Amazon Seller News & Suspension Alerts (2026) | AMZ Sellers Attorney® Blog

Marketplace Compliance  |  Updated August 15, 2026

Amazon's AI-Generated People Rule: What the contains-synthetic-performer Tag Actually Requires

By Kenneth G. Eade, Esq. — Founding Attorney, AMZ Sellers Attorney, Ltd.  |  California Bar No. 93774  |  Practicing law since 1980

What Amazon actually changed on July 22, 2026

Amazon's seller notice is narrow and technical. Before you add an image or video to a listing or to A+ Content, you must use an IPTC-compatible metadata editor to write the exact keyword contains-synthetic-performer into the file's dc:subject XMP field. Where the tag is present, Amazon surfaces an indicator telling shoppers the media includes AI-generated people.

Three features of the rule deserve attention that the operational coverage has not given them.

First, it is embedded metadata, not a form field. Nothing in Seller Central asks you the question. The disclosure travels inside the file. That means the compliance failure mode is not a seller who refuses to disclose — it is a seller who disclosed correctly and whose asset pipeline destroyed the tag. Resizing tools, compression utilities, CDN transforms, and export presets routinely strip XMP on write. A brand that tags at the design stage and hands off to an agency that re-exports for Amazon's dimension specs can arrive at upload with a stripped file and no notice that anything happened.

Second, the identification duty is entirely yours. Amazon does not classify your assets for you. It has published a rule and moved the judgment call onto the account holder. Every borderline asset in your catalog is a decision you made, and it is a decision you will have to defend if it is ever questioned.

Third, Amazon applied it globally. The rule originates in a single U.S. state statute. Amazon extended it to all worldwide stores as a term of the Business Solutions Agreement. A seller in Frankfurt or Shenzhen with no New York nexus whatsoever is now bound — by contract, not by law.

When does the tag apply, and when does it not?

Amazon's carve-outs are broader than most sellers assume. You do not need the tag if your image or video:

• Contains no people at all — which describes the majority of product catalogs;
• Features only real people, even if those people were altered with AI tools (relighting, background replacement, retouching, cleanup);
• Features only characters from films, television, streaming content, documentaries, video games, or similar expressive works;
• Features people who are not photorealistic — illustration, cartoon, stylized rendering.

The rule targets one thing: a convincing human being who never existed. The AI lifestyle model in your A+ module holding your product. The invented spokesperson in your shoppable video. The synthetic "customer" in your comparison chart.

The exemption is not an exemption from liability

This is the part I would flag hardest for any brand that has been buying AI creative for the past eighteen months.

The carve-out for real people altered with AI tools takes you outside the tagging rule. It does not take you outside anything else. If your AI pipeline produced an image that resembles an identifiable real person — whether you intended it or not, and generative models do this without being asked — you have left the disclosure statute and entered right-of-publicity territory. New York's synthetic performer amendment says expressly that it does not alter rights under New York Civil Rights Law §§ 50, 50-f, and 51, which govern name, image, likeness, and digital replicas.

Those statutes are considerably more dangerous to a seller than a $1,000 disclosure penalty. They carry private enforcement. They carry damages. And the same New York legislative session that produced the tagging obligation strengthened the digital replica regime alongside it.

So the compliance posture is not "tag it or don't." It is: identify what is in the frame, and understand which body of law that answer puts you in. A synthetic person nobody recognizes is a tagging question. A synthetic person who looks like somebody is a likeness question. The second one does not get solved by adding a keyword.

What is the New York law behind this?

New York General Business Law § 396-b, as amended by S.8420-A / A.8887-B, took effect on June 9, 2026. Governor Hochul's office described it as first-in-the-nation. It requires any person in the business of dealing in property or services who produces or creates a commercial advertisement to conspicuously disclose that a synthetic performer appears in it, where that person has actual knowledge of the synthetic performer.

A "synthetic performer" is a digital asset created, reproduced, or modified by computer using generative AI or a software algorithm, intended to give the impression of an audiovisual or visual performance by a human performer who is not recognizable as any identifiable natural person.

Penalties are $1,000 for a first violation and $5,000 for each subsequent violation. The statute contains no express private right of action, which points enforcement toward the New York Attorney General rather than plaintiff-side litigation.

The unsettled question nobody has answered

Read that definition again. It turns on a performance. Whether a static product photograph of an AI-generated model standing next to a blender constitutes a "visual performance of a human performer" is genuinely unresolved. The statute does not define "performer" or "performance," and it does not define "conspicuous" either. Commentators have read the terms expansively enough to capture background extras; a narrower reading would confine the statute to something closer to advertising talent.

I raise this not as an argument for non-compliance. Amazon's requirement binds you contractually whatever § 396-b turns out to mean, and the tagging burden is trivial. I raise it because sellers are being told a state statute clearly governs their product photos, and it does not clearly do anything of the kind. If you are ever asked to defend a classification decision, the ambiguity is on your side of the ledger, not Amazon's.

Where the real risk sits: Account Health, not the fine

A seller reading only the New York penalty schedule will conclude this is a rounding error. That is the wrong frame.

An untagged qualifying asset is not primarily a state law violation. It is a listing content violation under the Business Solutions Agreement. Content violations move through Amazon's enforcement stack the way every other content violation does: image suppression, then listing suppression, then a policy violation posted to your Account Health dashboard, then — in a pattern, or in combination with unrelated flags — deactivation of the account and a hold on your disbursements.

That is a $1,000 statutory problem converting into a five- or six-figure commercial problem, and it converts through a channel most sellers are not watching.

There is a second-order risk worth naming. Rules of this kind become instruments. A competitor who wants your A+ Content taken down now has a new complaint category to use against you, and Amazon's complaint intake is not built to adjudicate whether the woman in your lifestyle image is real. We have defended enough pretextual IP complaints to know how this pattern develops.

A practical audit for the next thirty days

1. Inventory the exposure. Pull every main image, secondary image, A+ module, brand story asset, and shoppable video added since roughly early 2025. You are looking for photorealistic humans. If your catalog has none, you are finished — document that conclusion and move on.
2. Classify each human asset into one of three buckets: real person (no tag), fully AI-generated person (tag required), or uncertain provenance (investigate before deciding).
3. Treat "uncertain provenance" as the priority. Assets from a former agency, a marketplace stock library, or a freelancer who has since disappeared are the ones you cannot classify by looking. If you cannot establish that a real human sat for the image, you do not know which rule applies.
4. Verify the tag survives your pipeline. Write the keyword, run the file through your actual resize and export process, then re-open the XMP and confirm contains-synthetic-performer is still present in dc:subject. Do this once per pipeline, not once per file.
5. Fix your vendor paperwork. Creative services agreements should require the vendor to identify AI-generated humans in delivered assets, warrant that no delivered asset depicts an identifiable real person without a release, preserve embedded metadata through delivery, and indemnify you for breaches of the first three.
6. Do not over-tag as insurance. Labeling a real human model as a synthetic performer is an inaccurate statement about your advertising, it may conflict with that model's release, and it costs you conversion. Accuracy is the safe harbor, not maximal disclosure.

Frequently asked questions

Related reading

• Amazon AAA and ICDR arbitration for suspended sellers and withheld funds
• Amazon seller legal updates and marketplace policy analysis

Attorney Advertising. This article is for general informational purposes and does not constitute legal advice, and is current as of August 15, 2026. Amazon policies and state law change frequently. Past results do not guarantee future outcomes. No attorney-client relationship is formed until a written engagement is signed.