The FTC’s proposed policy statement on the “suppression of accuracy” in AI systems closed its public comment window on July 31, with 40 filings now sitting in docket FTC-2026-0859. What the Commission does with them will determine whether undisclosed output-steering in consumer-facing AI becomes a Section 5 deception violation by winter.

The proposal, published July 1 under File No. P264200 and authorized by a 2-0 Commission vote, traces back to Executive Order 14365, which President Trump signed on December 11, 2025. Its logic is narrower than the headline suggests. Hallucinations aren’t the target. Deliberate configuration away from accuracy is: brand-safety filtering, ideological steering, or state-law compliance retrofits applied without prominent disclosure to the end user.

Two details make this a small-business story, not just a model-lab story. First, the FTC’s framework reaches deployers, not only developers, continuing the posture from its prior small-business notice on undisclosed AI pricing and the Cox Media / Mindsift settlement. Second, per Ballard Spahr’s Consumer Finance Monitor, disclaimers buried in terms of service won’t satisfy the standard; the disclosure must be conspicuous.

The most striking line in the proposal is what it forecloses. The FTC signals that trying to comply with Colorado’s AI Act isn’t a defense against a federal deception charge if the compliance work quietly bends outputs. State bias rules and federal accuracy rules point in opposite directions, and the agency is claiming the tiebreak.

PYMNTS notes the proposal describes an enforcement lens rather than binding law, and the FTC hasn’t said when, or whether, it’ll finalize. For a small team using a chatbot, an outreach generator, or a marketing-copy tool, the practical read is unglamorous: ask the vendor, in writing, what the system filters and why. That question is now a compliance record.

Sources

Sources