AI Systems Transparency Act

Developers of covered generative AI models would publish two versions of information: a clear disclosure for consumers and a more detailed one for independent researchers. Topics include data practices, safety measures, testing results, model capabilities and limitations, and changes between model versions.

The FTC, consulting with the National Institute of Standards and Technology, would set disclosure requirements for covered deployers of AI systems. The rules would be tailored to deployers and their systems, using requirements for developers as a guide.

The FTC would set criteria for identifying models or deployments likely to reach the user thresholds in the bill, and for designating a model or system as covered because it poses a significant safety risk. Before most designations take effect, the FTC would give the developer or deployer written notice and at least 30 days to respond; an imminent-risk designation could take effect immediately, with a later opportunity to respond.

The FTC would set rules requiring entities that significantly change a covered model’s behavior or capabilities to publish information about it. The Commission could tailor which disclosures apply to those entities.

Violations of the disclosure requirements would be treated as violations of FTC rules against unfair or deceptive acts or practices. The FTC would enforce the law using its existing powers under the Federal Trade Commission Act.