The transparency obligations in Article 50 of the EU AI Act apply from August 2, 2026. Two of them reach directly into the digital-human industry. First, an AI system that interacts with a person must make that clear at the point of contact — disclosure buried in terms of service does not satisfy it. Second, image, audio or video content that is AI-generated or manipulated must be marked in a machine-readable form, and deepfake content must be labelled as artificially generated.

The definition is what makes this expansive. Article 3(60) defines a deepfake as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear authentic. The European Commission's guidance is explicit that content resembling a real person falls in scope even where there was no intent to deceive. A licensed celebrity avatar, built with the artist's consent and marketed as an avatar, still carries a marking and labelling duty when it is put on the EU market.

What compliance actually looks like

The Commission published guidelines on transparency obligations alongside a Code of Practice on Transparency of AI-generated Content, and released a set of icons deployers may use to label AI-generated output. Adherence to a Code of Practice deemed adequate by the AI Office is a route to demonstrating compliance with Articles 50(2), (4) and (5) — not the only route, but the cheapest one to document. Providers carry the machine-readable marking duty at the model layer; deployers carry the visible labelling duty at the point of publication. Most avatar businesses are both.

The rule does not ask whether the avatar was authorised. It asks whether the output resembles a real person and whether the viewer was told. — EW analysis

Key Facts

  • Article 50 transparency obligations apply from August 2, 2026
  • AI systems interacting with people must disclose that at the point of contact
  • AI-generated or manipulated media must be marked machine-readably; deepfakes must be labelled, with no intent to deceive required
  • Deepfake is defined in Article 3(60) as content resembling existing persons, objects, places, entities or events that would falsely appear authentic
  • Commission guidelines, a Code of Practice on Transparency of AI-generated Content and an official icon set support compliance
  • Penalties for infringement reach EUR 15 million or 3% of total worldwide annual turnover, whichever is higher

Why it matters

The avatar and digital-human market has been sold on realism, and realism is now the trigger for a compliance obligation. That changes the product spec: provenance marking, an audit trail of consent, and a visible label become build requirements rather than legal afterthoughts, for anyone touching EU users. It also changes the competitive picture. Vendors that already ship content credentials and disclosure UI carry a low marginal cost; those that treated indistinguishability as the whole value proposition now have to add the one thing they were engineering away. Enforcement will take time to bite, and the practical test will be how the AI Office treats good-faith adherence to the Code of Practice. But the direction is set: in Europe, a synthetic person has to say so.

Frequently Asked

What changes on August 2, 2026?

Article 50 applies. AI systems must disclose themselves at the point of contact, AI-generated media must be machine-readably marked, and deepfake content must be labelled.

Does consent from the person depicted exempt an avatar?

No. The definition turns on resemblance to a real person, and the Commission's guidance covers content produced without any intent to deceive.

What does it cost to get this wrong?

Up to EUR 15 million or 3% of total worldwide annual turnover, whichever is higher.

Sources & Further Reading