On August 3, 2026, a coalition of digital rights groups — the Electronic Frontier Foundation, Public Knowledge, and Fight for the Future — filed joint comments urging the Federal Trade Commission to withdraw its proposed policy statement "Concerning the Suppression of Accuracy in Artificial Intelligence Systems." The FTC released the nine-page proposal on July 1, 2026, on a 2-0 vote, and closed public comment on July 31. The coalition's filing, and EFF's accompanying post, argue the agency has drafted itself a mandate no statute gives it: deciding which AI outputs are "accurate" enough to escape federal enforcement, and which state laws must yield to that judgment.
What the FTC Is Actually Proposing
The policy statement traces to Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," signed December 11, 2025. That order directed the FTC chair to explain, within 90 days, when state laws requiring changes to AI outputs are preempted by the FTC Act's ban on deceptive practices — extending a concern the administration first raised in its July 2025 "Preventing Woke AI in the Federal Government" order, which barred federal procurement of models that sacrifice "truthfulness and accuracy to ideological agendas." The FTC's answer is that steering a model's output away from what it calls the "most accurate answer," without disclosing that to users, can itself be a Section 5 deceptive act — and that state laws pushing companies toward such steering are therefore federally preempted.
The statement names Colorado directly. It criticizes the original Colorado Artificial Intelligence Act (SB 24-205) for allegedly incentivizing companies to quietly tune outputs toward "equitable" results to avoid discrimination liability. Colorado repealed and replaced that law with SB 26-189, a narrower transparency-and-disclosure framework for automated decision-making in employment, housing, credit, insurance and government benefits, signed May 14, 2026 — six weeks before the FTC's statement. The FTC's response, per legal analysis from Arnold & Porter, was that the narrowed law "poses many of the same concerns" and may still be preempted — leaving Colorado's law in force from January 1, 2027 under a cloud of contested federal authority.
The Case for the FTC's Concern
The underlying worry isn't fabricated. If a chatbot markets itself as neutral and factual while its developer quietly weights outputs toward a preferred narrative — political, commercial, or otherwise — undisclosed, that is a plausible consumer-protection harm resembling the kind of deceptive-practices cases the FTC has brought for decades over hidden defaults and dark patterns. A single, predictable federal standard also has real appeal for companies operating across 50 states, each potentially imposing its own definition of algorithmic fairness or accuracy. Regulatory fragmentation is a genuine compliance cost, and Congress has never passed a comprehensive federal AI statute to preempt it cleanly.
Why the Proposal Still Overreaches
The coalition's objection is not that AI accuracy is irrelevant to consumer protection — it's that the FTC has designed a mechanism far broader than that narrow harm. As EFF put it, "the government may not install itself as the arbiter of truth," and doing so here amounts to "a prior restraint on speech" because the agency would decide, case by case, which viewpoint-inflected outputs count as impermissible "suppression of accuracy." That is a strikingly different posture than policing hidden defaults in a subscription flow: it requires the FTC to adjudicate contested questions — about historical framing, social categories, or contested facts — that have no single "accurate" answer the government can certify.
The preemption claim compounds the problem. The FTC Act's deception provision was never written with AI model outputs, let alone state algorithmic-discrimination statutes, in mind. Federal agencies do not acquire implied preemption power merely by asserting that state laws create the same incentives their own theory condemns; that requires either explicit congressional delegation or a genuine, demonstrated conflict — not an executive order instructing an independent agency to find one. Colorado's SB 26-189 addresses documented harms in high-stakes automated decisions about jobs, housing and credit; treating disclosure-focused state consumer protection as presumptively suspect, days after the state itself narrowed the law to address federal concerns, suggests the goal is less about deception than about foreclosing state experimentation entirely.
There's also a chilling-effect problem the coalition is right to flag: once a federal agency claims authority to judge whether an AI system's outputs are ideologically "suppressed," companies have every incentive to tune models toward whatever the sitting administration currently considers neutral — the opposite of open, contestable discourse the First Amendment is meant to protect. An accuracy standard that shifts with each administration is not a standard; it's discretion dressed up as consumer protection.
The Better Path
The FTC already has tools to address genuine deception in AI marketing — misrepresenting capabilities, hiding material limitations, or concealing that a system optimizes for engagement over accuracy — without a freestanding "suppression of accuracy" doctrine that reaches into contested judgment calls and claims to override state legislatures. If Congress wants a uniform national AI framework, that is a legislative choice, not one an agency should engineer through a policy statement following an executive order. The FTC should narrow its final statement to concrete, provable deception and drop the preemption claim against states actively working — as Colorado has — to calibrate their own rules.