The sessions gathered providers of generative AI systems, developers of marking and detection tools, industry associations, civil society, international organisations, and academic and research institutions.
The first workshop, held on 17 November by the Working Group on Marking and Detection Techniques (WG1), focused on techniques that providers must implement under Article 50(2). Discussions examined feasibility, effectiveness, interoperability, standards, and governance considerations. Participants were invited to submit written input to support the first draft of the Code.
On 18 November, a joint workshop explored the interplay between Working Group 1 and Working Group 2 (WG2) on Disclosure of deep fakes and certain AI generated text, addressing how transparency obligations under Articles 50(2), 50(4) and 50(5) connect with each other and with other EU legislation. Stakeholders discussed cooperation between actors, the balance between standardisation and flexibility, and when labelling or notification is most useful for users.
The same day, WG2 held its first dedicated workshop on disclosure obligations for deep fakes and certain AI-generated text under Article 50(4). Participants debated risk-based approaches, exceptions for non-deceptive uses, various degrees of AI involvement, responsibilities across the value chain, and the potential of provenance and cryptographic verification. Concerns about information fatigue and the need for nuanced, context-appropriate labelling were also raised
Across all three meetings, the Chairs and Vice-Chairs presented preliminary reflections and invited further written submissions, which will inform the drafting of the first version of the Code of Practice.
Below you can download the minutes of the working groups meetings, together with the participant list (in PDF).