A Deception Theory Built for a New Target
On July 1, 2026, the Federal Trade Commission published a proposed policy statement, Concerning the Suppression of Accuracy in Artificial Intelligence Systems, and opened a comment window that ran through July 31 (Reference: FTC). The theory is straightforward on its face: AI companies market their chatbots and assistants as accurate and objective, so if a company alters outputs to serve an undisclosed ideological or other objective — including to comply with a state law — that gap between marketing and practice can itself be a Section 5 deceptive act, apart from any judgment about the content of the outputs themselves (Reference: Federal Register).
The statement did not originate with the FTC alone. Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," signed by President Trump on December 11, 2025, gave Chairman Andrew Ferguson 90 days to explain how the deception standard applies to AI models and, specifically, when state laws that force output alterations are preempted by federal law. The order carves out three categories from that preemption push — child-safety rules, AI compute/data-center infrastructure, and states' own procurement and use of AI — leaving most other state AI statutes exposed to the theory.
The Case For It
The steelman here is real. Users form expectations of neutrality when they ask a general-purpose model a factual question, and if a developer quietly tunes outputs to satisfy a patchwork of state mandates without disclosing it, that is a meaningful gap between representation and reality — exactly the asymmetry Section 5 exists to police. A single federal accuracy floor is also, in the abstract, more administrable for a nationally distributed product than fifty different state theories of what "unbiased" output requires. Colorado's Artificial Intelligence Act, the FTC's chief example, imposes liability on developers for discriminatory outcomes their customers cause downstream — a real compliance burden that plausibly pushes companies toward defensive, undisclosed output-shaping (Reference: Inside Privacy).
Where the Theory Strains
But the FTC isn't proposing to police a narrow category of undisclosed bias — it's proposing to become the arbiter of what "accurate" and "objective" AI output looks like, a job with no natural stopping point. The agency's own statement reportedly concedes that objective bias standards may be difficult to draw cleanly, which is a striking admission to build an enforcement theory on. EFF, joined by Public Knowledge and Fight for the Future, filed comments on August 3, 2026 urging the FTC to withdraw the proposal outright, arguing that "the government may not install itself as the arbiter of truth" and warning that a vague, contestable standard invites selective enforcement against whichever companies' outputs displease whoever holds the chair at a given moment (Reference: EFF).
That concern isn't hypothetical given the proposal's lineage. Executive Order 14365 sits alongside the administration's earlier "Preventing Woke AI" order, and the FTC's own framing — steering "toward undisclosed ideological ... objectives" — imports contestable political vocabulary into a consumer-protection statute built for concrete, falsifiable claims like mileage or weight-loss numbers, not open-ended judgments about model neutrality.
The Preemption Reach Is the Bigger Problem
The more consequential move is buried in the theory's implications, not its headline. The FTC's framing treats state laws that require output changes it deems deceptive as impliedly preempted wherever they conflict with the federal deception scheme. That's an aggressive reading of implied-preemption doctrine, asserted through a policy statement rather than a rule, a statute, or a court ruling. Policy statements don't bind courts and don't carry the force of law the way notice-and-comment rules do; using one to signal that a whole category of state consumer-protection and anti-discrimination law is preempted invites exactly the legal challenge that leaves developers with less certainty, not more, while the theory gets litigated.
The Better Path
This publication favors a single, predictable federal AI framework over a fifty-state patchwork. But the vehicle matters. A genuine national standard should come from Congress, where preemption's scope, its exceptions, and its enforcement mechanism can be debated and fixed in statutory text — not asserted by an FTC policy statement invoking authority the agency has never tested in court and cannot bind future commissions to. Ferguson's FTC is right that fifty divergent state AI-output mandates impose real costs on developers. It's wrong to try to fix that cost by having an enforcement agency define, case by case, what counts as ideologically distorted truth. The statement should be narrowed to concrete, verifiable representations — capability claims, safety claims, factual assertions a company makes about its own product — and leave the neutrality debate to the branch of government built to have it.