AI regulator, is expected to follow with analogous requirements if New York's program survives legal challenge.

For federal policymakers: A growing patchwork of state-level frontier AI registries creates compliance complexity that makes a federal preemption framework more attractive to the industry. That is a dynamic New York may be deliberately engineering: force federal action by making state fragmentation painful enough that industry lobbies for a unified national standard.

For global AI governance: New York's registration requirement arrives as the EU's AI Act approaches full enforcement. Companies building compliance programs for January 2027 now need to integrate dual requirements simultaneously.

What to Watch

Whether AI industry groups file for injunctive relief before the January 2027 full-compliance deadline — a legal challenge that would pause enforcement while constitutional questions are litigated. The argument would likely center on the dormant Commerce Clause, asserting that state-level regulation of interstate technology development is preempted by the federal government's exclusive authority over interstate commerce. How that argument plays out in federal court will determine whether the RAISE Act becomes a durable governance model or a legal cautionary tale.