That AI model in your product photo? You probably have to label it now

Aug 3

2 min read

As of August 2, 2026, the EU AI Act's Article 50 transparency obligations have taken legal effect. If your work reaches anyone in the EU, these rules can reach you, even if you are based somewhere else.

In short, if your work is seen in the EU and AI had a "hand" in it, you need to label it so.

Here is what has to be disclosed:

Chatbots must tell people they are talking to a machine.
AI-generated images, audio, and video must be marked as AI.
Deepfakes must be labeled as deepfakes. By the way, any realistic AI model qualifies as deepfake. It does not need to be modeled after a real person. If it looks believable that it could be a real person, it qualifies as “deepfake” in this context.
AI-written text presented as news or public-interest information must be disclosed, unless a human editor reviewed it and stands behind it.

What it means to you:

Ecommerce brands and marketers: If you use AI-generated models, sets, or props in imagery shown to EU shoppers, plan to label them. You are the "deployer," so the visible label is your job, not just the tool maker's.
Photographers and retouchers: Fully synthetic shots and heavily AI-altered images fall in scope. Composite work that would read as a real photo needs a label.
Illustrators and designers: Realistic AI-generated visuals delivered for EU-facing use should be flagged as AI.
Filmmakers and video editors: Synthetic faces or voices need a deepfake label. Obvious creative or fictional work gets a lighter touch, but the label still has to be there.
Musicians and producers: AI-generated audio counts as synthetic content and should be marked.
Writers, journalists, publishers: AI-written public-interest text needs disclosure, unless a real editor reviewed it and takes responsibility. Human editing is your exemption, so don’t replace your editor with a synthetic one.

The exceptions:

There is a carve-out for content that is clearly artistic, creative, satirical, or fictional, and the label is meant to be applied in a way that does not ruin the work. However, note that commercial advertising, like a product shot, does not qualify as artistic in the first place. So the AI-model product photo gets no relief here.

One more thing:

The AI tool makers will have the obligation to embed a hidden, machine-readable mark in the AI output by December 2, 2026. It will NOT replace the requirement to visually label the work as AI.

Also, there is no need to retroactively label the work published prior to Aug 2.

What happens if you do not comply?

Regulators can fine violations up to 15 million euros, or 3% of your worldwide annual revenue, whichever is higher. For small companies and startups, they use whichever number is lower, so the penalty scales to your size rather than hitting a flat figure.

How and how well will this rule be enforced? It is too early to tell. Certainly, no one wants to be the first one to find out. So remember to label AI-generated and deepfake content and/or seek legal advice when in doubt.