FTC Proposes Policy That Could Make Secretly Biased AI a Federal Law Violation

For years, critics on both sides of the political spectrum have accused major AI companies of secretly steering their models’ outputs — toward or away from certain topics, viewpoints, or conclusions — without telling users. The AI companies have denied systematic bias while simultaneously maintaining that content policy decisions are proprietary. The Federal Trade Commission has now proposed to make that opacity potentially illegal.

On July 7, 2026, the FTC published a proposed policy statement in the Federal Register addressing what it calls the “suppression of accuracy in artificial intelligence systems.” The statement argues that AI companies which deliberately alter their systems’ outputs to achieve undisclosed ideological objectives — rather than the best, most accurate response — may be committing deceptive acts in violation of Section 5 of the FTC Act. The commission authorized the notice by a 2-0 vote and opened a 24-day public comment period running through July 31, 2026.

What the FTC Is Actually Proposing

The statement is carefully framed. The FTC is not alleging that AI systems must be perfectly objective, or that editorial judgment is inherently deceptive. Rather, it’s arguing that AI companies have made a specific category of implicit and explicit representations to consumers: that their systems are designed to produce “the best, most accurate, and most faithful output possible within their technological and resource constraints.”

If that representation — made through marketing materials, product descriptions, and the general framing of AI as a tool for answering questions accurately — is false because the company secretly also pursues undisclosed objectives that distort outputs, then consumers are being deceived. The FTC cites data suggesting consumers accept AI outputs without independent fact-checking more than 90 percent of the time. That level of reliance, the agency argues, creates a heightened responsibility for accuracy and transparency.

The targeted behaviors are specific: deliberately altering AI model outputs to achieve ideological ends, without disclosure. The proposed statement does not prohibit content moderation, safety filtering, or even commercially motivated choices (such as refusing to generate certain categories of content). What it targets is the combination of undisclosed objectives and the reasonable consumer expectation of accuracy.

The compliance path is relatively clear. Under the proposed policy, an AI company can avoid Section 5 liability by making “clear, conspicuous, and adequate disclosures” that its system is designed to prioritize certain objectives over pure accuracy. In other words: tell users your model has a point of view, and you’re likely safe. Keep it hidden while claiming objectivity, and you’re potentially in violation.

The State Preemption Dimension

The proposed statement carries a second, strategically significant component: an analysis of state AI laws that require alteration of AI outputs, focusing specifically on Colorado’s Artificial Intelligence Act.

Colorado’s law, among other things, includes provisions requiring AI systems to avoid certain types of discriminatory output — which, the FTC’s proposed statement suggests, could itself mandate output distortions that violate the accuracy principle. The statement raises the possibility that federal consumer protection law could preempt state-level AI output mandates that create a conflict.

This is legally and politically complex. The AI industry has lobbied heavily against state-by-state AI regulation, preferring either federal preemption or no regulation at all. Consumer advocates and state attorneys general have pushed back, arguing that federal inaction has left states as the only practical check on AI harms. By introducing the preemption question, the FTC is inserting itself into a live battle between state and federal regulatory authority over AI.

FTC Chairman Andrew N. Ferguson stated that the commission “wants to hear from businesses and consumers about their experiences and concerns regarding the subversion of AI systems for ideological ends” — framing that signals the agency is particularly interested in politically motivated output steering, not just algorithmic bias arising from training data.

The Context: Why This, Why Now

The FTC’s proposal arrives in a specific political moment. Questions about AI bias have become a major political flashpoint, with conservatives arguing that major AI models systematically suppress conservative viewpoints and Democrats arguing that AI systems perpetuate racial and gender bias. Both sets of complaints, viewed uncharitably, involve companies making undisclosed choices about output that diverge from what users expect.

Chairman Ferguson’s framing — emphasizing “ideological ends” — signals where the current FTC sees the primary problem. The Trump-era appointee has been consistent in his concern about alleged left-wing bias in technology platforms, and this proposed statement can be read in part as an extension of that concern into the AI context.

But the underlying legal theory, if adopted, is not inherently partisan. A standard requiring disclosure of output-steering objectives would apply equally to a model that deprioritizes conservative viewpoints and one that deprioritizes liberal perspectives. It would apply to an AI customer service agent secretly instructed to never recommend competitor products, and to a medical information AI secretly calibrated to reduce healthcare utilization.

The practical implication — if the policy statement is adopted and enforced — is significant. AI companies would need to audit their systems for undisclosed output-influencing objectives, document those objectives, and either disclose them to users or eliminate them. For large-scale foundation model providers with hundreds of fine-tuning decisions layered into their systems, that transparency exercise is non-trivial.

Industry and Expert Response

Industry responses have been cautious. AI companies are unlikely to publicly oppose a “be accurate and transparent” standard, but the implementation challenges are real. What counts as an “undisclosed ideological objective”? Is safety filtering ideological? Is refusing to generate certain medical information ideological? The proposed statement does not answer these questions, which is partly the point of the comment period.

Legal experts have noted that Section 5 FTC enforcement against AI companies would be genuinely novel. The FTC has used its Section 5 authority against deceptive advertising, data privacy violations, and unfair competitive practices, but not against the internal calibration decisions of AI model providers. Establishing that such calibration constitutes a consumer-facing deception would require either consent decrees (essentially settlements) or litigation that could reach the Supreme Court.

The public comment deadline of July 31 means the FTC’s next move — either finalizing the statement, revising it, or withdrawing it — will likely come in the fall. Whatever the outcome, the proposal has already changed the conversation: AI accuracy is now explicitly on the federal enforcement agenda, and companies building and deploying AI models have been put on notice that their output calibration choices are a matter of consumer protection law.

The Bigger Picture

The FTC’s proposed statement is one piece of a rapidly evolving U.S. AI regulatory mosaic. The Senate has been debating the AI Patent Eligibility Restoration Act. The GSA finalized LLM safeguard requirements for federal procurement. Illinois signed its own AI regulation bill. The EU delayed implementation of its high-risk AI rules. At the UN, governments are debating global AI governance frameworks.

What’s emerging is not a coherent regulatory architecture but a patchwork of sector-specific, jurisdiction-specific interventions responding to immediate political pressures. The FTC’s proposal is unusual in that it attempts to apply an existing, durable legal standard — the prohibition on deception — to AI behavior, rather than creating a new AI-specific regime.

Whether that approach proves more durable than purpose-built AI legislation is an open question. But it is unambiguously the most concrete signal yet that the era of self-regulation for AI output decisions may be drawing to a close.