California AI Transparency Act: your AI images are stamped

Today is the day the California AI Transparency Act starts biting, and the interesting part is that it probably doesn't regulate you. It regulates the generator you used to make your product photos. Starting now, the image that came out of that tool carries a permanent, machine-readable stamp saying which system made it and when — and there's a free public tool anyone can use to check.

What actually happened

SB 942 passed in 2024 with a January 1, 2026 start date. AB 853 pushed the operative date to August 2, 2026 — the same day the EU AI Act's Article 50 transparency rules kick in, which is not a coincidence.

The obligations land on a "covered provider": anyone who produces a generative AI system with more than 1,000,000 monthly visitors or users that's publicly accessible in California. That's OpenAI, Google, Adobe, Canva, Meta — not your shop. As of today those providers must:

• Embed latent disclosure in every image, video, and audio file their system generates or substantially alters — provider name, system version, creation timestamp, unique identifier. In practice this means C2PA.
• Offer a free public detection tool with both a web interface and an API, that accepts uploads and URLs, and doesn't retain what you submit or collect your personal information.
• Offer manifest disclosure — a visible "AI-generated" label users can opt into, appropriate to the medium.

Civil penalties run $5,000 per violation, with each day counted separately.

Then AB 853 stacks the part that actually reaches your storefront. From January 1, 2027, large online platforms above 2 million monthly users must detect provenance data, surface it in the UI, name the system that created the content — and not strip provenance from uploaded files. From January 1, 2028, cameras sold in California ship with latent disclosure embedded by default.

Why the California AI Transparency Act matters for your business

You aren't the regulated party. You're the one whose catalog gets read.

Every AI-generated hero image, lifestyle shot, and background fill you ship from a major tool now travels with a durable label. In 2027, the platforms you sell and advertise on are legally required to detect that label and show it. Your product photography quietly becomes an attribute a marketplace, a search engine, or a buyer can query — and platforms are barred from quietly scrubbing it for you.

That's a merchandising decision you should make on purpose rather than discover in a support ticket. Three things worth doing this month: inventory which images in your catalog came out of a generative tool, decide per category where AI imagery is fine (background swaps, size variants) and where it isn't (the actual product), and stop treating provenance metadata as something your image pipeline is free to strip on resize. Most CDNs and thumbnail generators drop metadata by default. That's fine today. It's a compliance liability for the platform sitting downstream of you in five months.

The pattern is the one we keep writing about: rules that nominally target frontier labs land on operators through the tooling. If your image pipeline is yours, this is an afternoon of work. If it's four SaaS products stapled together, it's four support tickets.

Sources: California SB 942 (leginfo), California AB 853 (leginfo).