The Federal Trade Commission’s (FTC) proposed policy statement puts a new consumer-protection frame around AI model behavior: if an AI company represents that its system is designed to deliver accurate, objective, or user-directed outputs, the company may create a reasonable consumer expectation that the system is trying to provide the best answer it can within its technical limits. The FTC’s concern is that an AI provider could secretly steer outputs toward undisclosed objectives, including ideological objectives, while still marketing the system as accurate, useful, or fit for the user’s task. In the FTC’s view, that kind of hidden steering may be deceptive under Section 5 of the FTC Act, even if the company says it is doing so to comply with state law. 

The proposed statement is especially notable because it connects AI accuracy, disclosures, and state AI regulation in one enforcement theory. The FTC says AI products are not exempt from Section 5, and points to prior enforcement involving deceptive AI-related claims about performance, efficacy, and product capabilities. The proposal also aims at state AI laws that the FTC believes could pressure companies to alter model outputs, including Colorado’s revised AI law, which the statement says may create incentives for companies to suppress accuracy or prioritize other objectives without telling users. The FTC further notes that state law may be impliedly preempted where it is inconsistent with or otherwise conflicts with the federal consumer protection framework.

For AI developers and enterprise customers, the practical takeaway is that disclosures, product claims, and governance controls around model behavior are going to matter. The FTC acknowledges that companies can shape user expectations through truthful, non-misleading disclosures, but warns that those disclosures must be clear, conspicuous, prominent, and strong enough to counter any contrary impression created by marketing or product design. Comments on the proposed policy statement are due July 31, 2026, giving AI companies, customers, and other stakeholders a short window to weigh in on how the FTC should draw the line between permissible model governance and deceptive manipulation of AI outputs. Read the full statement here.