Labelling AI-written articles and synthetic media under Article 50

The European Union's AI Act introduces two key requirements for generative AI tools. Firstly, any AI-written text published on matters of public interest must be disclosed as AI-generated, unless a human editor has reviewed and approved the content, in which case the editor's name and the date of review must be displayed. Secondly, any realistic synthetic images, audio, or video that closely resemble real people, places, or events must also be labeled as artificially generated or manipulated.

These obligations are distinct duties on the deployer of the generative AI, with the labeling duty falling on the provider of the generation tool, and the editorial review exception applying to the text duty. To comply, agencies should maintain detailed records of each AI-generated article or media piece, including the reviewer's name, date, and the nature of the review. It is crucial to note that a superficial examination does not suffice - a substantive human review is required to satisfy Article 50(4).

For content that could be considered either public interest or not, such as marketing materials that blur the line between informative and commercial, the text should be treated as requiring disclosure. While artistic or satirical works may be exempt from the disclosure requirement, any use of synthetic media that depicts real individuals without their consent remains a legal risk beyond the scope of Article 50. The Commission's Code of Practice provides recommended icons for implementing these labeling requirements.