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)

FUTURE REVISION OF THE CODE

The AI Office and the AI Board will regularly monitor and evaluate the achievement of the Code’s objectives and its contribution to the effective implementation of Articles 50(2),

(4) and (5) AI Act, in particular in light of the rapidly evolving landscape of AI technologies and the risks of manipulation, disinformation, fraud and impersonation that the transparency obligations are designed to address.

The Board takes note of the commitments from the Signatories that aim to align their practices with the advancement of the state of the art and adoption of new standards as they emerge, and notes that this alignment could improve the interoperability of marking and detection solutions.

The Board considers that targeted revisions of the Code should be initiated without delay where more robust interoperability solutions become available that are not being widely adopted by Signatories. This absence of wide adoption could either be due to the flexibility afforded by the current measures or the open-ended nature of the criteria against which compliance is assessed, in particular for interoperability measures for the detection solutions, and solutions allowing detection mechanisms to be locally executable or made available through open-source implementations without reliance on remote provider- specific infrastructure. The Board notes that the determination of what constitutes a sufficiently robust solution should not be left solely to the appreciation of Signatories and highlights the crucial role of competent market surveillance authorities in gathering experience with enforcement. The staged implementation mechanism of Measure 3.4 and the short text exemption of Measure 1.1 represent two instances where revision should be considered where necessary as standards emerge. The Board should also consider a revision of the Code if the described technological solutions prove to be insufficient, impractical or disproportionately burdensome to implement.

In addition, the Board invites the AI Office to consider facilitating targeted updates to the Code at least every two years, to make sure that it is meeting its intended objectives. This would ensure that the Code keeps pace with emerging technologies and enforcement experiences, and that the marking and detection of AI-generated and manipulated content is carried out in line with the objectives of the transparency requirements.

The Board also notes that the Commission's guidelines on Article 50 AI Act were only available in their draft version at the time of this assessment.

The Board expects to be closely associated with the monitoring and evaluation of the Code's implementation and with any process leading to its revision, as a condition for the Code to continue meeting the objectives assigned to it.

CONCLUSION

In light of the foregoing, the Board considers that Sections 1 and 2 of the Code adequately cover the obligations provided for in Articles 50(2), (4) and (5) of the AI Act and facilitate their effective implementation.

Competent market surveillance authorities should therefore consider the Code as the only EU-wide practical tool, as of July 2026, assessed as adequate for the implementation of the AI Act’s transparency obligations for AI-generated content. This is without prejudice to the powers of the competent market surveillance authorities to request information and documentation and carry out investigations to determine whether the provider or deployer has actually implemented the commitments under the Code. Similarly, providers and deployers of AI systems falling within the scope of those obligations can rely on the Code to demonstrate compliance with their obligations under Article 50(2), (4) and (5) AI Act, regardless of their place of establishment, operation and competent market surveillance authority.

The Board will monitor the implementation of the Code to ensure it meets its objectives and aligns with the state of the art. The Board will pay close attention to monitor the progress in its implementation, in particular with respect to the interoperability of the marking and detection solutions. Regarding measure 3.4 in Section 1, it is noted that the chosen approach for interoperability in the medium term by 2 February 2027 does not predetermine the solutions that Signatories will ultimately adopt, which might affect the effectiveness of the Code.

Should implementation of the Code prove unsatisfactory, the Board will consider recommending to the Commission to consider a targeted revision of the Code, or the adoption of common rules as it may be appropriate.