European Commission advances AI transparency code under EU AI Act
Industry and civil society debated AI Act transparency obligations during European Commission workshops.
The European Commission’s AI Office has convened a new round of working group meetings and workshops on the forthcoming Code of Practice on Marking and Labelling of AI-Generated Content.
The discussions brought together providers of generative AI systems and models, technology companies, industry representatives, civil society organisations and academic experts. Feedback from the meetings will inform the third and final draft of the code, expected in early June.
The code is intended to support transparency obligations under the AI Act, including requirements linked to marking, labelling, disclosure and detectability of AI-generated content. It covers issues such as synthetic media, deepfakes and certain AI-generated text.
Working Group 1 focused on marking and detection obligations for providers, including a revised multi-layered approach, technical feasibility, benchmarking, compliance frameworks and possible third-party assessments. Industry participants raised concerns over compliance burdens, innovation and feasibility, while civil society and academic experts called for stronger safeguards in the public interest.
Working Group 2 examined disclosure obligations for deployers of generative AI systems, particularly deepfakes and certain AI-generated text. Discussions covered origin disclosure, user-facing labels, proportionality, governance measures, editorial control and the possible development of a uniform EU label.
Additional workshops explored how machine-readable marks, provenance data, visible labels, watermarking systems and an EU-wide icon could work together across the AI value chain. Participants also discussed coordination with other EU rules, including the Digital Services Act, while stressing the need to balance transparency, legal clarity, accessibility and innovation.
Why does it matter?
The code of practice will help determine how AI-generated content is marked, labelled and disclosed across the EU. Its development highlights the practical difficulty of turning transparency obligations into workable rules, particularly when regulators, companies and civil society disagree over technical feasibility, compliance costs, user experience and safeguards against deceptive synthetic media.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!