Creative Strategy / AI and Advertising
AI Disclosure Rules for Brands in 2026: New York, Amazon and the EU AI Act
Three separate AI disclosure obligations landed inside nine weeks in 2026. New York's synthetic performer law took effect June 9. EU AI Act Article 50 transparency obligations began applying August 2. Amazon's metadata requirement followed on July 27. Every article I can find explaining them was written by a law firm, for lawyers.
This is the production-side version. I make commercial content for brands, so my interest is narrow and practical: what does a marketing team have to track before the next campaign ships. Nothing here is legal advice, and the sections below name the statutes so you can hand them to counsel.
If you are generating a photorealistic person who does not exist and using them to sell something, that is now a tracked production decision in at least three places.
If the people and products in your content are real, none of these particular obligations attach, and your provenance chain stays simple.
For brands that want a clean source-content workflow, I shoot commercial product photography and UGC-style brand content using real products and real people, which keeps the paperwork side uncomplicated.
New York: synthetic performers in advertisements
New York General Business Law section 396-b was amended by S.8420-A and A.8887-B, signed December 11, 2025 and effective June 9, 2026. It requires a business that knowingly uses a synthetic performer in a commercial advertisement to conspicuously disclose it, subject to the statute's definitions and exceptions.
Three details matter to a production team.
- Reach, not residence. The obligation follows the audience. An advertiser based anywhere can be in scope if the ad could reach New Yorkers.
- Background figures count. The law is not limited to a synthetic spokesperson. A generated person in the background of a scene is still a synthetic performer.
- Penalties are per violation. $1,000 for a first violation, $5,000 for each subsequent one.
There are carve-outs, including certain advertisements for expressive works and audio advertisements, and the statute does not define conspicuous. That ambiguity is exactly why this belongs in front of counsel rather than in a creative brief.
Amazon: metadata for photorealistic AI-generated people
Amazon's requirement is narrower and purely mechanical. Buyer-facing media containing a photorealistic person generated entirely by AI has to carry the keyword contains-synthetic-performer in the XMP dc:subject field. Media with only real people, including real people edited with AI tools, and media with no people at all are exempt.
I cover the implementation, the exemptions and the ordinary image specification that is far more likely to suppress a listing in a separate guide.
My existing Amazon product image requirements guide covers the specification side.
EU AI Act Article 50
Article 50 transparency obligations began applying on August 2, 2026. The European Commission published guidelines on those obligations in July 2026. Broadly, providers carry machine-readable marking duties for certain AI-generated content and deployers carry disclosure duties in defined circumstances, including deepfakes.
The practical read is that this is more nuanced than a universal AI label, and that a brand selling into multiple markets needs someone to determine which role it occupies for which asset. That determination is not a marketing decision.
California: SB 942 and AB 853
California's AI Transparency Act, SB 942, focuses on transparency tooling and disclosure for AI-generated content, and AB 853 amended it, including changes to scope and timing. The direction of travel is what matters here. New York was first, it will not be last, and a campaign built to run nationally is going to encounter more than one standard.
This is really a content provenance problem
For most of my career the chain behind a commercial image was self-documenting. A brand knew who shot it, who appeared in it, what product was in front of the lens, what retouching happened and where the file came from.
Generative tools break that chain fast. A marketing team can receive a finished asset and be unable to answer basic questions about it: which model produced it, if the person is real, if a real person was the basis, if required metadata is present, if that metadata survived the last export, and if the same file is compliant everywhere it is scheduled to run.
That is not an argument against AI. It is an argument for a record.
The AI asset log a small brand can actually maintain
This does not need to be a policy document. It needs to be a repeatable line in your asset workflow. For every final asset, record:
- Was AI used, and at what stage
- Which tool
- Was a person generated
- Was a real person altered
- Was the product photographed or generated
- Where the asset will run, by platform and by market
- What metadata or disclosure that destination requires
- Who approved it
Nine fields. Do it at approval, not at audit.
Real photography has a quiet new advantage
As synthetic imagery gets cheaper, verifiable imagery gets more useful in specific contexts.
When I photograph a physical product in the studio, the brand knows exactly what the source is. When I photograph a real person, the brand knows who they are and manages releases and usage through a production workflow that has existed for decades. AI-assisted tools can still be used in post. The provenance underneath stays explainable.
That matters most in ecommerce, medical and aesthetic marketing, consumer products, food, automotive parts and any regulated category where product accuracy is not decorative.
I work this way across product photography, commercial video production, UGC-style brand content and creative strategy for brands that need the content and the plan behind it.
One tool worth having in a provenance workflow
If accuracy is the standard you are being held to, a colour reference in the first frame of every set is the cheapest documentation you can produce. It gives you a defensible starting point for what the product actually looked like, which is a useful thing to have when the question is if the image represents the item.
What to do this quarter
- Audit current advertising for synthetic people, including background figures.
- Flag which assets run on Amazon, in New York and in the EU.
- Preserve source files and metadata rather than only finished exports.
- Put the nine-field log into the approval step.
- Send the genuinely unclear cases to counsel before the campaign ships, not after.
Then build the real source library. Once you have accurate photography of the actual product and real people you have permission to use, you know precisely where the synthetic layer starts. In 2026 that line is worth being able to point to.
Commercial Content, Pittsburgh and Nationwide
Build the real library first
Product photography, UGC and commercial video made from actual products and actual people. Clean provenance, clear usage terms, and content you can run anywhere.
Start a ProjectFAQ
Do I have to disclose AI in my ads?
It depends on the jurisdiction and on what the AI produced. New York requires a conspicuous disclosure when a business knowingly uses a synthetic performer in a commercial advertisement, subject to the statute's definitions and exceptions. Other obligations apply in the EU. This is a legal question, so confirm your specific exposure with counsel.
What counts as a synthetic performer?
Under New York General Business Law section 396-b, 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 a human performance by someone who is not a recognisable real person. Background figures are covered too.
What is the penalty for not disclosing a synthetic performer in New York?
The law provides a civil penalty of $1,000 for a first violation and $5,000 for each subsequent violation. It reaches advertisers whose content could reach a New York audience regardless of where the advertiser is based.
Does Amazon require AI disclosure on product images?
Amazon requires specific metadata on buyer-facing media containing a photorealistic person generated entirely by AI. Media with only real people, including real people edited with AI tools, and media with no people at all are exempt.
What does EU AI Act Article 50 require?
Article 50 transparency obligations began applying on August 2, 2026. They cover marking and disclosure duties for certain AI systems and AI-generated or manipulated content, with different responsibilities for providers and deployers. The detail is more nuanced than a blanket AI label.
Is real photography exempt from AI disclosure rules?
Photography of a real product or a real person does not create a synthetic performer, so these particular disclosure obligations do not attach to it. Provenance is also simply easier to document, which is one reason brands in regulated categories keep real source content at the base of their library.
Related reading: what UGC content is and why brands are paying creators to make it, why UGC and product photography work together, and my Amazon product image requirements guide.
I am Ian Jones, a commercial photographer and videographer serving Pittsburgh and Western Pennsylvania with shipped product work nationwide. See what I do or start a project.
Nothing in this article is legal advice. Sources: New York General Business Law section 396-b as amended by S.8420-A and A.8887-B, effective June 9, 2026. Amazon Seller Central guidance on photorealistic AI-generated people. European Commission guidelines on Article 50 transparency obligations, July 2026, applying from August 2, 2026. California SB 942 as amended by AB 853.