Independent experts appointed as chairs and vice-chairs are responsible to draft the code. The meetings brought together stakeholders across the AI value chain participating in the code of practice process. These include providers of generative AI systems, developers of marking and detection tools, industry representatives, civil society organisations and academic experts.
Second round of working groups meetings
Working Group 2
The meeting of the Working Group on Disclosure of Deepfakes and AI generated text (WG2), held on 12 January 2026, examined section 2 of the draft code. It was centered around the disclosure obligations of deepfakes and AI-generated text. In particular, the discussion focused on the distinction between deceptive and non-deceptive uses, responsibilities across the value chain – including providers, deployers and intermediaries – as well as possible exceptions, proportionality and risk considerations. Participants highlighted the need to avoid information fatigue for users, while emphasising clarity, usability and context-appropriate transparency.
Working Group 1
On 14 January 2026, the Working Group on Marking and Detection Techniques (WG1) examined section 1 of the code that aims to facilitate implementation of providers’ obligations. Discussions focused on state-of-the-art marking techniques, including watermarking and metadata-based solutions, detection capabilities and their technical limitations, interoperability and standardisation challenges, as well as robustness, resilience against removal and governance aspects. Stakeholders highlighted feasibility considerations and the importance of aligning technical solutions with evolving international standards.
Workshops
Three additional workshops were organised with the Code of Practice participants and observers of the drafting process of the code (PDF). During the workshops, chairs and vice chairs addressed questions asked in advance.
WG1 workshop
On 21 January 2026, a WG1 workshop explored implementation challenges across different model architectures, trade-offs between transparency, innovation and system performance and cooperation between providers and downstream actors. Participants stressed the need for practical and technologically neutral guidance capable of accommodating diversity in system design, while ensuring meaningful and effective transparency.
WG2 workshop
On 22 January 2026, a WG2 workshop continued discussions on labelling obligations for deepfakes and AI generated text publications. It examined the proposed taxonomy distinguishing “AI-generated” from “AI-assisted” content, as well as the proportionality in labelling requirements and the responsibilities of online platforms and other intermediaries. The alignment with other EU legislation and the importance of ensuring coherent messaging to users across regulatory frameworks was also discussed.
Joint WG1 and WG2 workshop
On the same day, a joint WG1 and WG2 workshop explored the interplay between marking and labelling obligations. Participants reflected on how technical marking solutions can support disclosure obligations, the complementary roles of watermarking, metadata, labelling and provenance systems, and the balance between standardisation and flexibility in implementation. The need for coordination across actors to ensure end-to-end transparency throughout the AI value chain was also considered. There was a call for coherence between technical and legal requirements and a balanced approach favouring user awareness without creating disproportionate compliance burdens.
Second draft of the code of practice
Written contributions from stakeholders, received until 23 January 2026, will inform the second draft of the code of practice that will be published in early March.
Download the minutes of the meetings and the participant lists below.