Summary#
This bill would require companies that create or publish images, video, or audio that are created or substantially changed by generative AI to label that content. Labels must be clear, easy to find, and embedded in the content. Providers must also include a machine-readable disclosure inside the content that identifies the content as AI‑generated, the system and version used, the date and time created or modified, and other relevant information. Providers must make the content detectable by widely available detection tools or provide detection tools. Large online platforms must show these disclosures, not remove them, offer users a way to share provenance information, and try to limit misuse called the "liar's dividend." The Federal Trade Commission would enforce the labeling rules and could adopt standards and safe harbors. The bill also bans knowingly falsifying or removing disclosures, selling tools meant mainly to defeat disclosures, and knowingly distributing content that hides or fakes disclosures. There are exemptions for internal research use, some nonprofit libraries and archives, law enforcement activities, and good-faith research such as reverse engineering. The Attorney General, state attorneys general, and harmed providers or platforms may sue for violations. The bill directs NIST to create a working group with government, industry, academic, and civil society members to publish technical standards and guidance within a year.
What it means for you#
- If you use or see images, audio, or video online, the bill would require those items that were materially created or changed by generative AI to carry a visible label and embedded, machine-readable provenance information.
- Companies that build AI systems must add labels, embed machine-readable data, and help platforms detect AI content.
- Large online platforms must display labels, not remove them, give users options to share provenance data, and follow accessibility and anti-misuse guidance when technically and economically feasible.
Expenses#
- The bill does not include a government cost estimate or budget for implementation. No publicly available information.
- Penalties and remedies are specified in the bill: the Attorney General may seek statutory damages (the bill allows election of up to $2,500 per act of circumvention or up to $25,000 per violation, with possible tripling for repeated violations within 3 years). Courts may reduce awards for innocent violations and remit damages for certain nonprofit entities. The bill allows courts to issue injunctions and order impoundment or destruction of devices used in violations.
Proponents' View#
- The bill aims to increase transparency about digital content so people can tell when images, audio, or video were created or changed by generative AI.
- It would create consistent, machine-readable provenance standards and detection tools so platforms and users can verify origin and modifications.
- Supporters include measures to protect accessibility, research, and legitimate uses, and to limit harms like the "liar's dividend," where bad actors falsely claim real content is AI-generated.
Opponents' View#
No publicly available information.