FTC proposed AI bias disclosures could reshape how developers communicate truthfulness of outputs

The FTC is proposing a new policy that would require AI companies to disclose when their generative AI systems produce biased or less-than-truthful outputs. The proposal, backed by the FTC Act's consumer protection authority, focuses on transparency rather than banning bias altogether. For AI builders, this means preparing for clearer disclosure requirements and navigating potential conflicts with state-level AI laws.

What happened

The Federal Trade Commission issued a policy proposal on July 1, 2026 that aims to compel AI makers to disclose biases in their large language models. The FTC is seeking public comment until July 31, 2026, and plans to enforce the policy under Section 5 of the FTC Act.

The proposal does not ban biased AI outputs. Instead, it requires companies to inform users when a system prioritizes objectives different from what users expect. The FTC argues that consumers reasonably expect accurate information from AI, and any deviation without clear disclosure could be deceptive.

Why AI builders should care

The FTC frames misleading AI outputs as a consumer-protection concern that could erode user trust. Builders of AI products may need to redesign how they communicate model behavior to match the "reasonable consumer" standard the FTC intends to apply.

State-level laws, such as Colorado's revised Artificial Intelligence Act, add another layer. The FTC notes that compliance with state laws could inadvertently steer AI outputs in ways that require federal disclosures. The two obligations are not necessarily in conflict, but builders must address both sets of requirements.

Practical implications

Disclosures must be clear and conspicuous, proportional to the behavioral departure from user expectations. The more the AI diverges, the more prominent the disclosure needs to be. The FTC expects builders to update disclosures as models are continuously updated.

Crucially, the policy does not require unbiased AI. It requires honest representation. If a model is designed to favor a viewpoint or commercial partner, the maker must tell users. This shifts the compliance burden from content moderation to transparency in AI outputs.

Caveats

The proposal is in its early stages. Details may change after public comments are considered. Key ambiguities remain: how to define "truthful" outputs, what constitutes a sufficient disclosure, and how to measure the "reasonable consumer" baseline. Critics warn of regulatory overreach, while supporters argue it is a necessary safeguard as AI becomes ubiquitous.