The Federal Trade Commission released on July 1, 2026 a proposed policy statement arguing that artificial intelligence companies which secretly steer their systems' outputs toward undisclosed ideological objectives may be violating Section 5 of the FTC Act, and the document takes direct aim at Colorado's revised Artificial Intelligence Act as a law that could force exactly that kind of hidden manipulation.

The nine-page statement, titled "Federal Trade Commission's Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems," opens a public comment period running through July 31, 2026. The Commission approved the document by a 2-0 vote, and the notice was published on public inspection at the Federal Register on July 6, 2026, ahead of formal publication before the comment window closes.

For an industry that has spent two years wiring AI tools into search, creative production, and campaign measurement, the statement lands as a warning about a specific kind of business risk: designing a system to quietly favor one output over another, without telling the people who use it, according to the Commission.

What the statement actually says

The core argument extends a nearly ninety-year-old law to a new kind of product. Section 5 of the FTC Act prohibits unfair or deceptive acts or practices in commerce. The Commission has applied that same three-part deception test since 1984: there must be a representation, omission, or practice likely to mislead a reasonable consumer, and that misrepresentation must be material to the consumer's decisions.

According to the statement, AI companies have spent years marketing their products as tools designed to produce the best, most accurate output possible within their technical and resource limits. Consumers, the statement argues, have absorbed that message and now trust AI systems to try their best on medical questions, financial advice, relationship guidance, and ordinary homework help. The Commission does not treat that trust as incidental. It treats the marketing itself as the source of a legal obligation: if a company says its system aims for the best answer, and instead secretly optimizes for something else, the mismatch between promise and product can be deceptive under Section 5.

The statement is careful to separate two very different failure modes. Ordinary hallucinations, meaning wrong answers caused by a model's technical or resource limitations, sit outside the policy's target. Nothing in the document treats a language model's occasional factual error as, by itself, illegal. What the Commission does target is a deliberate design choice: training or configuring a model to prioritize an undisclosed objective, such as inserting a particular ideological framing, ahead of accuracy, while continuing to advertise the system as accuracy-first.

The historical injustices example

One passage stands out for how specific it gets. The statement describes a scenario in which a company trains its model to correct what its own developers consider "historical injustices" in the facts, without disclosing that override to users. That is presented as a textbook case of the kind of conduct the statement targets: a company overriding a factually correct answer and substituting a different one because of the developer's own agenda, while marketing the product as striving for accuracy.

The document draws on public statements from several AI companies to support its claim that these representations are widespread. It cites Anthropic's own product marketing describing Claude as a tool built to be "safe, accurate, and secure," alongside comparable language from OpenAI's ChatGPT materials and xAI's positioning of Grok as a "truth-seeking AI companion." The statement also references separate Anthropic research on how people use Claude for personal guidance, noting that a meaningful share of conversations touch on relationship navigation, personal development, and spirituality, as evidence that consumers rely on these systems well beyond narrow technical tasks.

A figure drawn from a Forbes column citing Anthropic research also appears in the statement: according to the citation, over 90 percent of users do not conduct further fact-checking on AI outputs. The Commission uses that figure to argue that the stakes of undisclosed steering are higher than they might otherwise be, since most people take an AI system's answer largely at face value.

Why Colorado is named directly

The statement does not stay abstract for long. It names Colorado's Artificial Intelligence Act, most recently amended through Senate Bill 26-189, as a law that could pressure companies into exactly the kind of concealed output manipulation the FTC considers deceptive. The bill that replaced Colorado's original statute passed the state House 57-6 on May 9, and the amended version, as PPC Land reported, introduced new automated decision-making technology disclosure and consumer rights rules that take effect January 1, 2027.

The FTC's statement argues that a law banning what it calls "algorithmic discrimination" could force AI developers to distort otherwise accurate outputs to avoid liability for statistically uneven results across protected groups. If a company does this quietly and continues to market its system as accuracy-first, the statement contends that the company has deceived consumers regardless of whether the underlying motive was compliance with state law. The FTC Act does not expressly preempt state law, but the statement argues that state requirements are impliedly preempted where they directly conflict with the federal deception framework, since a state cannot force a company to violate a federal statute and expect that compliance to serve as a defense.

This is not the first time Washington has singled out Colorado's AI statute. Executive Order 14365, signed by President Trump on December 11, 2025, specifically referenced Colorado's law when directing the FTC to prepare guidance on how state AI rules interact with Section 5. As PPC Land reported, the U.S. Department of Justice filed a complaint in intervention on April 24, 2026, joining a federal lawsuit brought by xAI against Colorado Attorney General Philip Weiser, arguing that Colorado's original statute compelled AI developers to discriminate in violation of the Equal Protection Clause. That case, filed under Civil Action No. 1:26-cv-01515-DDD-CYC in the U.S. District Court for the District of Colorado, produced an order on April 27, 2026, blocking Colorado's Attorney General from enforcing the original law until rulemaking concludes and the court rules on the underlying constitutional challenge.

Colorado's legislature responded by repealing and replacing the earlier statute. SB 26-189 cleared the state Senate and House in early May 2026 and now carries a January 1, 2027 effective date, with enforcement authority resting exclusively with the Colorado Attorney General under the Colorado Consumer Protection Act. Violations are classified as deceptive trade practices, carrying civil penalties of up to 20,000 dollars per violation. The FTC's new statement makes clear that the revised law still concerns federal regulators, noting in a footnote that Colorado "materially revised" the law but that the new version "poses many of the same concerns" as the original.

The disclosure standard the FTC describes

The statement does leave companies a path to compliance that does not involve abandoning any particular design choice. A company can prioritize objectives other than pure accuracy, according to the document, provided it discloses that trade-off clearly and conspicuously. The bar the FTC sets for that disclosure is demanding. A disclaimer buried in terms of service will not satisfy the standard, the statement argues, and a one-time disclosure that later disappears into fine print is unlikely to be enough either. The more a company's actual behavior departs from what a reasonable consumer would otherwise expect, the more prominent and persistent the disclosure needs to be, according to the statement's reasoning, which draws on established case law holding that a misleading headline is not cured by an accurate paragraph further down the page.

That standard echoes obligations advertisers already navigate elsewhere. The FTC's own 2024 rule against fake reviews and undisclosed paid testimonials, and its long-running guidance on influencer disclosures, rest on a similar premise: a disclosure that a reasonable person would miss does not count as a disclosure at all. The new AI statement applies that same logic to a different kind of product, treating an AI system's design priorities as a claim the company makes about itself, subject to the same clarity requirements as any other advertising representation.

A pattern, not an isolated action

This statement did not appear out of nowhere. It executes a specific instruction contained in Executive Order 14365, which directed the FTC to clarify how Section 5 interacts with state laws that require alterations to AI model outputs. The order itself built on earlier executive actions, including one from July 2025 aimed at removing regulatory barriers to AI adoption and another addressing what the administration termed "Preventing Woke AI in the Federal Government."

The FTC has been active on AI enforcement well before this statement. PPC Land reported that the Commission ordered seven companies operating consumer-facing AI chatbots to submit detailed reports on safety practices, data handling, and monetization by September 10, 2025, under Section 6(b) of the FTC Act. That inquiry focused on child safety rather than output accuracy, but it demonstrated the Commission's willingness to use compulsory process against AI-specific business models. Separately, PPC Land documented Operation AI Comply, launched in September 2024, which brought enforcement actions against companies including DoNotPay over claims that its chatbot could replace a human lawyer.

The new statement cites that same DoNotPay case, along with a more recent action against Air AI Techs over claims regarding the efficacy of its conversational AI product for customer service. It also references the Commission's decision in December 2025 to reopen and set aside a prior consent order against Rytr LLC, a generative AI company, a move the statement frames as evidence that the current Commission is careful not to burden AI innovation unnecessarily even as it pursues deception cases where the facts warrant it.

The FTC's broader 2026-2030 Strategic Plan, approved in April 2026 by a 2-0 Commission vote, already signaled that consumer protection work touching AI, data practices, and platform accountability would remain a priority through the back half of the decade. The new AI accuracy statement fits inside that broader agenda rather than standing apart from it.

What remains unsettled

Several open questions run through the statement without clear resolution. The document does not define precisely where routine safety fine-tuning, the kind nearly every AI developer already documents and discloses in some form, ends and prohibited "ideological steering" begins. Nor does it name a specific company as a target, leaving the practical scope of enforcement to future cases built on these principles.

The preemption theory the statement advances is also untested in court. Federal courts have not yet ruled on whether Section 5's deception framework actually displaces a state consumer protection statute like Colorado's, and the statement itself acknowledges that the FTC Act does not expressly preempt state law, resting its argument instead on implied conflict preemption, a doctrine that requires courts to find it genuinely impossible to comply with both state and federal requirements simultaneously.

There is also the question of process. This is a proposed policy statement, not a final rule and not an enforcement action. It does not create new legal obligations by itself. The document signals how the Commission intends to apply an existing, decades-old legal framework to a new class of products, and the 30-day comment window exists precisely so that businesses, consumers, and legal scholars can weigh in before the Commission finalizes its position.

Why this matters for marketers

For anyone running AI-assisted advertising tools, whether that is a chatbot answering customer questions, a generative creative engine, or an AI-powered bidding system, the statement introduces a new category of compliance question that sits alongside existing disclosure obligations already familiar to the industry. The FTC has enforced disclosure standards around fake reviews, undisclosed paid endorsements, and misleading influencer content for years. This statement extends a comparable logic to the design priorities baked into an AI system itself: if a product's behavior departs meaningfully from what a company's own marketing promises, silence about that gap can become the basis for a deception claim.

The Colorado dimension carries separate weight for any marketing organization operating across multiple U.S. states. Companies building AI tools for advertising, search, or customer engagement now face two federal signals, an executive order and this proposed statement, arguing that state-level AI bias and discrimination rules can conflict with federal consumer protection law. Whether that argument survives judicial review remains genuinely open, but the uncertainty itself is a planning problem. A company adjusting its AI system's outputs to satisfy Colorado's ADMT disclosure and consumer rights requirements, effective January 1, 2027, could find itself facing a federal argument that the same adjustment, if undisclosed to end users, creates separate exposure under Section 5.

None of this changes what advertisers and publishers should do about AI-generated content today. It does, however, add a federal enforcement theory to a state-by-state patchwork that has already produced litigation in Colorado, comprehensive legislation in Connecticut, and a wave of narrower measures in states such as California and Vermont addressing companion chatbots and AI-only mental health services. Comments on the FTC's proposed statement close July 31, 2026, giving businesses and consumers roughly a month to put their views on the record before the Commission finalizes its position.

Timeline

  • July 23, 2025 - Executive Order No. 14319, "Preventing Woke AI in the Federal Government," is signed, later cited in the FTC's new statement.
  • December 11, 2025 - President Trump signs Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," directing the FTC to clarify how Section 5 applies to AI models and to address conflicts with state AI laws.
  • December 22, 2025 - The FTC reopens and sets aside its 2024 consent order against Rytr LLC, cited in the new statement as evidence of restraint toward AI innovation.
  • April 9, 2026 - xAI files a federal lawsuit challenging Colorado's original AI statute, Senate Bill 24-205, on constitutional grounds.
  • April 24, 2026 - The U.S. Department of Justice files a complaint in intervention in xAI's Colorado lawsuit, arguing the state law compels discrimination in violation of the Equal Protection Clause.
  • April 27, 2026 - A U.S. District Court orders Colorado's Attorney General not to enforce SB 24-205 or any amending legislation until rulemaking concludes and the court rules on the underlying case.
  • May 9, 2026 - Colorado's House passes Senate Bill 26-189, repealing and replacing SB 24-205 with new automated decision-making technology rules effective January 1, 2027.
  • July 1, 2026 - The FTC releases its proposed policy statement on the suppression of accuracy in AI systems, approved 2-0, naming Colorado's revised AI Act as a law that could compel undisclosed output manipulation.
  • July 31, 2026 - The public comment period on the FTC's proposed statement closes.

Summary

Who: The Federal Trade Commission, chaired by Andrew N. Ferguson, issued the proposed policy statement. The document directly names Colorado's Artificial Intelligence Act, as revised by Senate Bill 26-189, and references marketing language from Anthropic, OpenAI, and xAI as examples of industry representations about AI accuracy.

What: A proposed policy statement arguing that AI companies which steer their systems' outputs toward undisclosed ideological objectives, rather than the accuracy consumers are led to expect, may be engaging in deceptive practices under Section 5 of the FTC Act. The statement distinguishes this conduct from ordinary hallucinations and outlines a disclosure standard companies could use to avoid liability.

When: The statement was released July 1, 2026. The public comment period runs through July 31, 2026.

Where: The action originates from the FTC in Washington, D.C., but its reasoning applies to any AI company marketing products to U.S. consumers, and it specifically addresses tension with Colorado's state-level AI law.

Why: The statement executes a directive from Executive Order 14365 and reflects the current administration's broader effort to establish federal primacy over a growing patchwork of state AI regulations. For marketers, it introduces a federal consumer-protection theory that could apply to any AI-powered tool whose actual design priorities diverge from what its marketing promises, while adding fresh uncertainty to compliance planning for companies already adjusting systems to satisfy state-level disclosure and discrimination rules.