FTC Proposes Rules on AI Systems That Alter Outputs Away from User Goals

The document targets generative AI and large language models and requires companies to disclose when systems steer outputs toward objectives different from those users reasonably expect. Forbes reported that the proposal states AI companies have represented that their systems aim to produce the best output that faithfully and accurately achieves users’ stated objectives.

It adds that consumers therefore have a reasonable expectation that AI systems will give truthful and accurate outputs.

The proposal notes that an AI company might be tempted to alter or steer output contrary to those expectations. It concludes that such steering is likely to deceive consumers in violation of Section 5 of the FTC Act, which prohibits deceptive acts or practices.

Companies can avoid potential deception by making truthful representations that clearly state they are prioritizing objectives different from those consumers requested, the proposal states.

The document remains open for public feedback until July 31, 2026.