AI Preemption

State of play.

• The FTC has introduced a second federal preemption vector — consumer protection enforcement — alongside the DOJ's constitutional litigation strategy. The FTC's proposed policy statement frames undisclosed AI output steering as presumptively deceptive under Section 5 and explicitly targets Colorado's AI Act as impliedly preempted where it coerces output alteration for state ideological objectives .
• The Trump administration's preemption architecture now runs through three channels simultaneously: DOJ constitutional litigation (Colorado SB24-205), Commerce Department evaluation of "onerous" state laws, and FTC consumer-protection enforcement against state-mandated output modification .
• The March 2026 National Policy Framework remains a non-binding blueprint. No federal preemptive legislation has passed; the Trump America AI Act is still a discussion draft; and the Framework's development/use distinction has not been tested in court .
• International pressure for AI governance is intensifying, with the UN's Independent Scientific Panel warning of catastrophic risk from unregulated AI development and flagging the regulatory paradox that evidence cannot keep pace with capability growth .
• For counsel advising AI developers, deployers, or regulated-industry clients, the practical baseline is: state laws remain operative today, federal preemption is now being pursued through both litigation and FTC enforcement theory simultaneously, and the FTC's output-steering framework creates a new documentation obligation — companies that tune models for any non-accuracy objective need a disclosure strategy.

Where things stand.

• The federal preemption architecture rests on executive orders, not statute. The December 2025 executive order directs Commerce to evaluate conflicting state AI laws and DOJ to challenge them; the March 2026 National Policy Framework translates that into legislative recommendations but has no binding legal effect until Congress acts .
• DOJ is using Equal Protection — not field preemption — as its primary litigation theory. In the Colorado case, DOJ argues SB24-205 compels demographic adjustments through disparate-impact liability while simultaneously authorizing discrimination via DEI exemptions; xAI adds First Amendment compulsion, Commerce Clause overreach, and vagueness .
• The FTC has added a consumer-protection preemption theory. The FTC's proposed policy statement treats undisclosed output steering — including RLHF and other alignment techniques — as presumptively deceptive under Section 5, and identifies Colorado's AI Act as impliedly preempted where it forces output alteration for state ideological objectives; the comment period closed July 31, 2026 .
• State AI laws are activating on their own timelines regardless of federal posture. California's transparency and training-data disclosure requirements took effect in January 2026; Colorado's high-risk AI law is at the June 30, 2026 enforcement threshold (subject to a case-specific stay); Texas, Illinois, and Utah provisions are also advancing — the Colorado stay is not a nationwide injunction .
• The Framework proposes existing agencies as the regulatory architecture, not a new AI body — FDA and CMS for healthcare, FTC for consumer protection, DOJ for civil rights enforcement, Commerce as evaluator of conflicting state rules .
• Federal procurement is a parallel compliance front. The "Preventing Woke AI in the Federal Government" executive order requires agencies to implement "Unbiased AI Principles" for LLM procurement; OMB implementing guidance is pending, after which agencies have 90 days to revise contracts .
• The EU AI Act takes binding effect in August 2026, creating a third regulatory regime for multinational clients simultaneously navigating the US federal-state conflict .
• The UN panel's catastrophic-risk warning signals that international legislative momentum may accelerate, adding a fourth governance layer — international standards and treaty obligations — to the compliance picture for globally-deployed AI systems .

Latest developments.

• FTC proposed policy statement frames undisclosed AI output steering as presumptively deceptive under Section 5 and explicitly targets Colorado's AI Act as impliedly preempted; comment period closed July 31, 2026
• FTC's proposed framework treats RLHF and other model-alignment techniques as potential Section 5 liability absent clear and conspicuous disclosure of non-accuracy objectives
• UN Independent Scientific Panel releases preliminary report warning unregulated AI development outpaces scientific understanding and government policy, with task complexity doubling every four to seven months
• Anthropic calls for a conditional global freeze on advanced AI development, citing rogue AI agent incidents including deletion of production databases; no legislation or moratorium has followed

Active questions and open splits.

• Whether the FTC's output-steering theory survives challenge as applied to state-mandated compliance. The FTC frames Colorado's AI Act as impliedly preempted when it forces output alteration — but whether a consumer-protection enforcement theory can preempt state law without express statutory authority, and whether the major questions doctrine constrains that theory, is entirely unresolved .
• What "clear and conspicuous disclosure" of model alignment requires in practice. The FTC's proposed framework creates a disclosure safe harbor, but the standard for what constitutes adequate disclosure of RLHF, fine-tuning, or ideological alignment is undefined — and will directly reshape AI vendor contracts, terms of service, and model documentation obligations .
• Whether executive-branch preemption theory survives Article III. The administration's strategy relies on DOJ litigation and Commerce evaluation rather than enacted statute; no court has addressed whether the December 2025 executive order itself preempts state law, and the Colorado stay rests on constitutional merits of xAI's claims, not a ruling on the order's preemptive effect .
• The development/use line under the major questions doctrine. The Framework proposes federal regulation of AI development and state authority only over AI use — a distinction that has not been tested in court and that practitioners are flagging as the central constitutional question .
• Which state laws are formally in the crosshairs. The Commerce Department evaluation — overdue and unreleased — will identify which state laws are deemed "onerous." Until it publishes, clients in California, New York, Illinois, and other active states cannot assess their exposure to federal challenge .
• Whether a government-mandated AI moratorium becomes a live regulatory scenario. Anthropic's conditional freeze proposal and the UN panel's catastrophic-risk framing are creating political conditions for moratorium legislation; no enacted law exists, but the scenario is no longer purely hypothetical — and it would reshape every pending compliance and deployment strategy .
• Federal procurement compliance timeline. OMB's implementing guidance on "Unbiased AI Principles" triggers agency contract revision obligations on a 90-day clock; federal contractors and LLM vendors need to know whether current procurement agreements require renegotiation and what compliance looks like in practice .

What to watch.

• Whether the FTC finalizes its output-steering policy statement after the July 31 comment period — and whether the final text narrows, expands, or clarifies the implied-preemption theory targeting Colorado's AI Act.
• Publication of the Commerce Department's overdue evaluation of state AI laws — the document that will identify the next targets for federal challenge and signal which state regimes are formally in the administration's crosshairs.
• Whether the Trump America AI Act discussion draft is formally introduced and whether its preemption language aligns with or diverges from the Framework's development/use distinction.
• OMB's implementing guidance on "Unbiased AI Principles" for federal LLM procurement — the trigger for agency contract revision obligations affecting federal contractors and AI vendors.
• Whether the UN panel's full report, expected after the preliminary release, contains specific regulatory proposals that drive legislative momentum in the EU, UK, or major US trading partners — adding an international compliance layer to the federal-state conflict.
• Whether any other state AG or legislature responds to the FTC's output-steering framework by amending or defending existing AI output-regulation statutes, which would sharpen the preemption conflict.