Does the US Require an AI Label on Virtual Try-On? SB 942 and the FTC, Explained

US retailers weighing AI disclosure often reach for the EU AI Act first, since it names the obligation directly. That rule reaches a US store selling into Europe, covered in an earlier post on the EU AI Act's disclosure requirement, but a domestic US store needs a different map. Two things changed on the US side in 2026, and they land on different parties.

What is SB 942 and what changed on August 2, 2026?

SB 942, the California AI Transparency Act, was signed in 2024 with an original start date of January 1, 2026. AB 853, signed October 13, 2025, pushed that date to August 2, 2026, deliberately lining it up with the EU AI Act's own enforcement timeline for high-risk systems.

From that date, a covered provider under SB 942 has three obligations: offer a free public tool that lets anyone check whether a piece of content came from its system, give users the option to add a visible manifest disclosure to AI generated images, and embed a latent disclosure, hidden provenance data, in content its system creates. A second wave of obligations aimed at large online platforms and AI hosting platforms is scheduled for January 1, 2027, under the same amendment.

Who does SB 942 actually apply to: platforms or retailers?

SB 942 defines a covered provider narrowly: an entity that builds a generative AI system with more than one million monthly users in California. That threshold sweeps in the handful of companies operating foundation models at that scale, Google, OpenAI, Anthropic, Meta, and comparable image and video generation platforms.

• A fashion retailer buying a try-on tool is a customer of a covered provider, not one itself. The one million monthly user threshold is built for scale, so a store running virtual try-on on its product pages does not carry SB 942's disclosure duties directly, even though the images it shows are AI generated.
• The obligation sits with whoever built the underlying model. Corlen's own generation engine runs on Google's Gemini image model family, which is squarely inside SB 942's scope as a covered provider in its own right. That does not transfer SB 942's specific duties down to a retailer using a tool built on top of it.
• This is a different structure from the EU AI Act. The EU rule puts its disclosure duty on the deployer, meaning the business showing the content to a shopper. SB 942 puts its main duties on the provider building the system. A US-only store can be effectively outside SB 942's direct reach while still needing to think about disclosure for an entirely separate reason, covered next.

What does the FTC require separately from SB 942?

SB 942 is a California statute naming AI specifically. The FTC's authority is older and broader: Section 5 of the FTC Act bars unfair or deceptive practices, and it does not need a new AI-specific law to reach an AI generated image that misleads a shopper about what they are buying.

In January 2026 the FTC stood up a dedicated AI enforcement unit and raised its maximum penalty to over 53,000 dollars per violation. Its updated Endorsement Guides now treat AI-generated content as its own disclosure category, separate from the older paid-endorsement disclosure a store may already run. A post marked as an advertisement does not also cover the AI generated question, and a label that says AI generated does not also cover the commercial relationship question. Both have to be present where either applies.

The FTC's standard for what counts as adequate disclosure has stayed consistent across years of guidance: clear, conspicuous, and visible at the moment a person encounters the content, not a click away in a bio or a policy page. A retailer showing a shopper an AI generated try-on result without saying so is closer to this exposure than to SB 942's.

What should a US fashion retailer actually do?

1. Label every AI generated image visibly, on the image itself or immediately beside it, rather than in a footer disclaimer a shopper is unlikely to read.
2. Say plainly that a virtual try-on result is a preview, not a guarantee of exact fit, color, or fabric, so the disclosure covers both the AI generated question and the accuracy expectation at once.
3. Do not rely on a vendor's own AI watermark or metadata as a substitute for a visible label on the retailer's own product page. SB 942's latent disclosure lives inside the file and is built for detection tools, not for a shopper glancing at a page.
4. Check whether the store also sells into the EU. If it does, the separate deployer obligation under the EU AI Act's Article 50 applies on top of anything covered here, not instead of it.

None of this is legal advice, and neither law hands a retailer a template to copy. A store with real exposure, meaning real EU sales or a marketing claim that could plausibly mislead a shopper about a product's actual appearance, should confirm its approach with counsel rather than assume either law's silence means no obligation exists.

How does Corlen handle AI-generated disclosure today?

Corlen labeled its outputs before either 2026 rule arrived. Every result on the in-store kiosk carries a visible AI GENERATED PREVIEW label directly on the image, and Corlen's terms state that a preview is an approximation, not a guarantee of exact fit, color, or fabric appearance. That was a product decision made for trust with shoppers, not a response to SB 942 or the FTC's enforcement push, and it happens to line up with what both now expect.

On a Shopify store, virtual try-on runs directly on the product page through Corlen's Shopify app, with the same visible label built into every result. A team building on WordPress or a custom stack gets the same generation and labeling behavior through the developer API instead, so the disclosure travels with the image rather than depending on a merchant remembering to add it separately.

For the EU side of this same question, see the earlier post on the EU AI Act's virtual try-on disclosure rule. For what actually happens to a shopper's photo rather than how the finished result gets labeled, see is virtual try-on safe.

See the label in place directly: try Corlen on your own photo, no account needed for a first result.