FTC Proposes AI Accuracy Rules Targeting 'Hidden Agendas'

The Federal Trade Commission has opened public comment on a policy statement that could redefine how AI developers must disclose the objectives behind their models. The proposal, issued last week with comments due by the end of July, advances the theory that companies could violate deception prohibitions under Section 5 of the FTC Act if they tune models to achieve "undisclosed ideological objectives."

Why it matters

This marks the first time a federal regulator has formally proposed treating the ideological tuning of AI systems as a potential consumer protection violation. The policy could force companies to disclose when models are adjusted to produce outputs that deviate from what users would reasonably expect based on training data alone — a disclosure requirement that could fundamentally reshape AI product marketing and documentation.

Consumer expectations as the measuring stick

The proposed statement rests on the assertion that generative AI interfaces have created explicit or implicit consumer expectations that outputs should match reality. The FTC acknowledges this standard isn't absolute — users might reasonably expect chatbots to balance "succinctness, clarity, relevance, accuracy and other objectives" or even request intentionally inaccurate outputs for entertainment purposes.

The core requirement is disclosure. Any "unexpected objectives" or "hidden agenda" underlying model performance must be clearly communicated to consumers, even if another law mandates the adjustment. The FTC argues that failing to disclose such modifications constitutes deception.

State anti-discrimination laws under fire

Following instructions from the White House's December 2025 executive order, the proposed statement explicitly targets state laws that impose liability for discriminatory AI outcomes. The FTC singles out Colorado's AI Act — now repealed and replaced — arguing such laws incentivize companies to "falsify or artificially steer" outputs to avoid disparate impacts, thereby deceiving consumers.

The original Colorado law would have applied only when AI systems were substantial factors in consequential decisions regarding employment, housing, financial services, healthcare, education, or government services. The FTC claims even the diminished replacement law creates problematic incentives, stating it's "predictable that an AI company might suppress accuracy and interpose other objectives, such as so-called 'equity,' to avoid liability under this law, but fail to disclose these ulterior objectives."

Competing visions for consumer protection

Two former FTC attorney advisors, Gaurav Laroia and Charlotte Slaiman, recently proposed an alternative approach in Tech Policy Press: a large-scale public messaging campaign to reset consumer understanding of how generative systems work. They suggested funding such efforts through settlements similar to the Tobacco Master Settlement Agreement, arguing that "we are currently fighting an asymmetric battle on messaging with the companies."

Both approaches acknowledge the same problem — consumers place excessive reliance on AI outputs — but differ on solutions. The FTC proposal emphasizes disclosure requirements and policing hidden objectives, while the public education approach focuses on cultivating digital literacy about the predictive, probabilistic nature of these systems.

The details were first reported by the International Association of Privacy Professionals.