The Federal Trade Commission opened comments on a proposed policy statement about “suppression of accuracy” in artificial intelligence systems.
The Federal Register published the notice on July 7. Comments are due July 31, 2026, under docket FTC-2026-0859. The agency says the proposed statement concerns how Section 5 of the FTC Act applies to companies that market AI systems.
The core theory is consumer deception. The FTC says AI companies often represent, explicitly or implicitly, that their systems aim to produce the best output possible for users’ objectives within technical and resource limits. If a company secretly steers an AI system toward unexpected objectives instead, the agency says that can mislead consumers.
The statement is not a final rule and does not create a new AI statute. It is a proposed enforcement-policy statement and request for comments.
The policy target is hidden objective shifting
The FTC is not saying every model output must be perfectly accurate.
The statement recognizes that AI systems pursue multiple objectives such as clarity, relevance, succinctness, and accuracy. It also notes that users can ask for intentionally inaccurate or entertainment-focused output.
The proposed problem is different: a company markets an AI system as solving the user’s task, but quietly changes the system’s objective so the output serves something else without a clear disclosure. The agency frames that as a mismatch between what consumers reasonably expect and how the system is actually designed to behave.
The notice explicitly discusses state AI laws as one possible pressure point. The FTC says companies may try to alter outputs to comply with state rules, but that state-law pressure would not automatically protect a company from Section 5 deception analysis.
AI safety teams should read the disclosure angle carefully
This proposal sits in a tense place.
AI systems do need safety policies, refusal rules, ranking choices, and product constraints. Those constraints often change outputs away from what a user literally requested. The FTC’s proposed statement is about whether the company has represented the product honestly enough for consumers to understand those objectives.
That means the practical work is documentation and user-facing clarity. If an AI system prioritizes safety, compliance, brand policy, or domain-specific guardrails over raw task completion in some cases, companies need to decide what has to be disclosed, where, and how.
The statement also raises a measurement problem. If an output was altered by a safety policy, a compliance policy, a ranking policy, or a model limitation, teams need logs and evaluation language that can distinguish those causes. Otherwise the company may not be able to explain how the system behaved when regulators, customers, or courts ask.
