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)

In light of the above, the Board considers that Section 1 of the Code, in its current form, adequately facilitates the effective implementation of the transparency obligations in Articles 50(2) and (5) AI Act and providers can rely on the Code to demonstrate compliance with their obligations.

The Board will monitor the implementation of the Code and in particular its Measure 3.4 on the interoperability solutions for the watermarking detection solutions by 2 February 2027. In case of lack of progress on the implementation of this interoperability measure, the Board reserve its right to recommend to the Commission to initiate a review of the Code on this specific aspect or consider adopting common specifying rules.

ASSESSMENT OF SECTION 2: LABELLING OBLIGATIONS FOR DEPLOYERS

Section 2 of the Code addresses the transparency obligations applicable to deployers of AI systems under Article 50(4) and (5) AI Act. These obligations cover two distinct categories: the disclosure of deepfakes, defined as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful; and the disclosure of AI- generated or manipulated text published with the purpose of informing the public on matters of public interest without human review or editorial control. Article 50(5) further requires that disclosure be provided in a clear and distinguishable manner at the latest at the time of the first interaction or exposure.

Commitment 1 – Disclosure of Deep Fakes and Published Text

Commitment 1 addresses the consistent and effective disclosure of the artificial origin of deepfakes and published text. It does so through common EU icons provided in Annex 1 of Section 2, together with design specifications for Signatories that opt not to use the common icon, and placement specifications ensuring that disclosure reaches natural persons no later than the time of their first exposure. The combination of design and placement requirements ensures that Signatories can implement the obligation across a range of use cases, sectors and content modalities while maintaining the clarity and distinguishability required by Article 50(5) AI Act.

Commitment 2 – Internal Processes

Commitment 2 ensures that Signatories put in place or maintain the internal processes, awareness measures and review mechanisms necessary for sustained compliance.

Measure 2.1 mandates compliance processes and documentation, including review mechanisms to remedy potentially mislabelled content for Signatories who regularly generate deepfakes or published text. Measures 2.2 and 2.3 further address awareness and literacy among personnel and cooperation with competent authorities. These measures can rely on existing processes and take into account the size and resources of Signatories, ensuring proportionality.

Commitment 3 – Disclosure for Artistic, Creative and Similar Works

Commitment 3 enables deployers to discharge their obligations in respect of deepfakes that form part of evidently artistic, creative, satirical, fictional or analogous works or programmes in a manner that does not hamper the display or enjoyment of the work.

It does so by allowing contextual disclosure, while still requiring compliance with the design specifications of Measure 1.1 and the minimum requirements of Article 50(5) AI Act.

Commitment 4 – Human Review and Editorial Control for Published Text

Commitment 4 facilitates the effective implementation of the exception to the transparency obligation under Article 50(4) subparagraph 2 for AI-generated text that has undergone human review or editorial control. By distinguishing between media service providers already subject to editorial standards and regulatory frameworks and other actors, the Code establishes a pertinent differentiation that takes into account existing regulations, as long as they are consistent with the human review or editorial control requirements of Article

50(4). The other Signatories are required to establish a similar editorial policy with minimum elements specified in the Code that ensure its effectiveness.

Overall Assessment of Section 2

In light of the above, the Board considers that Section 2 of the Code adequately facilitates the effective implementation of the transparency obligations under Articles 50(4) and (5) AI Act, and that deployers of AI systems can rely on it to demonstrate compliance with those obligations.

The Board will monitor, where appropriate in cooperation with other EU bodies and competent regulators, to what extent, in practice, labels are actually retained during further processing, platform distribution, or third-party applications.