The Federal Trade Commission has issued a proposed policy statement addressing AI models that steer outcomes toward undisclosed ideological objectives.
The FTC issued the proposal on July 1, 2026, signalling a significant escalation in federal scrutiny of how artificial intelligence systems handle user requests.
According to the FTC, the practice of steering AI outputs may compromise accuracy and run contrary to reasonable consumer expectations about how AI models function.
The proposed policy statement follows Executive Order 14365, which directed the agency to clarify how Section 5 of the FTC Act applies to AI models and emerging state laws regulating those models.
Consumers generally expect AI systems to strive for accurate responses and faithfully carry out their requests, subject to ordinary technological limitations, the FTC contends.
The FTC argues these expectations are reasonable given the inherent nature of AI systems and explicit or implied representations that companies have made about those systems for years.
AI developers may engage in “steering” outputs, the FTC says, by deliberately configuring systems to advance objectives other than accuracy, such as achieving ideological goals or avoiding controversial topics.
The proposed policy statement specifically points to Colorado’s recently enacted Artificial Intelligence Act, which can hold companies liable for discriminatory AI outputs resulting from customer use of those products.
The FTC takes the position that compliance with a state requirement does not automatically eliminate potential federal deception concerns if consumers are not adequately informed about how those requirements affect outputs.
The proposed policy statement also suggests that state laws compelling modifications to AI outputs could, in some instances, conflict with federal consumer protection principles.
Although the statement does not provide a formal preemption rule, its discussion of emerging state AI laws appears to reflect an intent to establish a national AI policy framework that limits the patchwork of state-level regulation.
The FTC sets a high bar for what constitutes adequate disclosure of output steering, making clear that burying such information in terms of service is insufficient.
A one-time disclosure that is “subsequently hidden” would not meet the standard either, with the regulator noting that the further a practice strays from user expectations, the more prominent any disclosure must be.
Ordinary AI mistakes, including wrong answers caused by real technical limitations commonly known as hallucinations, do not on their own violate Section 5 of the FTC Act.
Blocking illegal content and preventing cyberattacks also do not raise concerns under the proposed framework, though overstating how rarely a system makes mistakes could still be considered deceptive.
The comment deadline for the proposed policy statement was July 31, 2026, with comments accepted at regulations.gov under Docket No. FTC-2026-0859, referencing “AI Policy Statement; Matter No. P264200.”