FTC seeks comment on AI 'accuracy' policy under Section 5

The FTC's newest AI move is less an accuracy rule and more a warning shot at how models get tuned. According to the FTC's announcement, Chairman Andrew N. Ferguson said the goal is to "hear from businesses and consumers about their experiences and concerns regarding the subversion of AI systems for ideological ends." The Commission voted 2-0 to open a public comment window that runs through July 31, 2026.

The proposed policy statement, titled the "Suppression of Accuracy in Artificial Intelligence Systems," argues that AI companies that distort their systems' outputs to achieve undisclosed "ideological objectives" can be deceiving consumers under Section 5 of the FTC Act, which prohibits unfair or deceptive business practices. That framing puts a lot of standard industry practice on the table, since the FTC says training chatbots to avoid responses that discriminate against specific groups of people may run afoul of the same statute.

Why this matters if you ship a consumer model: the guardrails and post-training tuning that model providers currently treat as a safety obligation could be read by an enforcement agency as an undisclosed manipulation of the product consumers thought they were buying. The statement also floats a preemption theory, saying Colorado's Artificial Intelligence Act is "impliedly preempted to the extent it conflicts with a federal regulatory scheme," which is a direct signal to state legislatures drafting their own AI rules.

The honest caveats. This is a proposed policy statement, not a rule, and not an enforcement action; the comment period exists precisely so the shape of any future case can be argued. What the reporting doesn't give you is a workable line between the "ideological" tuning the Commission wants to police and the routine safety fine-tuning nearly every frontier lab already documents, nor does the announcement name a specific target company. Take the specifics as reported, not settled.

The people who benefit soonest are compliance teams that already keep clear records of what their models are trained to refuse, and why. The July 31 docket is where the definitions get contested, and the filings from labs, civil society groups, and state attorneys general are worth reading closely, since they will preview how any eventual Section 5 case is fought.