The California AI law known as the AI Transparency Act became operative on August 2, 2026, requiring large generative AI providers to place hidden provenance disclosures in covered images, video, and audio. It also mandates free detection tools, creating verification duties for technology companies, media platforms, publishers, and users handling synthetic content.
Key Takeaways
- The California AI law became operative on August 2, 2026.
- Covered providers must embed latent disclosures in qualifying synthetic media.
- Providers must offer free tools to check media created or altered by their systems.
- Platform requirements begin in 2027, followed by capture-device rules in 2028.
The California AI law requires generative AI systems to attach hidden source information to synthetic media. It covers systems with more than one million monthly users that are publicly accessible in California.
The measure focuses on provenance, not a visible stamp on every file. Providers must embed a latent disclosure in AI-generated images, video, audio, or combined formats that their detection tools can identify.
California created the framework through Senate Bill 942. Assembly Bill 853 moved the operative date from January 1 to August 2, 2026, and added future duties for platforms, AI hosting services, and device manufacturers. The update joins other new California laws affecting compliance and consumer information.
California AI Law Places Provenance Inside Media Files
When technically feasible and reasonable, the latent disclosure must convey the provider’s name, the generative AI system’s name and version, the creation or alteration date and time, and a unique identifier.
The information may appear in the file or through a permanent website. The disclosure must follow widely accepted standards and be extraordinarily difficult to remove when technically feasible.
Visible labeling works differently. Providers must offer an option to identify content as AI generated, but the law does not require users to select it.
A file can contain machine-readable provenance without displaying a badge or watermark. Editors may need to inspect embedded information instead of relying on the screen.
The law also covers licensed systems. Contracts must require licensees to preserve the disclosure function. A provider that learns it was disabled must revoke the license within 96 hours, and use must stop.
Free Detection Tools Create a Public Check
Covered providers must offer a free AI detection tool. Users must be able to upload media or submit a URL to assess whether that provider’s system created or altered it.
The tool must report detected system provenance data without revealing personal provenance data. It must also support an application programming interface, allowing outside services to run checks without visiting the provider’s website.
Privacy restrictions limit data collection. Providers cannot retain submissions longer than necessary, retain personal provenance data, or collect users’ personal information. Contact details may be kept when users opt in with feedback.
The process is provider-specific, not a universal test for synthetic media. A negative result from one company’s tool does not establish that a file was created without AI. It may only show that the tool did not detect provenance connected to that provider’s system.
For publishers and platforms, these tools may support routine screening. They can indicate a declared system origin or modification history, but they do not determine whether every statement, scene, or depicted event is accurate.
Related AI and social media rules address other forms of disclosure and platform responsibility. The California AI Transparency Act centers on provenance attached to generated or altered media.
Rules Expand to Platforms and Devices
Beginning January 1, 2027, large online platforms with more than two million unique monthly users during the preceding 12 months must detect qualifying provenance data attached to distributed content.
Platforms must indicate when system provenance data is available and let users inspect it. They also may not knowingly strip compliant system provenance data or digital signatures when preservation is technically feasible.
Generative AI hosting platforms face a related rule on the same date. They may not knowingly make available a generative AI system that fails to place the disclosures required by the act.
On January 1, 2028, requirements extend to certain capture devices first produced for sale in California on or after that date. Covered cameras, mobile phones with built-in cameras or microphones, voice recorders, and similar devices must offer latent disclosures and embed them by default when technically feasible and consistent with widely adopted specifications.
Violations can bring a $5,000 civil penalty per violation. Each day of noncompliance by a covered provider, large online platform, or capture-device manufacturer may count separately. The attorney general, a city attorney, or county counsel may bring an enforcement action.
For organizations handling generated or altered media, the California AI law shifts attention from visible labels alone to embedded provenance, detection access, privacy controls, licensing terms, and preservation across distribution systems.
Frequently Asked Questions
When Did the California AI Law Become Operative?
The California AI law officially became operative on August 2, 2026. AB 853 delayed the earlier January 1 date established under SB 942.
Does Every AI-Generated File Need a Visible Watermark?
No. Covered providers must embed a latent disclosure in qualifying images, video, and audio, while offering users an option to add an easily perceived label.
Which Providers Are Covered?
The law covers providers producing generative AI systems with more than one million monthly visitors or users that are publicly accessible in California. The standard concerns reach and availability rather than headquarters alone.
What Does the Detection Tool Show?
The tool must assess whether the provider’s system created or altered submitted media and report detected system provenance data. It cannot output personal provenance data.
What Requirements Come Next?
Large platforms and AI hosting services face additional duties beginning January 1, 2027. Certain newly produced capture devices become subject to disclosure rules beginning January 1, 2028.
