The FTC AI accuracy policy comment period is set to close Friday, with the agency’s draft statement putting AI companies on notice that some undisclosed steering of chatbot outputs could be treated as deceptive under Section 5 of the FTC Act. The fight is less about hallucinations than about who gets to decide when an AI answer is improperly shaped by values.
The Federal Trade Commission, led by Chair Andrew Ferguson, frames the draft Policy Statement Addressing AI Accuracy as a consumer-protection move. The agency says AI firms have marketed their systems as tools that try to produce accurate, useful responses, and that users may reasonably expect those systems not to be skewed by hidden ideological aims.
The draft says an AI company could deceive users if it changes or guides outputs against those expectations, including in an attempt to comply with a state law such as Colorado’s revised Artificial Intelligence Act. The FTC also says companies may reduce that risk by clearly telling users when a model is built to prioritize objectives other than the ones users requested or would expect.
What would the FTC AI accuracy policy do?
A policy statement does not create a new statute or formal rule. It tells companies how the FTC says it may apply existing law, in this case Section 5, which bars unfair or deceptive acts or practices.
The mechanism is straightforward: if an AI provider markets a system as accurate or objective, the FTC says undisclosed steering of outputs may make that marketing deceptive. That gives the agency a potential enforcement theory against model behavior, not just against privacy claims, billing tricks, or fake product promises.
That is where the constitutional objection starts. Critics argue that model responses involve editorial choices by private companies, including decisions about what to include, omit, rank, refuse, or qualify. In 2024, the Supreme Court said in Moody v. NetChoice that government may not decide the proper balance of private expression by forcing platforms to correct what officials view as bias.
Former FTC lawyer Keith Fentonmiller has argued that the proposal is an attempt to dictate private companies’ editorial policies and runs into the First Amendment. Another former FTC lawyer, Aaron Rieke, mocked the proposal in a LinkedIn post asking the commission to publish a quarterly list separating disfavored “ideology” from acceptable “common sense” values, so companies would know which viewpoints create enforcement risk.
Why Colorado’s AI law is in the middle of this
The FTC’s Colorado reference is unusually pointed. The draft says steering done to satisfy a state AI law may still be deceptive under federal consumer-protection law. That matters because Congress has considered, but has not enacted, broad federal preemption of state AI rules.
As written, the FTC statement would not erase Colorado’s law. It would tell companies that obeying a state requirement could still expose them to FTC scrutiny if the agency believes the resulting system behavior contradicts what consumers were led to expect.
The commission is likely to receive comments attacking both the legal theory and the practical problem it creates for AI developers. The final statement, if Ferguson’s FTC issues one, would show how aggressively the agency wants to turn claims about AI accuracy into a lever over model design and output policy.
This story draws on original reporting from Techdirt.
