If you used AI to put a model in your product photos, Amazon now wants a hidden tag inside the file. Since July 22, 2026, sellers must embed the keyword contains-synthetic-performer in the image metadata before uploading it to a listing or A+ Content. No tag, no compliance. And Amazon is not the only one asking.
Three separate rules landed inside nine weeks. Most sellers have not noticed any of them, because none of them break your listing. They just quietly make it non-compliant.
What Amazon actually requires
Amazon published the policy in Seller Central on July 22, 2026 under the title “You must now tag AI-generated people in listings and A+ content.” The wording is specific. According to Amazon’s own announcement, sellers “must use an IPTC-compatible metadata editor to add the keyword ‘contains-synthetic-performer’ to the dc:subject (XMP) field” of the image or video file before adding it to listings and A+ Content in any worldwide store.
When the tag is present, Amazon says it will add a customer-facing indicator noting that the image includes AI-generated people.
The trigger is narrow. Per Amazon’s announcement, the requirement applies only to photorealistic AI-generated people. You do not need the tag if the media:
- Features only real people, even if edited with AI tools
- Features characters from movies, TV, streaming, documentaries, or video games
- Contains no people at all
- Contains people who are not photorealistic (illustration, cartoon, stylized)
So an AI-upscaled photo of your actual founder is fine untagged. An AI-generated model wearing your hoodie is not.
What to do this week
Audit your last 12 months of lifestyle images and A+ modules. Flag every asset where a human appears and nobody on your team can name that human. Those are your candidates. Then add the keyword using any IPTC-capable tool (Adobe Bridge, ExifTool, Photo Mechanic all write to dc:subject) and re-upload.
One trap worth knowing: metadata is fragile. It gets stripped by resizing, format conversion, screenshotting, and plenty of DAM systems. A compliant master file can arrive at Amazon non-compliant because a resize step in the middle silently wiped the field. Verify the tag on the exact file you upload, not the one you exported.
The law that caused it
Amazon did not do this for fun. New York enacted a first-in-the-nation synthetic performer disclosure law, signed by Governor Kathy Hochul as S.8420-A/A.8887-B, which amended New York General Business Law § 396-b and took effect June 9, 2026.
The law defines a “synthetic performer” as a digital asset created or modified by generative AI or a software algorithm that is meant to give the impression of a human performance, where the asset is not recognizable as any identifiable natural performer. Anyone dealing in property or services who creates a commercial advertisement containing one must conspicuously disclose it, where they have actual knowledge.
Penalties, per the statute: $1,000 for a first violation, $5,000 for each subsequent violation. There is no private right of action, so enforcement sits with the state rather than plaintiff firms. Law firm analyses from Reed Smith and Cooley both note the statute never defines “conspicuous,” which means placement and formatting norms will be settled by enforcement, not by the text.
The law is a New York law. Amazon applied the policy globally anyway. That is the part sellers keep missing: platform policy travels faster and wider than the statute behind it.
Two more deadlines you probably missed
California SB 942. The California AI Transparency Act was signed in September 2024 with a January 1, 2026 start date. AB 853, signed October 13, 2025, pushed the operative date to August 2, 2026 to line up with the EU. It became operative nine days ago. It targets large generative AI providers with more than one million monthly users, requiring free AI detection tools plus visible and embedded latent disclosures on generated content. Further obligations for hosting platforms, large online platforms, and device manufacturers phase in through 2027 and 2028.
You are probably not the regulated party here. Your image generator is. But the practical effect flows downhill: the tools you use will start stamping provenance into outputs whether you ask them to or not.
EU AI Act Article 50. Also August 2, 2026. Article 50(4) transparency obligations covering deepfake labelling became applicable on that date. Article 50(2) requires providers of AI systems generating synthetic image, audio, video or text to mark outputs in a machine-readable format so they are detectable as artificially generated. The AI Omnibus provisional agreement from May 2026 gives generative systems already on the market before August 2 until December 2, 2026 to meet the machine-readable marking requirement.
Same theme as the wave of AI chatbot disclosure rules that hit stores earlier this year. Regulators are converging on one idea: if a machine made it, the machine has to say so.
Build the check into your pipeline, not your memory
Manual tagging fails the moment you scale past a few dozen SKUs or hand production to an agency. Three things worth setting up now:
1. Tag at creation, not at upload. Whoever generates the asset adds the keyword immediately. If that is an agency or freelancer, put it in the brief and the deliverable spec.
2. Add a verification step before publish. ExifTool will read dc:subject from the command line, so a one-line check can run against a folder before anything goes to Seller Central. Treat a missing tag on a flagged asset the same way you would treat a missing barcode.
3. Keep a register. A simple sheet listing which SKUs use synthetic people, which tool generated them, and when they were last verified. If enforcement ever knocks, “we have a documented process” is a very different conversation from “we think we tagged most of them.”
None of this is expensive. It is a metadata field and a checklist. What it protects against is the scenario where a marketplace suppression or a state penalty arrives long after the person who made the image has left the company.
The bigger pattern
Compliance requirements are increasingly showing up inside the file rather than on the page. Metadata tags, machine-readable marks, provenance credentials. You cannot see them in a screenshot, which is exactly why they get lost.
Sellers who already handle regulatory detail well should recognise the shape of this. It is the same operational discipline that the EU packaging rules landing this week demand, and the same lesson from accessibility cases where partial compliance still lost in court. Partial does not count. Documented does.
The rule itself is trivial to follow. Missing it is trivial too, which is the whole problem.
Sources
- Amazon Seller Central, “You must now tag AI-generated people in listings and A+ content,” July 22, 2026 — sellercentral.amazon.com
- Amazon Seller Central Help, “How to tag media that contains an AI-generated person” — sellercentral.amazon.com
- Office of Governor Kathy Hochul, announcement on first-in-the-nation AI advertisement disclosure law — governor.ny.gov
- New York State Senate, Bill S8420A (amending GBL § 396-b) — nysenate.gov
- Reed Smith, “Fake performer, real penalty: What advertisers need to know before June 9” — reedsmith.com
- Cooley, “New York Enacts ‘Synthetic Performer’ Disclosure Law for Advertisements,” January 29, 2026 — cooley.com
- Morgan Lewis, “New California AI Disclosure Rules Become Operative,” August 2026 — morganlewis.com
- Troutman Pepper, “California AI Transparency Act Amendments Signed Into Law” (AB 853) — troutmanprivacy.com
- European Commission, “Transparency obligations under Article 50 of the AI Act” — digital-strategy.ec.europa.eu
- EU Artificial Intelligence Act, Article 50 text — artificialintelligenceact.eu
- GeekSeller, “Amazon Introduces New Rules for AI-Generated Images (July 2026)” — geekseller.com
Last reviewed: August 11, 2026
Affiliate disclosure: this site may earn commissions from links in this article, at no extra cost to you. It does not affect which tools or policies we cover.
