Conclusion of the Artificial Intelligence Board on the assessment of the Code of Practice on Transparency of AI-generated Content (Article 50(7) AI Act)

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”)

The exponential growth in AI systems capable of generating synthetic audio, images, video and text at scale presents a fundamental challenge to democratic societies and to the integrity of the information ecosystem on which they depend. As the AI Act explicitly recognises, the outputs of such systems are increasingly indistinguishable from authentic human-generated content, giving rise to serious risks of disinformation and manipulation at scale, fraud, impersonation and consumer deception. These risks affect how citizens form their opinions, exercise their fundamental rights and participate in democratic life. In the context of ever-increasing exposure to AI-generated and manipulated content, preserving those abilities of individuals and our collective societies is a responsibility that the Union has chosen to address through binding law. The transparency obligations introduced by Article 50 AI Act constitute a foundational pillar of that response, establishing for the first time a horizontal, enforceable duty to mark and enable the detection of AI-generated and manipulated content and clear labelling for deep fakes and certain published text across the Union.

The Code of Practice on Transparency of AI-generated content (the “Code”) is a voluntary instrument drawn up pursuant to Article 50(7) of Regulation (EU) 2024/1689 (the “AI Act”). Its purpose is to serve as a guiding document for demonstrating compliance with the obligations regarding the marking, detection and labelling of artificially generated or manipulated content provided for in Articles 50(2), 50(4) and 50(5) AI Act, while recognising that adherence to the Code does not constitute conclusive evidence of compliance with these obligations.

The Code was prepared through an iterative, multi-stakeholder drafting process that began on 5 November 2025 and involved more than 180 participants: providers of generative AI systems, developers of marking and detection techniques, deployers, civil society organisations, academic experts, standardisation bodies and other specialized organisations, and very large online platforms. Member States were closely involved throughout as observers, and via the AI Board and its dedicated subgroup on transparency. The AI Office also invited public bodies and agencies from across the world to participate as observers.

The Code establishes eight commitments divided across two sections. Section 1 operationalises the obligations in Article 50(2) and (5) AI Act with four Commitments applicable to providers of AI systems generating or manipulating synthetic content. Section 2 operationalises the obligations in Article 50(4) and (5) AI Act with four Commitments applicable to deployers of AI systems able to generate or manipulate deepfakes or text published in the public interest without human review or editorial control. Subject to assessment of the Code as adequate, providers and deployers who adhere to the Code and comply with its commitments may rely on it to demonstrate compliance with

their corresponding obligations under the AI Act. This applies in the entire EU market without affecting or altering the powers of market surveillance authorities to investigate and determine whether the provider or deployer has actually implemented the commitments under the Code.

The Code further includes a set of optional measures which, while not legally required, are conducive to the objectives of Article 50(2), (4) AI Act and encouraged by the Code. These measures address, among other things, the provision of richer provenance metadata, the development of forensic detection capabilities, enabling functionalities for perceptible labelling for downstream deployers, and investment in advancing the state of the art of marking and detection techniques and use of an EU-wide icon for labelling. Beyond their individual contribution to transparency, they foster harmonisation of implementation approaches across Signatories and across the Union, reducing fragmentation and supporting a coherent European transparency ecosystem.

THE BOARD’S ROLE IN ASSESSING THE CODE

Article 50(7) AI Act cross-references to Article 56(6) AI Act that provides for an adequacy assessment by both the Commission and the European Artificial Intelligence Board (the ‘Board’). Although the AI Omnibus has introduced amendments to the procedure for the assessment in Articles 50(7) and 56(6) AI Act, those amendments have not yet become effective in a timely manner before the application of Article 50 AI Act on 2 August 2026. The adequacy assessment of the Code will therefore be conducted by both the Commission and the Board under the existing rules, while it will also remain valid for the future as long as the conclusions are aligned. This Opinion sets out the assessment by the Board. The Commission’s assessment is published separately.

This Opinion does not prejudge any future opinion that the Board may take in view of the enforcement of the AI Act, nor does it pre-empt the Board from assessing the Code as inadequate in the future following its regular monitoring and evaluation of the Code’s contribution to the effective implementation of Article 50 AI Act.

The Board welcomes the Code and appreciates the time, effort, and dedication put into delivering it by the independent expert Chairs and Vice-Chairs and all stakeholders involved in this process. This Code serves as a significant benchmark for compliance with the transparency obligations in Article 50(2), (4) and (5) AI Act.

THE BOARD ASSESSMENT OF THE CODE

The Board’s assessment is based on the criterion of “effective implementation” of the obligations in Articles 50(2), (4) and (5) AI Act, read in light of the additional criteria set out in Recital 135. Recital 135 specifies that the Code should support practical arrangements for making detection mechanisms accessible, facilitate cooperation with other actors along the value chain, and enable the public to effectively distinguish AI- generated content.

The present Board adequacy assessment of the Code is limited only to the measures that are mandatory and does not affect the possibilities for Signatories to implement also other measures that are optional under the Code and can contribute to the effective implementation of the transparency obligations under Article 50 of the AI Act.

ASSESSMENT OF SECTION 1: MARKING AND DETECTION OBLIGATIONS FOR PROVIDERS

Section 1 of the Code addresses the four distinct but interlinked elements of the transparency obligation under Article 50(2) and (5) AI Act: the marking of outputs in a machine-readable format; the detectability of those outputs as artificially generated or manipulated; the technical requirements applicable to marking and detection solutions (effectiveness, reliability, robustness and interoperability); and the requirement that the result of detection be provided to natural persons in a clear and distinguishable manner.