Requires synthetic content creations system providers to include provenance data on certain content made available by the provider

Showing the cosponsors who are on Agora, A to Z. The official record may list more.

• Democratic8
• Republican1

This bill has cosponsors from more than one party.

Shown in neutral colors by design; Agora is non-partisan.

◆ The sponsor's argument

Why Andrew Gounardes says this is needed

This legislation seeks to ensure transparency and accountability in the use of generative artificial intelligence (AI) by requiring the attach- ment and preservation of metadata by synthetic content creation systems, large public-facing platforms, and state agencies. The rapid advancement of AI technologies has made it increasingly difficult for consumers to distinguish between human-created and AI-generated content, raising concerns about misinformation, manipulation, and the erosion of trust in digital media. By mandating that AI providers include clear provenance data, this bill will enable the public to identify the origins and alterations of AI-generated content. The bill would only impose obli- gations on companies that provide users synthetic content and would not impose obligations on companies that merely make available model weights or source code. Companies would also not have to attach this kind of data to types of files which cannot support it. In line with current industry best practices, this light-touch bill would help establish clear demarcations between what is real and fake online. The data attached would be embedded within the content itself, allowing for an icon, tag, or other visible manifestation of detected provenance data that would, by default, come with all AI-generated or -modified content. The Stop Deepfakes Act imposes three requirements synthetic content creation providers have to attach metadata, large online platforms have to preserve metadata, and state agencies have to attach the data to the extent practicable. This kind of data is attached to images, videos, and audio all the time, and thus it is a relatively small burden to add a bit more data to ensure that AI-generated content is clearly indicated.

This is the sponsor's own case for the bill, quoted from their memorandum. It is not Agora's description of what the bill does.