From 2 August 2026, the EU AI Act's transparency obligations apply to synthetic media. If you generate ad creative with AI, and a meaningful number of agencies now do, this is in scope.
The obligation is not onerous. The risk is not noticing it applies.
What the rule is for
The concern is simple: a person should be able to tell whether what they are looking at depicts something that happened. The Act requires that content which has been artificially generated or manipulated, and which appreciably resembles real people, places or events, is disclosed as such.
The term "deepfake" makes people picture a fabricated video of a politician. The obligation is broader and quieter than that, and the quiet end is where advertising sits.
What this catches in advertising
The honest answer is: more than most teams assume.
AI-generated people. A synthetic spokesperson, a generated "customer", a face that does not belong to anyone. Appreciably resembles a real person, because it is designed to.
AI-generated places and scenes. A generated interior, storefront or location presented as though it is the real premises.
Manipulated footage of real people. Altering what someone appears to say or do. This is the central case and the least ambiguous.
AI-generated testimonial-style content. Where this overlaps with other law it gets serious quickly, because a fabricated endorsement is not only a labelling problem. In the US it also runs into the FTC's rule on fake reviews and testimonials.
What is generally outside the concern: AI used as a tool on something real. Upscaling, background removal, colour correction, a generated abstract pattern that does not depict anything.
The line to reason from is depiction. Does this content invite a viewer to believe they are seeing a real person, place or event that is not real?
What a compliant disclosure does
The label has to actually inform the person looking at the ad. That means:
- Visible to the viewer, not buried in metadata a human never sees.
- Present at the point of viewing, not on a page they would have to go and find.
- Clear in plain language. "AI-generated" communicates. A small icon with no explanation does not.
A disclosure that technically exists but that no viewer would notice is the failure mode worth designing against, because it is also the one that is easy to ship by accident.
Why agencies carry more of this risk
Volume and distance. An agency produces more creative than any one brand, often with AI in the pipeline precisely because it scales, and often several steps removed from the person who will answer for it.
There is also a records problem. In six months, can you say which assets in your library were AI-generated? If creative sits only inside client ad accounts, the context disappears with it. Knowing what you generated is a prerequisite for labelling it.
What to do now
Mark it at creation. The moment to record that an asset is synthetic is when it is made. Reconstructing it later is guesswork.
Decide the label once. Agree the wording and placement as a standard rather than per campaign, so it is consistent and nobody has to make a judgement call under deadline.
Audit what is already live. Creative generated before anyone was thinking about this is still running.
Watch the overlap. Synthetic testimonials and endorsements sit at the intersection of this rule, the FTC's rules on endorsements and fake reviews, and platform policy. That intersection is the expensive one.
The short version
From 2 August 2026, synthetic media depicting real-seeming people, places or events has to be labelled in the EU. Most AI tools in an ad pipeline are fine. Generated people and places presented as real are not, unless disclosed. The practical work is marking assets as synthetic at creation and agreeing one clear label, and the hardest part is knowing what you have already made.
