If you've built a moodboard on Midjourney, run product shots through an AI upscaler, or fed your own archive into a fine-tuned model to generate a lookbook, you've been operating in a legal void. As of January 1, 2026, that void has a floor under it. California's AB 2013, the Generative AI Training Data Transparency Act, requires the companies behind these tools to tell the public what their models were trained on. For an independent designer, that's not abstract policy. It's the first real window into whether the tool you rely on was built, in part, on scraped runway photos, stolen sketches, or someone else's copyrighted prints.

What changed

Starting January 1, 2026, developers of generative AI systems available to Californians must post documentation on their own websites describing the data used to train those systems, before the system goes public and again whenever there's a substantial update. The statute, codified at Civil Code sections 3110 and 3111, defines generative AI broadly, as systems that can generate synthetic content, text, images, video, and audio, that emulates the structure and characteristics of its training data. The disclosure has to be a high-level summary covering things like the sources of the datasets, intellectual property status, including use of copyrighted, trademarked, patented, or public domain data, with licensing details where applicable, and whether personal or aggregate consumer information is involved. Critically, the law reaches backward. Any generative AI system or service made available to Californians since January 1, 2022 is covered, casting a wide net over existing generative AI platforms as well as those which have yet to be built.

Not everyone is complying cleanly. OpenAI, Anthropic, and Google each published the required documentation on their websites by January 1, with varying degrees of detail, though none of them names the specific datasets behind their models. But xAI sued the California Attorney General on December 29, 2025, days before the law took effect, arguing the disclosures amount to a taking of trade secrets and compelled speech. In March 2026 a federal judge denied xAI's request to block enforcement while the case proceeds, finding the company hadn't shown it was likely to win, and xAI has appealed to the Ninth Circuit. So this law is live, but it's also being tested in court, and its ultimate shape isn't settled.

Who it affects

The statute's definition of "developer" is wider than it sounds. It covers a person, partnership, state or local government agency, or corporation that designs, codes, produces, or substantially modifies an artificial intelligence system or service for use by members of the public. "Substantial modification" includes a new version, new release, or other update that materially changes the system's functionality or performance, including the results of retraining or fine-tuning. That matters for designers, because it isn't just about OpenAI and Midjourney. If you commission a fine-tuned model trained on your own archive, license it out, or offer it as a public tool, whether a plugin, a custom Instagram filter, or a client-facing rendering app, you may become a "developer" under this law yourself. Most independent designers won't cross that line. But studios and small brands increasingly experimenting with custom AI pipelines for prints, textures, or virtual samples should know the threshold is lower than they'd assume.

On the other side of the ledger, this law gives designers concerned about their own work being absorbed into training sets a new, if imperfect, tool. The disclosures won't name specific designers or garments. They're high-level by design, and the law itself does not define "high-level," leaving open the types of information that must be disclosed, and provides no compliance mechanism. Still, for the first time, you can look at a platform's public disclosure and see, in general terms, whether it drew on copyrighted material, purchased or licensed data, or scraped publicly available content without much filtering. If you want to know whether your own designs are likely sitting in one of those datasets, I walk through that question in Your Designs Are Probably in an AI Training Dataset.

What to do about it

Read the disclosures before you commit to a tool, not after. If you're licensing an AI image generator for lookbooks, ad creative, or virtual samples, check the developer's published summary. It won't tell you whether your specific work is in there, but it will tell you the platform's general posture on copyrighted material, which matters if you're later asked to defend your creative process in a dispute.

Keep your own records. If any part of your design workflow uses a fine-tuned or custom model, even informally, document what you trained it on and how you use it. That paper trail protects you if licensing questions ever arise, and it positions you to comply quickly if you ever cross into "developer" territory yourself. The provenance tool is built for exactly this kind of record-keeping.

Don't expect enforcement muscle yet. AB 2013 does not include a specific enforcement mechanism. The legislative commentary suggests the law will likely be enforced under California's Unfair Competition Law, which authorizes enforcement by the California Attorney General, district attorneys, and other government prosecutors. There's also a private right of action under that statute, but only when a plaintiff has been injured and lost money or property as a result of violations, a high bar for most disputes.

Watch the xAI litigation. If that lawsuit succeeds in narrowing what developers must disclose, the transparency you're counting on could shrink fast. And watch the copyright cases running alongside it, which I map in the AI copyright litigation map. This law is a start, not a shield. Treat every disclosure as a data point, not a guarantee. I track AB 2013 and the rest of the state AI laws in the tracker.

Sources

This article is editorial analysis, not legal advice. For questions about your specific situation, consult a qualified attorney.