Overview
On July 1, the Federal Trade Commission (FTC) issued a proposed policy statement (the Statement) advancing a significant new theory under Section 5 of the FTC Act: companies that intentionally steer AI outputs away from what consumers reasonably expect—including to comply with state AI governance laws—may be undertaking a deceptive practice. Although framed as guidance rather than a rule, the Statement extends traditional deception principles beyond affirmative misrepresentations about AI products to AI model design and optimization, while also advancing a conflict-preemption theory that could substantially affect emerging state AI laws.
Statement
The FTC’s Statement begins with the premise that consumers generally use AI systems to obtain accurate predictions, recommendations, or decisions in furtherance of their intended objective; therefore, consumers reasonably expect AI systems to optimize for that objective absent contrary disclosure.[1] After a brief discussion of the administration’s recent actions with respect to AI and the prohibition against deceptive practices under Section 5 of the FTC Act, the Statement points out that in marketing AI systems as problem-solving tools, AI companies have "explicitly and implicitly" represented that their AI systems produce the best outputs reasonably achievable given technological constraints.[2]
The Statement notes that consumers have learned to "trust" AI systems across a wide variety of applications and objectives, including education, health, finance, and personal relationships.[3] It notes one AI developer who found that "consumers accept AI system outputs without conducting any further fact checking over 90% of the time."[4] Given AI developers' marketing and the inherent value proposition of AI systems as problem-solving tools, the FTC concludes that consumers reasonably expect AI systems to prioritize accurate outputs. Although AI systems may optimize for multiple objectives, the Statement treats accuracy as the default consumer expectation absent contrary disclosure.[5]
Accordingly, the FTC takes issue with AI companies who might "steer their systems’ outputs toward objectives other than those that consumers ask for or expect," such as "by training a model surreptitiously to produce ideologically motivated distortions in a response to a factual question, such as to correct what the developer believes are ‘historical injustices’ in the facts."[6] Importantly, the Statement also asserts that this steering can happen when a company is "pressured by state law to alter its technology’s outputs," citing Colorado’s Artificial Intelligence Act, which requires AI companies to "avoid output that might lead to disparate impacts in various contexts, including when a customer’s foreseeable use of that output could itself create a disparate impact."[7] The Statement also identifies other potential motivations for steering, including responding to public pressure regarding politically sensitive outputs or advancing a developer's own policy preferences.[8]
The Statement then establishes the FTC’s belief that AI companies who steer the outputs of AI systems away from the objectives set or reasonably expected by users—including, generally, accuracy—are likely engaging in a deceptive practice in violation of Section 5 of the FTC Act.[9] The FTC therefore concludes that intentionally steering AI outputs away from users' reasonable expectations may constitute deception under Section 5 regardless of the developer's motivation. Importantly, however, the Statement does not suggest that inaccurate outputs or "hallucinations" alone violate Section 5. Rather, its focus is on deliberate design choices that prioritize objectives other than those consumers reasonably believe the AI system is intended to achieve.
The FTC acknowledges that there are ways that a company may change a consumer’s reasonable expectations. The Statement includes as an example a clear and conspicuous disclosure that an AI system is "designed to produce outputs that prioritize certain objectives over what users request and otherwise expect."[10] The Statement nevertheless suggests that sufficiently clear and conspicuous disclosures may alter consumers' reasonable expectations, while emphasizing that generalized disclaimers or boilerplate disclosures are unlikely to suffice. It provides relatively little guidance regarding what disclosures would be adequate in practice.
Insights into the Statement
The Statement establishes the FTC’s default assumption that consumers reasonably expect an AI system to be accurate, and that pursuing any other objective that may alter such output, including obeying state laws regulating the discriminatory impact of AI systems, may be deceptive under the FTC Act. Although the Statement does not preempt any state law, it expressly articulates the FTC's view that state laws requiring AI developers to engage in conduct the FTC considers deceptive may be subject to implied conflict preemption. Whether courts ultimately adopt that position remains an open question.
The Statement may be particularly significant for financial institutions and fintech companies, which increasingly deploy AI in fraud detection, underwriting, customer service, compliance, marketing, and other decision-making functions. Companies subject to both federal consumer protection requirements and state AI governance laws should evaluate whether existing AI governance frameworks appropriately address potentially competing objectives such as accuracy, fairness, explainability, and regulatory compliance.
The brief discussion at the end of the Statement provides at least one pathway for complying with both the FTC Act and state laws such as Colorado’s Artificial Intelligence Act: clear and conspicuous disclosures, provided persistently and prominently, that state that the AI system in question is designed to produce outputs prioritizing, e.g., compliance with state law over consumer expectations of accuracy.
The Statement is part of a broader federal effort to establish a national AI policy framework. Consistent with Executive Order 14365 and the Administration's AI National Policy Framework,[11] it reflects an increasing emphasis on federal standards and the potential preemption of conflicting state AI requirements.[12] Additional actions are likely to follow, further preempting state laws that are seen to conflict with the government’s policy objectives. For example, Executive Order 14365 also directed the Special Advisor for AI and Crypto and the Assistant to the President for Science and Technology to jointly prepare a legislative recommendation setting a statutory, federal AI policy framework that preempted state AI laws.[13]
The Statement leaves important legal questions unresolved, including the scope of implied conflict preemption, the circumstances under which disclosures may alter consumer expectations, and how courts will evaluate competing federal and state AI requirements. Nevertheless, it provides a clear indication of the FTC's current enforcement priorities and the legal theories the agency is likely to advance.
Practical Implications
Although the Statement remains subject to public comment, companies developing or deploying AI systems should consider whether:
- Any public statements regarding AI accuracy align with the objectives the model is designed to optimize.
- AI governance documentation appropriately identifies competing optimization objectives.
- Compliance with state AI laws could materially affect user expectations regarding model outputs.
- Disclosures adequately explain circumstances in which the AI system prioritizes objectives other than accuracy.
- Governance processes appropriately document decisions regarding model steering and optimization.
[1] See Fed. Trade Comm’n, Federal Trade Commission’s Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems at 1–2 (2026) (the "Statement").
[2] See Statement at 6.
[3] See id. at 6–7.
[4] Id. at 7.
[5] See id.
[6] Id.
[7] Id.
[8] Id. at 8.
[9] See id.
[10] Id.
[11] Scott A. Sinder et al., White House Releases National AI Policy Framework As Congress Weighs Competing Legislative Paths, Steptoe LLP (Mar. 27, 2026), https://www.steptoe.com/en/news-publications/steptechtoe-blog/white-house-releases-national-ai-policy-framework-as-congress-weighs-competing-legislative-paths.html.
[12] 90 Fed. Reg. at 58499.
[13] 90 Fed. Reg. at 58501.
