California's AI Transparency Act takes effect for GenAI providers
Ship AI images, video, or audio to 1M+ California users and today you owe a free detection tool, C2PA provenance, and labels — at $5,000 a day per gap.
Copy markdown
What you owe as of today
Any generative-AI system with 1M+ monthly California users must now run a free, public AI-detection tool (web UI and API, no logging), embed a latent C2PA-style provenance marker in every AI image, video, and audio file, and let users add a visible “AI-generated” label. Text output is exempt.
The math on non-compliance
Penalties run $5,000 per violation, and each day out of compliance counts as a separate violation — so one missing watermark left unfixed for a month is $150,000. The statute has no first-offense grace period.
Wrapping someone else's model won't save you
The duties bind the provider of the GenAI system and flow down through mandatory contract terms to every licensee, so building on a third-party API doesn't move the obligation off you. Under the 1M-user line, you're out of scope — for now.
The deadlines behind the deadline
Providers must comply today; large online platforms and GenAI hosting platforms get until Jan 1, 2027, and capture-device makers until Jan 1, 2028. The same day, the EU AI Act's transparency rules begin enforcing too — so a global launch now faces provenance and AI-labeling duties on both sides of the Atlantic at once.