On July 7 the FTC published a proposed policy statement in the Federal Register recasting undisclosed AI output-steering as deception under Section 5, and asserting that conflicting state AI laws are impliedly preempted. The commission authorized the notice on a 2-0 vote. Comments close July 31.

The document, “Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems,” concedes that the FTC Act doesn’t expressly preempt state law, then invokes implied preemption “to the extent” a state regime conflicts with the federal scheme. It singles out Colorado’s Artificial Intelligence Act, which it says “appears to coerce” developers into altering outputs to avoid disparate-impact liability. Hallucinations traceable to technological limits are carved out. Intentional steering for ideology, political pressure, or compliance with state law isn’t.

The timing is the story. One day earlier, on July 6, Governor JB Pritzker signed Illinois’ frontier AI law, which passed the Illinois House unanimously and was backed by OpenAI and Anthropic. It mandates annual audits, risk frameworks, and 72-hour incident reporting, with a 24-hour window for imminent risk of death. It takes effect January 1, 2028.

The statement traces its authority to Executive Order 14365, signed by President Trump on December 11, 2025, which called for a “minimally burdensome national policy framework for AI—not 50 discordant State ones.” Outside counsel analyses read the notice as offering a safe harbor: “clear, conspicuous, and adequate” disclosures that a system prioritizes objectives other than what users expect can defuse Section 5 exposure. Fine print won’t do it.

Deception doctrine, historically a consumer-protection tool against snake-oil labeling, is being repurposed as a federalism instrument. The vehicle for shrinking state authority is a disclosure regime.

Sources

Sources