⚡ Quick Summary
Published by the European Artificial Intelligence Board, this conclusion sets out the Board's adequacy assessment of the Code of Practice on Transparency of AI-generated content, a voluntary instrument drawn up pursuant to Article 50(7) of Regulation (EU) 2024/1689 (the AI Act). The assessment applies the criterion of “effective implementation” of the obligations in Articles 50(2), (4) and (5) AI Act, read in light of the additional criteria in Recital 135, which requires practical arrangements for accessible detection mechanisms, cooperation along the value chain and public ability to distinguish AI-generated content.
The document examines the Code's eight commitments in two sections. Section 1, addressed to providers, covers marking (Commitment 1), detection (Commitment 2), technical requirements of effectiveness, reliability, robustness and interoperability (Commitment 3), and testing, verification and compliance (Commitment 4). Section 2, addressed to deployers, covers disclosure of deepfakes and published text, internal processes, artistic and creative works, and human review and editorial control.
The Board concludes that both sections adequately facilitate effective implementation, that providers and deployers can rely on the Code to demonstrate compliance, and that market surveillance authorities should treat it as the only EU-wide practical tool assessed as adequate as of July 2026. It commits to monitoring, especially the interoperability measure due by 2 February 2027.
🧩 What’s Covered
The document follows the Code's structure, assessing each commitment in turn.
- Context and instrument: the Code is a voluntary instrument under Article 50(7) AI Act covering the marking, detection and labelling duties of Articles 50(2), (4) and (5); adherence is not conclusive evidence of compliance.
- Drafting process: an iterative multi-stakeholder process begun on 5 November 2025 with more than 180 participants, including providers of generative AI systems, marking and detection developers, deployers, civil society, academic experts, standardisation bodies and very large online platforms.
- Structure of the Code: eight commitments in two sections — four for providers under Articles 50(2) and (5), four for deployers under Articles 50(4) and (5) — plus optional measures on provenance metadata, forensic detection and an EU-wide icon.
- Board's role: Article 50(7) cross-references Article 56(6); as the AI Omnibus amendments were not effective before the application of Article 50 on 2 August 2026, the Commission and the Board assess under existing rules, the assessment covering only mandatory measures.
- Section 1 (providers): marking (Measures 1.1–1.2), detection (Measures 2.1 and 2.3), effective, reliable and robust solutions (Measures 3.1–3.3), staged interoperability by 2 February 2027 (Measure 3.4) and testing, verification and compliance (Commitment 4).
- Section 2 (deployers): disclosure of deepfakes and published text via the common EU icons in Annex 1 or specified alternatives, internal processes, contextual disclosure for artistic and satirical works, and human review and editorial control.
- Future revision and conclusion: the Board invites targeted updates at least every two years, flags revision where interoperability solutions are not widely adopted, and finds both sections adequate, making the Code the only EU-wide practical tool assessed as adequate as of July 2026.
💡 Why it matters?
For providers and deployers within scope, the opinion indicates how the Code can be used to demonstrate compliance with the Article 50 transparency duties, while stating that market surveillance authorities retain powers to investigate actual implementation. It highlights practical dependencies: accessible detection, preservation of marks along the value chain, and interoperability between watermarking detection mechanisms, which the Board will examine after 2 February 2027. Compliance, legal and engineering teams marking or labelling synthetic content can read the document as the Board's account of where the Code is considered sufficient and where it is not.
❓ What’s Missing
The document assesses the Code rather than reproducing it: the full text of the commitments, the icons in Annex 1 of Section 2 and the technical specifications are not included, and the Commission's separate adequacy assessment and its guidelines on Article 50 (available only in draft) are not reproduced. The assessment is limited to mandatory measures, so optional measures appear only in outline. The Board acknowledges open questions: possible circumvention of the single-layer marking exception for closed environments, the short text exemption under Measure 1.1, and whether sufficiently robust interoperability solutions will emerge. No publication date is printed.
👥 Best For
Compliance and legal teams at providers and deployers of generative AI systems preparing to rely on the Code; market surveillance and regulatory staff tracking enforcement expectations; product and engineering teams implementing watermarking, provenance metadata and labelling; and policy analysts following EU transparency requirements for AI-generated content.
📄 Source Details
Conclusion of the Artificial Intelligence Board on the Assessment of the Code of Practice on Transparency of AI-generated content pursuant to Article 50(7) of Regulation 2024/1689 (Artificial Intelligence Act or “AI Act”), European Artificial Intelligence Board, 7 pages, English. No publication date, version number, reference number or URL is printed. The input was the text extracted from all seven pages of the PDF, with page breaks marked in the extraction.
