Guide
EU AI Act Article 50: what advertisers must label, and when
By the LikePact team · Updated
From 2 August 2026, Article 50 of the EU AI Act requires AI-generated content to be identifiable. Providers of generative AI tools must mark their outputs in a machine-readable way. Brands that publish AI-generated images or video of real-looking people, places or events, called “deep fakes” in the Act, must clearly disclose that the content is AI-generated.
This guide explains what that means for advertising. It is general information, not legal advice. For a specific campaign, check with your own counsel.
Key dates
| Date | What applies |
|---|---|
| 2 August 2026 | Article 50 transparency obligations apply across the EU, including disclosure of deep fakes. |
| 2 December 2026 | End of the transition period for machine-readable marking (Article 50(2)), for generative AI systems already on the market before 2 August 2026. The Digital Omnibus agreed this transition. Systems placed on the market after 2 August get no transition period. |
Breaches of the Article 50 obligations can be fined up to €15 million or 3% of worldwide annual turnover, whichever is higher.
Who must do what
Article 50 splits duties between providers, who build or supply the AI system, and deployers, who use it, such as a brand or agency publishing an ad.
| Article | Who | Obligation |
|---|---|---|
| 50(1) | Providers | Tell people when they are interacting with an AI system, such as a chatbot |
| 50(2) | Providers of generative AI | Mark generated audio, image, video and text in a machine-readable, detectable format |
| 50(3) | Deployers | Inform people exposed to emotion recognition or biometric categorisation |
| 50(4) | Deployers | Disclose that deep fake image, audio or video content has been AI-generated or manipulated |
For advertisers, the key duty is Article 50(4). Your image or video tool may already mark its output under 50(2), but that doesn’t disclose anything to your audience. The visible disclosure is yours.
Is my ad a deep fake?
The Act defines a deep fake as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful (Article 3(60)).
Some practical consequences:
- An AI-generated ad showing a real, recognisable person is a deep fake, even when that person has licensed their likeness and approved the ad. Consent makes the content lawful to make. It doesn’t remove the duty to disclose it.
- Photorealistic AI scenes of real places or products can also qualify.
- Clearly stylised illustration that nobody would take as real is less likely to qualify.
For content that is part of an evidently artistic, creative, satirical or fictional work, the duty is lighter: you disclose in a way that doesn’t spoil the work. Most advertising is not “evidently” artistic, so don’t rely on this exception without advice.
How to disclose
The Act requires the information to be clear and distinguishable, given at the latest at the first exposure to the content, and accessible (Article 50(5)). In practice:
- Put a visible label on the asset itself. Metadata is often stripped when files are uploaded to social platforms, so it cannot be your only disclosure.
- Keep the label legible at the size the asset is actually seen, including mobile feeds and stories.
- Label video from the start, not only in the end card.
- Use the platform’s own AI disclosure as well. YouTube, TikTok and Meta all have one, and their policies apply on top of the law.
The European Commission’s code of practice on transparency of AI-generated content proposes a common “AI” icon. Following it makes your label recognisable across brands.
Machine-readable marking and C2PA
Article 50(2) machine-readable marking is mostly a duty for the tool provider. Brands still benefit from keeping it intact and adding their own provenance. The open standard for this is C2PA (Content Credentials). A signed manifest travels with the file and records that it is AI-generated, which tool produced it, and what changed.
Two good habits:
- Don’t strip metadata in your export and compression pipeline.
- Sign the final file with your own C2PA manifest, so the published version, not just the raw output, carries its provenance.
Checklist for brands and agencies
- List every campaign asset that uses AI-generated imagery, video or audio.
- Flag the ones that show real people, places or events realistically. These are your deep fakes.
- For each one showing a real person, keep written consent for that use: the licence, the scope and the approval of the final file.
- Add a visible AI label to the asset itself, legible at the size it is seen.
- Keep or add machine-readable marking (C2PA) to the final export.
- Use each platform’s AI disclosure setting when publishing.
- Keep an audit trail: which file ran, where, when, and on what consent.
Outside the EU
- United Kingdom. There is no equivalent of Article 50 yet. Advertising codes still prohibit misleading ads, and platform policies apply.
- New York. Since 9 June 2026, ads must conspicuously disclose “synthetic performers”. These are AI-generated humans who are not recognisable as a real person. Licensed real people are covered by New York’s separate right-of-publicity and digital-replica rules.
- Other US states. Right-of-publicity laws, such as Tennessee’s ELVIS Act, protect a real person’s likeness. Using it without consent is the risk, whether or not the ad is labelled.
How LikePact handles Article 50
LikePact treats every asset made with a creator’s likeness as a deep fake under the Act. On approval:
- Visible AI label. LikePact burns a label into the image, or the brand confirms its own label is already in the file. That is the only option for video.
- C2PA signature. The final file is signed with a manifest that records it as AI-generated, with the original as its parent and a reference to the consent record.
- Consent record. The approval is written to an append-only, hash-chained ledger: who approved which file, when, and under which licence.
That gives a brand the label, the manifest and the evidence. Compliance stays with the brand that publishes the ad, but nothing is missing when someone asks for proof.
Frequently asked questions
Does Article 50 apply to brands outside the EU?
If the content is shown to people in the EU, assume it applies. The AI Act covers deployers outside the EU when the output is used in the EU.
Is a C2PA manifest enough to comply?
No. C2PA is machine-readable marking. Article 50(4) also requires a disclosure your audience can see.
Do we need to label AI-retouched photos?
Light editing that doesn’t change what the image shows usually isn’t a deep fake. Content that would make people believe something happened that didn’t, such as a person saying or doing something they never did, needs a label.
Does the creator’s consent remove the labelling duty?
No. Consent and disclosure are separate duties. You need the creator’s consent to use their likeness, and you need to disclose that the content is AI-generated.