FTC Proposes Rules Against Hidden AI Output Steering

The Federal Trade Commission has published a proposed policy statement that would treat hidden manipulation of AI system outputs as a likely violation of federal consumer protection law, even when companies adjust responses to comply with state regulations.

Under the draft policy released July 7, 2026, AI companies that steer their systems toward goals users don't expect—such as ideological alignment, equity objectives, or avoidance of sensitive topics—would be considered deceptive under Section 5 of the FTC Act, which prohibits unfair or deceptive acts or practices. The proposal applies regardless of whether the steering is done for profit, political reasons, or legal compliance.

According to Spencer Fane, the law firm that first reported the details, the FTC's position rests on the premise that AI marketing creates an implicit promise: that systems will deliver the most accurate answer possible within technical limits. When companies adjust outputs for reasons beyond accuracy without clear disclosure, they break that promise.

State law compliance offers no safe harbor

The most contentious element of the proposal targets state AI regulations directly. The FTC states that complying with state law does not excuse deception under federal standards. The agency specifically names Colorado's revised Artificial Intelligence Act (S.B. 26-189), arguing that state laws requiring output adjustments to avoid discriminatory outcomes may conflict with Section 5's consumer protection mandate.

The FTC goes further, asserting that such state laws are likely preempted because they "obviously conflict" with federal prohibitions against consumer deception. This sets up a potential legal clash between federal and state authority over AI governance.

Disclosure requirements and technical limits

Companies can steer outputs if they inform users clearly and prominently. Burying disclosures in terms of service or showing them once in fine print does not meet the standard. The more a practice deviates from user expectations, the more prominent the disclosure must be.

The policy distinguishes between intentional steering and technical failures. Ordinary AI errors—often called hallucinations—do not violate Section 5 on their own, nor does blocking illegal content or preventing cyberattacks. However, overstating system reliability could still constitute deception.

Why it matters

This proposal extends beyond foundational AI developers to any company deploying consumer-facing AI systems. Under third-party risk management principles, businesses using AI tools from other providers could face compliance obligations even if they don't build the underlying models. The preemption argument also creates immediate tension for companies subject to state AI laws, forcing them to navigate potentially conflicting requirements while courts resolve the federal-state authority question. The FTC's approach could reshape how companies balance accuracy, compliance, and transparency in AI deployment.

Broader implications

The proposed policy responds to Executive Order 14365, issued December 11, 2025, which called for a unified national AI framework instead of state-by-state regulation. The FTC's statement also references a separate executive order on fair lending that challenges disparate impact liability theories.

Spencer Fane notes that the FTC's preemption logic could extend to other regulatory contexts, potentially allowing future administrations to use consumer protection law as either a shield against state regulations or a sword to override state policies on different grounds.

Public comments on the proposal are due July 31, 2026, through regulations.gov under Docket No. FTC-2026-0859. Companies with stakes in how the final policy defines adequate disclosure, expected objectives, or preemption should consider submitting feedback before the deadline.

These details were first reported by Spencer Fane attorneys Jack Amaral, Greg Ewing, Kirstin D. Kanski, and Mike G. Silver.