New California Law Will Require AI Transparency and Disclosure Measures

On September 19, 2024, California Governor Gavin Newsom signed into law the California AI Transparency Act, which will require providers of generative artificial intelligence (AI) systems to: (a) make available an AI detection tool; (b) offer AI users the option to include a manifest disclosure that content is AI generated; (c) include a latent disclosure in AI-generated content; and (d) enter into a contract with licensees requiring them to maintain the AI system’s capability to include such a latent disclosure in content the system creates or alters. The California AI Transparency Act goes into effect January 1, 2026, and is the nation’s most comprehensive and specific AI watermarking law.

Key Definitions

The law’s provisions apply to “covered providers,” defined to mean “a person that creates, codes, or otherwise produces a generative artificial intelligence system that has over 1,000,000 monthly visitors or users and is publicly accessible” within California. Additionally, a “generative artificial intelligence system” is defined as:

An artificial intelligence that can generate derived synthetic content, including text, images, video, and audio, that emulates the structure and characteristics of the system’s training data.

The law also defines “artificial intelligence” to mean:

An engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.

Covered providers will be required to comply with the law as of the date it goes into effect; i.e., January 1, 2026.

Requirements for Covered Providers

The law’s transparency requirements will require covered providers to make disclosures about the use of generative AI systems. First, covered providers will be required to make available—at no cost to a user—AI detection tools that will allow users to assess whether image, video, or audio content has been created or altered using a generative AI system. Subject to certain technical requirements and privacy protections, these tools will also be required to provide system provenance data regarding the content, in order to allow users to verify the devices, systems, or services used to generate such content and content authenticity.

Additionally, covered providers will be required to provide users with both “latent” disclosures in AI-generated content and an option of including a “manifest” disclosure in such content. Manifest disclosures—meaning disclosures on AI-generated content that are easily perceived, understood, or recognized by a natural person—will be required to identify the content as AI-generated in a clear, conspicuous, appropriate, and permanent manner. Latent disclosures—those that are not manifest, but are instead present in the metadata of AI-generated content—will be required to convey (to the extent technically feasible and reasonable, either directly or through a link to a permanent website) the covered provider’s name, the name and version number of the generative AI system, the time and date of the content’s creation or alteration, and a unique identifier. The latent disclosures must also be detectable by the AI detection tool, consistent with widely accepted industry standards, and must be either permanent or extraordinarily difficult to remove.

Finally, covered providers that license their generative AI systems to third parties will be required to ensure that licensees maintain these disclosure requirements. If covered providers know that third-party licensees are not capable of including such disclosures, they will be required to revoke their licenses within 96 hours.

The law will be enforced by the California Attorney General, a city attorney, or a county counsel, and provides for civil penalties of $5,000 per day if a covered provider is found to be in violation.

Takeaways for Interested Parties

California joins Colorado, Utah, and Illinois in requiring transparency regarding the use of AI. However, unlike those other states, California’s law is the first AI law to create comprehensive and specific requirements related to watermarking. Companies developing generative AI systems should be mindful of these specific technical requirements as they invest time and resources developing technology within scope of this law. Licensors and licensees of cover AI systems should consider updating their agreements to address these contractual requirements.