California law mandates hidden provenance tags for AI-generated media

California’s AI Transparency Act, which took effect on Aug. 2, obliges major generative-AI companies—those with more than one million monthly users and public access in the state—to embed a hidden provenance record in every AI-generated image, video or audio file. The latent disclosure must, when feasible, list the provider’s name, the AI model and version, the creation or alteration date and a unique identifier, and it must be difficult to remove.

Providers also have to offer a free detection service that can confirm whether a file was produced by their system, while restricting the collection and retention of users’ personal data. Beginning Jan. 1, 2027, large social media, file-sharing and search platforms must detect and display any embedded provenance data, and they may not strip it when technically possible. From Jan. 1, 2028, newly sold phones, cameras and voice recorders in California must embed similar hidden tags by default, indicating manufacturer, device model and timestamp. The law aims to give consumers a way to trace the origin of synthetic media, though it does not verify the truthfulness of the content and leaves gaps for media from smaller providers or older files.