California's AI Transparency Act became operative on Aug. 2, 2026, requiring covered generative-AI providers to offer a free detection tool and place difficult-to-remove provenance disclosures in generated images, video, and audio. Later phases add provenance duties for large online platforms in 2027 and capture devices first produced for sale in California in 2028.
California's AI Transparency Act became operative on Aug. 2, 2026, after a 2025 amendment delayed the original start date. The law now requires covered providers of widely used generative-AI systems to support detection and provenance disclosures for generated or altered images, video, and audio.
What applies to generative-AI providers
The law defines a covered provider as a person or company that creates a generative-AI system with more than 1 million monthly visitors or users that is publicly accessible in California. Covered providers must make a free, publicly accessible detection tool that can assess whether their own system created or altered submitted media and can return system provenance data without exposing personal provenance data.
Providers must also offer users an option for a clear, human-visible disclosure and must include a latent disclosure in generated image, video, or audio content. The latent disclosure is intended to carry information such as the provider, system and version, creation or alteration time, and a unique identifier. The statute qualifies some requirements by technical feasibility and ties disclosures to widely accepted industry standards.
California can pursue a civil penalty of $5,000 per violation, and each day of noncompliance can count as a separate violation. The law assigns enforcement to the state attorney general, city attorneys, and county counsel rather than creating a private consumer action.
Platform and device phases follow
Assembly Bill 853 expanded the original 2024 law and established two later phases. Starting Jan. 1, 2027, large online platforms must detect standards-compliant provenance data, make system provenance information available through their interfaces, and avoid knowingly stripping compatible provenance data or digital signatures when technically feasible. GenAI hosting platforms also face restrictions on knowingly offering systems that do not place the required disclosures.
Starting Jan. 1, 2028, manufacturers must support latent disclosures for capture devices first produced for sale in California on or after that date. The enacted text requires a user option and default embedding, subject to technical feasibility and widely adopted specifications. This is narrower and more precise than saying every phone or camera made in California must retain provenance.
The EU comparison
The Aug. 2 date aligns with the European Union's AI Act Article 50 transparency obligations. The European Commission says providers must add machine-readable marks that help detect AI-generated or manipulated content, while deployers must disclose deepfakes and some AI-generated public-interest material. The EU framework has a different scope and enforcement structure, so California compliance does not establish EU compliance.
For product and ML teams, the shared engineering problem is maintaining trustworthy provenance across generation, export, editing, storage, and distribution. The laws do not mandate one identical implementation, but both make machine-readable marking and user-accessible disclosure part of the compliance surface.
Key Points #
- 1California's Aug. 2 phase requires covered providers with more than 1 million monthly users or visitors to offer a free detection tool and support manifest and latent disclosures for generated media.
- 2From Jan. 1, 2027, large online platforms must detect and expose compatible provenance data; GenAI hosting platforms also gain disclosure-related duties.
- 3From Jan. 1, 2028, qualifying capture devices first produced for sale in California must offer and default to latent disclosure, subject to technical feasibility and adopted standards.
Scoring Rationale #
The operative law directly affects large generative-AI providers and establishes phased provenance requirements for platforms and capture devices. It is relevant to media-generation and distribution pipelines across a major market, while its scope and technical-feasibility qualifications limit the immediate reach.
Sources #
Primary source and supporting public references used for this report.
View 3 more sources #
Governor Newsom signs bills to crack down on sexually explicit deepfakes and require AI watermarkinggov.ca.govCommission publishes guidelines on transparency obligations for providers and deployers of certain AI systemsdigital-strategy.ec.europa.euCalifornia Leads US With New AI Transparency Lawkqed.org
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.