The FTC opened a public comment docket on July 1, 2026 for a proposed policy statement, “Suppression of Accuracy in Artificial Intelligence Systems,” that asserts Section 5 authority over how AI companies tune their models and identifies Colorado’s newly enacted AI bias law as impliedly preempted. The commission voted 2-0. Comments close July 31.

The statement, issued pursuant to Executive Order 14365 (signed December 11, 2025), argues that AI systems are implicitly marketed as maximally accurate, and that consumers accept AI outputs without fact-checking more than 90% of the time. Steering a model toward undisclosed objectives, on that theory, becomes a deceptive practice. Chairman Andrew N. Ferguson said the agency “wants to hear from businesses and consumers about their experiences and concerns regarding the subversion of AI systems for ideological ends.” Hallucinations, notably, are carved out: the statement draws a line between technological limits and intentional steering.

The Colorado piece is where the deference ends. SB 26-189, enacted May 14, 2026, is named directly. The FTC characterizes the law’s effect as pressuring developers to “suppress accuracy and interpose other objectives, such as so-called equity, to avoid liability,” and declares it “impliedly preempted to the extent it conflicts with a federal regulatory scheme.” Compliance with Colorado, the statement adds, is no defense to a federal deception claim.

There’s an escape hatch. Firms can disclose “clear, conspicuous, and adequate” notice that a model prioritizes objectives other than accuracy, and step outside Section 5’s reach.

None of this is a rule. It creates no new obligations on its own. What it does is stake a jurisdictional claim: model tuning is federal turf, and the state laws being drafted to police it are, in the FTC’s telling, already gone. The comment docket is where that claim gets tested against everyone it would bind.

Sources

Sources