// blog · analysis · policy2026-05-255 min read

Three states and a Senate — the frontier AI patchwork locks in despite federal preemption pressure

California SB 53, New York RAISE Act, and now Illinois SB 315 (52-5 Senate passage May 22). Three states with binding frontier-AI transparency laws covering ~30% of US tech employment. The White House National Policy Framework wants federal preemption. The patchwork is locking in faster than the federal response can match.

For most of 2024-2025 the frontier-AI policy conversation was speculative: which state might pass which framework, whether the federal government would preempt, what the EU AI Act would look like after the omnibus negotiations. May 2026 is the cycle when the speculation resolved into facts on the ground.

California has SB 53 (Transparency in Frontier Artificial Intelligence Act). New York has the RAISE Act (Responsible AI Safety and Education Act). Illinois SB 315 just passed the state Senate 52-5 on May 22, advancing toward the same regulatory structure. Three states with binding frontier-AI laws, covering approximately 30% of US tech-industry employment and a much larger share of headquartered AI companies. The patchwork is real and growing.

What each framework requires

The three state frameworks share a structural shape but differ in specifics. All three require frontier-model developers (Meta, OpenAI, Anthropic, Google, plus the next tier of well-funded labs) to adopt transparency frameworks, submit to third-party audits, and report catastrophic-risk capabilities. New York's RAISE Act is the strongest on the disclosure dimension — narrower trade-secret exemptions, dedicated state-level enforcement office, faster enforcement velocity than California's AG-routed approach.

California SB 53 was the first to ship and has the most procedural maturity. Illinois SB 315 is explicitly modeled on the New York approach with refinements. Each framework adds incremental requirements that the others lack — California requires risk-mitigation documentation, New York requires safety-protocol publication, Illinois adds an explicit catastrophic-risk-reporting threshold.

The federal preemption question

The White House released its National Policy Framework for AI on March 20, explicitly urging Congress to replace the state-law patchwork with uniform federal regulation. The framework recommends preempting state AI laws that impose "undue burdens" while preserving state police powers, zoning authority, and rules governing states' own use of AI.

The political reality is that federal preemption legislation is hard to pass even when the executive branch favors it. The Trump administration's pivot toward pre-release evaluation requirements for frontier models (a meaningful shift from its early 2025 opposition to AI oversight) creates some bipartisan space for federal legislation, but Congressional action on AI regulation has moved slowly through 2026. The most plausible scenario is that the state patchwork continues to grow — Texas, Washington, Massachusetts all have draft frameworks in various legislative stages — and federal preemption happens (if at all) in 2027-2028 timeframe, with the existing state regimes operating in the interim.

The compliance cost compounds

For frontier-AI vendors operating in the US market, three-state compliance plus pending federal pre-release-evaluation requirements plus the EU AI Act enforcement (Q1 2026 €250M in fines) plus the EU omnibus updates (covered this morning) creates a meaningful regulatory burden. The cost-of-compliance calculation now favors:

The cost falls disproportionately on the mid-tier closed-frontier vendors. Cohere, AI21, Mistral-when-it-exceeds-the-SME-threshold — these face the same compliance burden as the largest vendors without the same revenue base to absorb the cost. The structural consequence is consolidation: mid-tier vendors are pressured to merge, acquire, or be acquired to spread compliance overhead across larger revenue bases.

The competitive moat that regulation creates

For Anthropic, OpenAI, and Google specifically, the regulatory patchwork is unpleasant but creates a real competitive moat against startup entrants. A new frontier-lab spinning up in 2026 has to comply with three state frameworks plus federal pre-release-evaluation expectations from Day 1 — which means compliance lawyers, third-party auditors, transparency-documentation infrastructure, all as fixed costs before the first model ships. That's a meaningful barrier to entry that didn't exist when Anthropic itself launched in 2021. The incumbents benefit from the regulation they nominally oppose; the structural moat compounds with their existing capability and capital advantages.

WCBU Peoria — Bill regulating powerful AI models advances Illinois → · New York State — Governor Hochul Signs Nation-Leading Legislation AI Frontier → · FPF — California's SB 53 First Frontier AI Law Explained →