The EU's AI Labelling Rules Are Live. What Must Brands Label?

Since 2 August 2026, anyone using AI in the EU must clearly label deepfakes, and AI-written text on matters of public interest that no human has reviewed. For most brand teams, that means labelling synthetic people, voices and likenesses in campaigns. It usually does not mean labelling marketing copy that a person has reviewed and signed off. This is a practical reading of the European Commission’s guidance, not legal advice.

Who has obligations under Article 50?

The AI Act splits duties between two roles. Providers build AI systems: they must design chatbots so people know they are talking to an AI, and mark generated content in a machine-readable way, such as watermarks or metadata. Deployers use AI systems in their own work. A brand or agency generating campaign assets with an image or video model is a deployer.

That split matters. A brand cannot rely on the tool’s invisible watermark to meet its own duty: the Commission’s FAQ says deployers cannot use the provider’s machine-readable marking to satisfy their labelling obligation.

What must a brand label?

Deepfakes. The Act defines these as AI-generated or manipulated image, audio or video content that resembles existing persons. In marketing that covers, for example:

• a real ambassador’s face or voice generated or altered with AI;
• a synthetic spokesperson who looks like a real, identifiable person;
• a dubbed or lip-synced version of a real person’s video made with AI.

The label must be clear and visible (or audible) “upon first exposure at the latest”. For evidently artistic, creative, satirical or fictional work, the duty is lighter: disclose in a way that does not spoil the work.

Some text. Deployers must label AI-generated or manipulated text “published with the purpose of informing the public on matters of public interest”, unless it has undergone human review or editorial control. The Commission specifies that spell-checking alone does not count, and that editorial responsibility means a person holds ultimate legal responsibility for the publication.

What probably doesn’t need a label?

Based on the Commission’s wording, routine brand copy that people review before publishing (product descriptions, social posts, emails) falls outside the text duty, both because it is reviewed and because it is generally not “on matters of public interest”. Images of products, scenes or invented people who do not resemble anyone real are not deepfakes under the definition.

Two caveats. First, “public interest” is not limited to news: brand content about health, the environment or social issues may come closer to the line. Second, other laws still apply. Consumer protection rules against misleading advertising already require honesty about what an ad shows, AI or not.

When does it apply, and what about old content?

Article 50 has applied since 2 August 2026. Providers of generative systems already on the market before that date have until 2 December 2026 to meet the machine-readable marking duty. Content created before 2 August 2026 does not need to be labelled retroactively, although the Commission encourages it.

Enforcement sits mainly with national market surveillance authorities. The Commission’s FAQ puts maximum fines at €15 million or 3% of worldwide annual turnover, with proportionality for smaller companies.

What is the Code of Practice?

The Commission has published a voluntary Code of Practice on transparency of AI-generated content, which it and the AI Board have assessed as adequate. Signing it gives legal certainty. Non-signatories must show they comply by other, equally adequate means, and may face more requests for information.

What should a brand team do now?

1. Inventory your AI use. List which tools generate images, video, voice and text in your campaigns, and who uses them, including agencies and creators.
2. Flag likenesses. Any asset that shows or sounds like a real person, made or altered with AI, gets a visible label. Make this a checkbox in creative approval.
3. Keep a human sign-off on record. For text, a documented editorial review is what takes content out of the labelling duty. Make sure your workflow records who approved what.
4. Update contracts. Require agencies and creators to tell you when they use AI on likenesses or voices, and to apply labels in the EU.
5. Agree a label style. Check the Code of Practice for its recommended labelling approach, then use one consistent, visible mark across channels. That is simpler to enforce than case-by-case wording.

Transparency also has a brand upside. Audiences are increasingly wary of synthetic media; being plainly honest about what is real is a trust signal, not just a compliance cost.