The Federal Trade Commission voted 2-0 on July 1 to propose a policy statement declaring that AI companies which steer model outputs toward undisclosed objectives may violate Section 5 of the FTC Act, and it named Colorado’s Artificial Intelligence Act as the first casualty. The document, published in the Federal Register on July 7 as 91 FR 41638, argues Colorado’s statute is “impliedly preempted to the extent it conflicts with a federal regulatory scheme.”

The theory is doctrinally clean and strategically aggressive. Section 5 deception has always covered affirmative misstatements, material omissions, and inadequate disclosures. The Commission is extending that logic to the internal objectives of an AI model itself: if a system is tuned toward goals a consumer wouldn’t expect, and the tuning isn’t disclosed clearly and conspicuously, that’s deception. Fine-print terms of service won’t cure it.

Colorado is the intended test case. The FTC contends the state’s AI Act “appears to coerce companies into altering the output of their AI models” to satisfy disparate-impact obligations, meaning compliance with Colorado becomes, in the Commission’s framing, a Section 5 violation. Suppressing accuracy to avoid state liability is itself deceptive.

The action traces to Executive Order 14365, signed December 11, 2025, which directed the FTC to confront state laws requiring alterations to “truthful outputs of AI models.” Chairman Andrew Ferguson framed the proposal as “expanding America’s global dominance in artificial intelligence.”

Comments close July 31 at docket FTC-2026-0859. The finalized statement will almost certainly become the vehicle for litigating preemption directly. It’s a familiar move: reclassify a state civil-rights framework as a federal consumer-protection problem, and let the courts sort the hierarchy.

Sources

Sources