The FTC's AI Accuracy Statement Is a Federal Preemption Weapon Aimed at State AI Governance Laws
The FTC's proposed policy statement makes state-mandated AI fairness measures — bias mitigation, disparate-impact avoidance — legally indistinguishable from deceptive output steering under § 5 of the FTC Act. Until a disclosure safe harbor is defined after July 31, enterprises simultaneously subject to Colorado-style mandates and FTC scrutiny have no fully compliant path.
What the FTC Actually Said
On July 1, the FTC published a proposed policy statement addressing whether AI companies that steer outputs toward undisclosed ideological objectives may be engaging in deceptive acts or practices under § 5 of the FTC Act. The title is revealing: Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems.
The most consequential sentence may be this: “These prohibitions apply even when a company engages in a deceptive act or practice in order to comply with a State law. Although the FTC Act does not expressly preempt State law, State law is impliedly preempted to the extent it conflicts with a Federal regulatory scheme.” The FTC is telling enterprise compliance teams that following state AI fairness law may itself be a federal offense.
Motivation is legally irrelevant under this framing — a company doing the right thing for the wrong-framed reason is still liable. The statement carves out AI “hallucinations” as arising from limitations rather than design decisions. That carve-out is narrower than it looks: deliberate post-training alignment, RLHF tuning, and bias-mitigation layers are all design decisions.
Colorado Is Named — But It Won’t Be the Last
Colorado’s AI Act “appears to coerce companies into altering the output of their AI models to comply with and advance the state’s ideological objectives,” and is “impliedly preempted to the extent it conflicts with a federal regulatory scheme.” As covered in Colorado’s AI Governance Retreat Didn’t End the Story, Colorado had already retreated to a lighter transparency regime under SB 26-189 — yet the FTC still named that revised statute by section number. The preemption theory doesn’t require a strong state law to bite; it bites whatever law pressures output changes.
The Illinois Human Rights Act and California’s FEHA (recently amended to cover AI) face the same logical exposure.
The Compliance Trap — Illustrated
| Obligation | Required Action | FTC Risk |
|---|---|---|
| Colorado SB 26-189 § 6-1-1707 | Avoid AI outputs causing disparate impact | Undisclosed output steering → § 5 deception |
| Fair lending / ECOA | Model bias mitigation in underwriting | Same theory; disparate-impact EO cited in footnote |
| Internal AI safety policy | RLHF, content filters, refusal tuning | Permissible only with sufficiently prominent disclosure |
A company may prioritize objectives other than pure correctness — but only with clear and conspicuous disclosure. The statement rules out burying disclaimers in terms of service. No template for what would suffice is provided. That undefined safe harbor is the governance gap that matters.
The Preemption Theory and Its Limits
The statement extends traditional deception principles to AI model design while advancing a conflict-preemption theory that could substantially affect emerging state AI laws. The historical parallel is Geier v. American Honda (2000), where the Supreme Court found implied preemption from a federal safety standard without an express preemption clause. The FTC attempts a structurally similar move: asserting that § 5’s purpose overrides state law compelling the very conduct § 5 prohibits. Courts have not always been receptive to agency-asserted implied preemption absent congressional intent — this will be litigated.
A 2-0 Commission vote — in a body whose independence is now doubtful following a Supreme Court decision holding the president can fire commissioners at will — signals this is political as much as legal.
What to Watch
The comment period closes July 31. The docket at Regulations.gov (FTC-2026-0859) is where the real action happens: AI developers, civil rights organizations, and state attorneys general will submit conflicting records shaping whether the FTC finalizes or withdraws the preemption language.
My read: The FTC will finalize some version before Q4 2026 — but courts will strip the implied preemption claim on first challenge unless Congress acts. The deeper lasting effect is the disclosure standard: even if preemption fails, enterprises that cannot document why their AI outputs are configured as they are face a credible § 5 deception theory on the merits. Model risk teams should treat this as a documentation and disclosure problem today, regardless of legal outcome. For context on how SR 26-2’s model risk guidance already left a structured governance gap, see SR 26-2’s GenAI Carve-Out Creates a Structured Governance Gap.