The FTC AI accuracy policy proposed July 1 is drawing criticism from Keith R. Fentonmiller, a former senior attorney in the Federal Trade Commission’s Division of Advertising Practices, who says the agency is using advertising law to police what AI chatbots are allowed to say.
The FTC’s proposed policy statement on “Suppression of Accuracy in Artificial Intelligence Systems” says AI developers likely engage in false advertising when they steer model outputs toward goals users do not expect. According to the proposal, companies that market AI systems as helpful may create an expectation that the systems will give maximally accurate answers, making undisclosed shaping of outputs potentially deceptive.
Fentonmiller argues the FTC has not identified a specific misleading ad or deceived consumer to justify that theory. His criticism is sharper than a normal policy disagreement because it goes to mechanism: large language models are built through design choices that affect what they generate, and the FTC’s proposal treats some of those choices as possible deception when they are not disclosed.
What is the FTC AI accuracy policy?
The FTC proposal says AI companies may face false-advertising liability if they present a system as useful or accurate while steering its answers toward other objectives without clear disclosure. The agency says developers could reduce risk by clearly and conspicuously telling users when a system prioritizes something other than “the best answer possible.”
Fentonmiller says that approach collides with First Amendment doctrine. He points to Brown v. Entertainment Merchants Association, where the Supreme Court held that video games receive First Amendment protection, and Moody v. NetChoice, where the court treated platforms’ decisions about what expressive content to present as protected editorial judgment.
On that view, what the FTC calls steering may look a lot like editing. That is the part the proposal needs to answer cleanly, because a model’s output rules, refusal behavior, ranking choices and safety layers are all editorial in the ordinary sense, even when implemented in code instead of a newsroom meeting.
Fentonmiller uses Truthly, a chatbot marketed around a Catholic worldview, as an example of the disclosure problem. Truthly says it is not impartial and presents its information as filtered through truth and morality. He argues that a religious chatbot could still face trouble if the FTC deemed its “biased but true” positioning confusing under the agency’s disclosure standards.
He also says the proposal’s logic could reach media companies that promise accuracy while making editorial choices. Fentonmiller cites the New York Times’ “All the News That’s Fit to Print,” Newsmax’s “real news” branding, Breitbart’s stated goal to report truth accurately and fairly, and One America News’ “credible source” marketing. In 2004, then-FTC Chairman Timothy Muris declined to pursue a complaint over Fox News’ “Fair and Balanced” slogan, saying judging news content belonged to the public rather than a government agency.
The proposal also sits awkwardly beside recent FTC AI enforcement. Fentonmiller notes that Chairman Andrew Ferguson supported action against DoNotPay over “robot lawyer” claims and against Workado over claims about an AI-detection product’s accuracy. Ferguson had also criticized a Biden-era order against Rytr, a generative AI writing tool, saying the FTC was targeting a product that helps people speak. As chairman, Ferguson later vacated the Rytr order.
Fentonmiller ties the new proposal to a changed agency structure after Trump v. Slaughter, which he says stripped the FTC of statutory independence and left a two-member, one-party commission. He argues that the policy risks making AI developers answer to shifting political definitions of accuracy, rather than to concrete evidence of deception.
This story draws on original reporting from Techdirt.
