The Federal Trade Commission on July 7 published a proposed policy statement, 91 FR 41638, arguing that AI systems secretly steered toward undisclosed objectives violate Section 5 of the FTC Act. It also announces that Colorado’s Artificial Intelligence Act is “impliedly preempted to the extent it conflicts with a federal regulatory scheme.”

The Commission cleared the Federal Register notice on a 2-0 vote, acting under Executive Order 14365, which President Trump signed December 11, 2025 directing the FTC to address state laws that force alteration of AI outputs. File No. P264200 is now the vehicle by which federal consumer-protection doctrine gets pointed at a state statute.

The doctrinal move is elegant, if aggressive. The statement adapts the FTC’s three-part deception test: undisclosed steering of outputs is potentially actionable; hallucinations arising from technical limits aren’t; and liability can be cured with disclosures that are “clear, conspicuous, and adequate.” The materiality prong leans on a striking claim from the Federal Register text itself, that consumers accept AI outputs without independent fact-checking more than 90% of the time.

Colorado is the named target. Its AI Act, the statement says, “appears to coerce” companies into altering outputs to advance state “ideological objectives.” Translation: a state that tells a model what to say is, in the Commission’s read, making the model lie to users, and Section 5 gets there first.

This is preemption-by-policy-statement, arriving before any court has weighed in. Comments are due July 31, 2026 at regulations.gov, Docket FTC-2026-0859. The doctrinal fight, though, won’t be settled in the docket. It’ll be settled wherever Colorado decides to defend its statute.

Sources

Sources