New Guidelines from the European Commission clarify transparency obligations under Article 50 of the AI Act

IRIS 2026-8:1/9

Justine Radel-Cormann

European Audiovisual Observatory

The Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the Artificial Intelligence Act (AI Act), available since 20 July 2026, provide further detail of each paragraph of Article 50. The Guidelines are non-binding and serve as practical guidance to assist interested parties in ensuring compliance with the transparency obligations under Article 50. Overall, the Guidelines explain the types of content covered, the actors concerned and the technical characteristics required for markings and detection mechanisms to support transparency. 

Article 50(2) applies to providers of AI systems, including providers of general-purpose AI systems. Providers must ensure that output generated or manipulated by their AI systems is detectable as artificially generated or manipulated. The information must be provided to natural persons likely to be exposed to the content in a clear and distinguishable manner, no later than at the time of their first interaction with or exposure to it.

The Guidelines also identify actors for whom this information is important, including competent authorities, independent researchers, civil society organisations, media, trusted flaggers and fundamental rights defenders. The Guidelines detail the content falling within this obligation, for example, 3D images, 3D videos or digital twins providing a virtual replica of a person, physical object or system, unless they fall within the applicable exceptions, including where they qualify as industrial or business-to-business applications.

The required marking must be provided in a machine-readable format. This means that it must be structured so that software applications can identify, recognise and extract it without human intervention. The Guidelines also explain that detection mechanisms should identify whether content has been generated or manipulated by AI, typically through technical markers or signatures verifying its origin.

Finally, providers should promote interoperability and rely on publicly available industry-standard detection solutions. The Guidelines further address the effectiveness, interoperability, robustness and reliability of these technical solutions.

Article 50(4) concerns deployers. Where deployers generate or manipulate content and the result constitutes a deepfake, they must disclose that the content has been artificially generated or manipulated. The Guidelines detail the four cumulative criteria necessary for a deep fake: the (i) resemblance of (ii) existing (iii) persons, objects, places, entities or events that (iv) would falsely appear to a person to be authentic or truthful.