On 1 July, the Federal Trade Commission issued a proposed policy statement arguing that state laws requiring AI developers to alter model outputs to meet state-defined accuracy or fairness standards may conflict with, and be preempted by, federal law, most directly the FTC Act's Section 5 prohibition on deceptive trade practices. The Commission approved the accompanying Federal Register notice by a 2-0 vote and opened it for public comment through 31 July, according to the FTC's own announcement. The statement, published in full in the Federal Register, names Colorado's Artificial Intelligence Act as the clearest example of the kind of state requirement it has in mind, and Chairman Andrew Ferguson has tied the move to a presidential executive order on the "truthful outputs" of AI models, as reported by law firm Steptoe.
What the statement actually argues
The FTC's theory is that when a company markets an AI system as accurate and objective, but a state law then requires it to alter outputs to satisfy state-defined fairness or ideological-balance criteria, the resulting gap between what was marketed and what the state compels can itself become a deceptive practice under Section 5, one the FTC's own federal enforcement authority is better placed to police than a patchwork of state statutes. The statement is careful to frame this as a concern about "undisclosed" alteration specifically: a developer that discloses how and why it adjusts outputs is in a different position to one that quietly does so while continuing to market the system as objective. That distinction matters more than the headline "FTC vs state AI laws" framing suggests, because it means the statement isn't a blanket objection to state AI regulation, only to a specific mechanism, compelled and undisclosed output alteration, that it argues sits in tension with federal deception law.
It's worth being precise about what this document is and isn't. A policy statement is the FTC signalling how it intends to interpret and enforce its existing authority; it is not a rule with the force of law, it doesn't repeal or suspend Colorado's AI Act or any other state statute, and the preemption argument it makes would ultimately need to be tested and won in federal court, likely over a period of years, to have binding effect. Comments close on 31 July, after which the Commission can finalise, revise, or shelve the statement entirely.
Why this lands awkwardly for enterprises mid-build
The timing is the interesting part for enterprise compliance teams. Colorado's AI Act, Illinois's newly signed SB 315 (which we covered here two days ago), and a growing list of other state AI statutes have driven real compliance build-out over the past year: documentation processes, bias testing regimes, incident reporting workflows, vendor due diligence questionnaires, often built specifically to the letter of a given state's requirements. A federal statement questioning the legal durability of one of those requirements, even a fairly narrow one focused on compelled output alteration, understandably raises the question of whether that build-out effort is about to be wasted.
It almost certainly isn't, for three practical reasons. First, this statement targets one specific mechanism, compelled alteration of outputs, not the broader categories most enterprise AI governance work is actually built around: risk assessment, incident disclosure, vendor transparency and human oversight, none of which this statement questions. Second, preemption is a genuinely difficult legal argument to win outright against a state's police-power regulation, and state attorneys general, including Colorado's, are highly likely to defend their statutes vigorously if this moves from a policy statement toward actual enforcement action or litigation. Third, even organisations confident the FTC's position will eventually prevail have no reliable timeline for when that might happen, and building a compliance programme around a legal outcome that hasn't occurred yet is a weak substitute for building one around the law as it currently stands.
The sensible position to hold this month
Treat the FTC's statement as a genuine signal worth tracking, not a green light to pause state-level AI compliance work. Governance and legal teams should read the statement, note the specific mechanism it targets, and flag any programme elements that rely narrowly on compelled-output-alteration requirements as ones to revisit if the FTC's position hardens. Everything else in a well-built AI governance programme, the parts aligned with the wider divergence between US and international AI regulation we've written about separately, remains exactly as necessary as it was before 1 July.
- Read the FTC's proposed statement and identify which, if any, of your current state AI compliance obligations rely specifically on compelled alteration of model outputs, as distinct from disclosure, testing or incident reporting requirements.
- Continue building to the strictest state standard your organisation is exposed to; a policy statement in a comment period is not grounds to relax an active compliance programme.
- If your organisation wants to weigh in, comments are open on Regulations.gov under Docket FTC-2026-0859 until 31 July 2026.
- Ask legal counsel to flag when, and if, this statement moves from proposal to enforcement action or litigation, since that transition is the point at which practical risk actually changes.
- Keep AI output disclosure practices, explaining when and why a system's outputs are adjusted, in place regardless of the preemption debate, since disclosure is the distinction the FTC's own theory turns on.
Federal and state AI regulation are heading in different directions again, and this is unlikely to be the last time a federal body questions a state requirement your compliance team has already built around. If you'd like help keeping your AI governance programme resilient to that kind of regulatory back-and-forth, email sales@halfteck.com.