Joint letter to the European Commission on the AI Act draft Guidelines

Brussels, 17 June 2026

egta, together with European partners representing the audio and audiovisual media community in Europe, has signed a joint letter to the European Commission on the draft Guidelines under the AI Act. Our members produce and distribute trusted media content across the EU, contributing significantly to media pluralism and serving as a credible counterweight to disinformation.

We welcome the Commission’s efforts to provide clarity on the transparency obligations under Article 50 of the AI Act. However, we have serious concerns about several aspects of the draft Guidelines that risk producing outcomes that are disproportionate, technically unworkable, and ultimately counterproductive.

Our overarching concern is that the draft Guidelines adopt an overly broad interpretation of the provisions in Article 50(4), (5) and Article 3 No. 60. We therefore urge the following amendments:

• Narrow and clarify the deepfake definition: Limit it to content presenting an actual risk of deception, rather than extending to merely “realistic” content. Fictitious, non-deceptive content as well as minor editorial or technical adjustments should be explicitly excluded.
• Clarify the timing of disclosure to avoid persistent labelling: Continuous or persistent labelling obligations across content would compromise audience engagement and must be avoided.
• Ensure proportionate and technically workable labelling standards that account for human perceptual limits and avoid a “banner blindness” effect.

As currently drafted, the Guidelines would generate excessive over-labelling across a broad range of professional media content, creating substantial technical and operational burdens. More importantly, they risk undermining audience trust in media by creating a perception of unreliability through visual clutter, rather than addressing genuine concern. This is particularly significant given that media service providers already operate under strict standards, ensuring regulatory oversight and human editorial responsibility.

First, the deepfake definition in the draft guidelines far exceeds the scope foreseen in the AI Act. By equating the criterion of “existing” with a broader notion of “realistic” content, the guidelines depart significantly from the text and intent of Article 3(60). Similarly, by assessing audience impact against the most easily misled viewer rather than an average recipient, the draft lowers the standard far below that applied in Article 50(1) and the Commission’s established practice.

Taken together, these choices would extend the labelling obligation to content that is clearly fictitious or non-deceptive, generating legal uncertainty, undermining the risk-based approach of the AI Act and producing systemic over-labelling. The definition and related obligations should not extend to content that is clearly fictitious or unrealistic and does not create genuine confusion about authenticity. Furthermore, not only minor tech-nical but also minor content adjustments should be excluded from the notion of deep-fake.

Secondly, the Guidelines misread the timing rule foreseen in the AI Act. Article 50(5) fixes the latest moment for disclosure. It does not create a continuous obligation. The reading in paragraph 132 that deepfakes must be labelled persistently or at later stages of a live broadcast is therefore disproportionate and has no basis in the law. This affects audiovis-ual as well as audio content, where persistent or repeated disclosure is particularly im-practical and risks disrupting the listening experience. The guidelines themselves acknowledge the risk of “banner blindness”: labelling large volumes of non-deceptive content will desensitise audiences, undermining the transparency objective in precisely the cases where it matters most.

The Guidelines, as drafted, risk achieving the opposite of their stated objective: rather than building audience trust in AI transparency, they would flood viewers with labels on non-deceptive content, eroding the very signal they are meant to convey. Targeted adjust-ments to the deepfake definition and the implementation/labelling requirements and the allocation of responsibility are essential to ensure the Guidelines are both legally sound and workable in practice.

We would welcome the opportunity to jointly discuss and further elaborate on this matter with you and your team at your convenience.

Signatories (in alphabetical order):

1. Arbeitsgemeinschaft Privater Rundfunk (APR)
2. Asociace komerčních televizí (AKTV)
3. Association of Commercial Television and Video on Demand Services in Europe (ACT)
4. Association of European Radios (AER)
5. European Association of Television and Radio Sales Houses (egta)
6. European Broadcasting Union (EBU)
7. Verband Österreichischer Privatsender (VÖP)
8. Verband Privater Medien (VAUNET)