FTC Warns Correcting AI Chatbot Bias May Violate Federal Law

The U.S. Federal Trade Commission (FTC) published a draft policy statement on July 1 warning that AI companies training their chatbots to avoid ideologically uncomfortable answers may be violating federal unfair trade practices law. Public comments on the document will be accepted through July 31.

The document, titled Proposed Policy Statement Concerning The Suppression Of Accuracy In Artificial Intelligence Systems, does not directly ban bias in language models. Instead, it requires companies to clearly disclose to users when their systems prioritize goals other than maximum accuracy in their responses.

The FTC wants to hear from companies and consumers about their experiences and concerns regarding the subordination of AI systems to ideological goals - Andrew N. Ferguson, FTC Chairman

Section 5 and the line of legality

The Commission grounds its position in Section 5 of the FTC Act, a classic tool for pursuing unfair and deceptive trade practices that has so far been applied mainly to advertising and consumer protection. The FTC argues that users have a legitimate expectation that AI systems aim for truthful and accurate answers, and that any departure from that principle without clear disclosure amounts to a deceptive practice.

The key caveat, however, is that companies can avoid liability if they make truthful and non-misleading statements about their system's actual goals. In practice, this means a chatbot provider can knowingly moderate responses along ideological lines, as long as it clearly communicates this to users instead of presenting the model as a neutral source of facts.

Dispute over Colorado's law

The FTC explicitly questioned whether Colorado's revised AI-related rules comply with federal law. The Commission argues that the state law pressures technology companies into modifying chatbot responses in ways aligned with the state's ideological goals. Colorado replaced its earlier strict discrimination risk-assessment requirements with lighter disclosure obligations for decisions involving education, healthcare, and employment.

The dispute highlights the tension between state regulation, which governs how AI systems behave, and the FTC's federal guidance, which governs what companies must disclose. Compliance lawyers note that the two levels of regulation do not necessarily conflict, but they can significantly complicate compliance planning for companies, especially where state and federal rules overlap across jurisdictions.

Political backdrop

The FTC's initiative did not emerge in a vacuum. It follows the Trump administration's December executive order directing the Commission to examine the legality of state AI regulations, as well as a January executive order aimed at so-called Woke AI, demanding ideological neutrality from chatbots. These moves draw on research such as the Washington Post analysis, which found that OpenAI's ChatGPT shows the strongest left-leaning tilt among the chatbots studied, while Google's Gemini balances liberal and conservative positions in roughly 90 percent of its answers.

For companies like OpenAI, Google, and Anthropic, this opens a new regulatory front in the United States, this time concerning not model safety or copyright, but how they present their own operating principles to users. The challenge will be determining exactly what counts as deceptive behavior and what form of disclosure is sufficient to avoid liability, without burying users in yet more legal notices.

Relevance for Polish companies and users

Polish companies using American language models in their products should watch how this case develops, since changes to disclosure policies at providers like OpenAI or Google could translate into new notices and restrictions visible in Europe as well. At the same time, EU regulations, including the AI Act, are heading in a partly different direction than the American debate over ideological neutrality, focusing more on risk assessment and transparency for high-risk systems than on the political leanings of a model's answers.

The public comment period closes on July 31, after which the FTC may adopt the policy as proposed or in amended form. AI companies operating in the U.S. market have about a month to prepare their positions and, if needed, adjust their disclosure practices.

Sources: Forbes (forbes.com), San.com (san.com), Federal Trade Commission (ftc.gov)