California AI Transparency Act.

Authored by Assembly Member Wicks, with Senator Becker as principal coauthor and Assembly Member Zbur as a coauthor, the California AI Transparency Act sets a staged regime of provenance and disclosure requirements that across GenAI developers, large online platforms, hosting platforms, and capture-device manufacturers. The measure structures implementation in three phases: the chapter becomes operative in late summer 2026, platform- and hosting-disclosure obligations begin in early 2027, and latent disclosures for devices first produced in California after 2028 are added later. It also establishes civil penalties, a mechanism for injunctive relief in certain enforcement scenarios, and a severability clause.

Key mechanisms include a redefined, broader set of terms to govern provenance and disclosure. The act adds obligations for large online platforms to detect provenance data that aligns with widely adopted specifications, provide a user interface that discloses the availability and details of provenance data, and enable users to inspect system provenance data through the platform UI, by download, or via a linked website or third party. It further directs that platforms should not knowingly strip system provenance data or digital signatures, to the extent technically feasible. GenAI hosting platforms are required not to knowingly host a GenAI system that fails to place disclosures. For capture-device manufacturers, starting in 2028, devices first produced for sale in California must offer a latent disclosure option containing the manufacturer name, device name and version, and creation timestamp, and must embed latent disclosures by default where technically feasible, subject to alignment with established standards.

The act expands enforcement provisions through a civil-penalty structure of five thousand dollars per violation, with each day of noncompliance treated as a discrete violation, and it permits injunctive relief and reasonable attorney’s fees for certain enforcement actions, including third-party licensee scenarios. Compliance is tethered to a phased operative timeline: the overall chapter becomes effective on August 2, 2026; the platform and hosting obligations take effect January 1, 2027; and device latent-disclosure requirements become operative January 1, 2028. The framework relies on references to “widely adopted specifications” from an established standards-setting body to govern provenance data, without naming a specific standards body.

Taken together, the measure delineates a multi-stakeholder regime that broadens the scope of California’s AI transparency framework by introducing platform-level provenance detection and disclosure, hosting-platform disclosures, and device-level latent disclosures for new-generation capture devices, while outlining a staged timeline, enforcement mechanisms, and the policy rationale of improving visibility into the provenance and authenticity of AI-generated or modified content. The approach situates provenance data within a standards-based ecosystem and raises questions about interoperability, feasibility, and verification procedures as implementation proceeds.