Blab AI
FTC Proposes Policy to Curb Deceptive AI Output Steering
On July 7 2026, the Federal Trade Commission set the stage for a new era of AI transparency with a bold proposal that could redefine how companies market and build their artificial intelligence tools. The agency’s policy statement, published in the Federal Register, extends the FTC Act’s prohibition on deceptive acts to firms that steer AI systems toward hidden goals—whether ideological agendas, compliance with state laws, or other undisclosed objectives.
The FTC’s core argument hinges on consumer expectations. Users buy AI products on the promise that the tool will deliver the most accurate answer possible. When a model is tuned to produce outputs that serve a different, unseen purpose, the agency says the company is likely misrepresenting its product. The proposal cites examples such as tweaking factual answers to fit an ideology, embedding “equity” objectives, or sidestepping politically sensitive topics without informing users.
State‑law compliance is not a defense. The statement specifically targets Colorado’s revised Artificial Intelligence Act (S.B. 26‑189). The FTC contends that a company that modifies its outputs to satisfy a state law can still violate Section 5. The agency argues that state laws that require AI firms to deceive consumers conflict with the FTC Act’s purpose of protecting consumers from such conduct, and therefore may be pre‑empted.
According to the FTC, a company’s motivation is irrelevant. Under long‑standing precedent, if a firm deceives a customer, the reason—whether profit, public pressure, employee politics, or state‑law compliance—is immaterial. The proposal treats all hidden‑steering practices the same.
Disclosure is the only way to comply, but the bar is high. A firm can steer its outputs if it informs users clearly and prominently. The FTC says a brief mention in a terms‑of‑service document or a single fine‑print notice is insufficient. The further the practice deviates from what users reasonably expect, the more prominent the disclosure must be.
The proposal does not cover ordinary AI mistakes. Technical errors that produce hallucinations or incorrect answers are not deceptive in themselves. Likewise, blocking illegal content or preventing cyberattacks does not raise issues. However, overstating how rarely a system makes mistakes could be deceptive.
Comments on the proposal are due July 31 2026. The FTC invites the public to file comments at regulations.gov under docket No. FTC‑2026‑0859, referencing “AI Policy Statement; Matter No. P264200.”
The policy statement follows Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence,” signed on December 11 2025. The order directs the FTC to explain how state laws that force changes to AI outputs can conflict with federal law. The proposal also draws on a separate executive order related to fair‑lending enforcement, which the current administration treats as a non‑viable legal theory.
The FTC’s legal test for deception comes from its 1983 Policy Statement on Deception. A practice is deceptive if it is likely to mislead a reasonable consumer and the misleading claim matters to the consumer’s decision. The agency notes that this covers implied claims and half‑truths, and that companies must be able to back up the claims they make.
For AI developers and deployers, the proposal raises the stakes on training, fine‑tuning, and content‑moderation choices that alter what models say. The FTC’s position could also affect companies that rely on third‑party AI systems, as the agency’s third‑party risk‑management principles would extend the compliance burden beyond service providers.
If the policy statement is finalized later this summer or by the fall, companies will need to inventory the goals their systems prioritize and compare those goals against marketing claims. Where a gap exists, they must decide whether a clear, prominent disclosure is sufficient. The FTC has stated that terms‑of‑service disclosures alone are not enough.
The proposal is a preliminary step. The agency has not yet issued a rule, and litigation could shape the final outcome. Until then, companies that deploy AI systems should review their marketing and product claims about accuracy, objectivity, and capability, and document compliance choices, especially if they are subject to state AI laws.
In short, the FTC’s July 7 proposal signals that hidden steering of AI outputs toward undisclosed goals may be deemed deceptive under federal law, that state‑law compliance does not excuse such conduct, and that clear, prominent disclosure is required to avoid violating Section 5 of the FTC Act.
The FTC’s core argument hinges on consumer expectations. Users buy AI products on the promise that the tool will deliver the most accurate answer possible. When a model is tuned to produce outputs that serve a different, unseen purpose, the agency says the company is likely misrepresenting its product. The proposal cites examples such as tweaking factual answers to fit an ideology, embedding “equity” objectives, or sidestepping politically sensitive topics without informing users.
State‑law compliance is not a defense. The statement specifically targets Colorado’s revised Artificial Intelligence Act (S.B. 26‑189). The FTC contends that a company that modifies its outputs to satisfy a state law can still violate Section 5. The agency argues that state laws that require AI firms to deceive consumers conflict with the FTC Act’s purpose of protecting consumers from such conduct, and therefore may be pre‑empted.
According to the FTC, a company’s motivation is irrelevant. Under long‑standing precedent, if a firm deceives a customer, the reason—whether profit, public pressure, employee politics, or state‑law compliance—is immaterial. The proposal treats all hidden‑steering practices the same.
Disclosure is the only way to comply, but the bar is high. A firm can steer its outputs if it informs users clearly and prominently. The FTC says a brief mention in a terms‑of‑service document or a single fine‑print notice is insufficient. The further the practice deviates from what users reasonably expect, the more prominent the disclosure must be.
The proposal does not cover ordinary AI mistakes. Technical errors that produce hallucinations or incorrect answers are not deceptive in themselves. Likewise, blocking illegal content or preventing cyberattacks does not raise issues. However, overstating how rarely a system makes mistakes could be deceptive.
Comments on the proposal are due July 31 2026. The FTC invites the public to file comments at regulations.gov under docket No. FTC‑2026‑0859, referencing “AI Policy Statement; Matter No. P264200.”
The policy statement follows Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence,” signed on December 11 2025. The order directs the FTC to explain how state laws that force changes to AI outputs can conflict with federal law. The proposal also draws on a separate executive order related to fair‑lending enforcement, which the current administration treats as a non‑viable legal theory.
The FTC’s legal test for deception comes from its 1983 Policy Statement on Deception. A practice is deceptive if it is likely to mislead a reasonable consumer and the misleading claim matters to the consumer’s decision. The agency notes that this covers implied claims and half‑truths, and that companies must be able to back up the claims they make.
For AI developers and deployers, the proposal raises the stakes on training, fine‑tuning, and content‑moderation choices that alter what models say. The FTC’s position could also affect companies that rely on third‑party AI systems, as the agency’s third‑party risk‑management principles would extend the compliance burden beyond service providers.
If the policy statement is finalized later this summer or by the fall, companies will need to inventory the goals their systems prioritize and compare those goals against marketing claims. Where a gap exists, they must decide whether a clear, prominent disclosure is sufficient. The FTC has stated that terms‑of‑service disclosures alone are not enough.
The proposal is a preliminary step. The agency has not yet issued a rule, and litigation could shape the final outcome. Until then, companies that deploy AI systems should review their marketing and product claims about accuracy, objectivity, and capability, and document compliance choices, especially if they are subject to state AI laws.
In short, the FTC’s July 7 proposal signals that hidden steering of AI outputs toward undisclosed goals may be deemed deceptive under federal law, that state‑law compliance does not excuse such conduct, and that clear, prominent disclosure is required to avoid violating Section 5 of the FTC Act.
