SB 942, amended by AB 853, has come into force. Beyond one million monthly users, a provider must mark its images, videos and sounds, and publish a verification tool.
The California AI Transparency Act (SB 942), as amended by AB 853, has been in force since 2 August 2026. It targets general AI providers that are accessible to the public in California and exceed one million visitors or monthly users—a threshold that focuses on large platforms.
The text imposes two levels of marking. The latent marking consists of source metadata embedded in the file, indicating the system at the origin of the content; it must be applied as soon as technically feasible and reasonable. Manifest marking is an option offered to the user to display a visible indication on an image, video or sound generated or modified by IA. In addition, there is an obligation to maintain a free and public detection tool, allowing anyone to check if content comes from the system concerned. Finally, licensing contracts must preserve these capabilities among third parties who redistribute the model.
The civil penalty is set at $5,000 per offence, each day counted for a separate offence. AB 853, signed in October 2025, had shifted the application date to August 2026 and added obligations for hosting platforms, which will take effect on 1 January 2027.
Read the full article on the original source
Go to the source →