Code of Practice on Transparency of AI-Generated Content (marcatura ed etichettatura dei contenuti generati o manipolati dall'IA)

Signatories will implement marking and detection solutions that operate seamlessly across multiple systems, actors, contexts and technical implementations to enable detection of AI- generated or manipulated content, regardless of the marking technique deployed in different AI systems, to the extent technically feasible. At the time of publication of this Code, relevant interoperability standards and/or best practices are yet to be developed, except for digitally signed metadata. Therefore, a staged implementation of interoperability requirements will be adopted as follows. In the initial stage as of the entry into application of Article 50(2) AI Act, Signatories will:

(a) adopt relevant established standards or best practices to implement metadata marking and detection, thereby enabling open detection of such markings;

(b) provide public information to the Commission and relevant stakeholders on how to integrate and access detection solutions for the other type(s) of marking technique(s) that they have implemented in their AI system. To facilitate dissemination of such information, Signatories are encouraged to provide relevant information on their websites and to populate publicly available registries that may be created to streamline integrations.

(c) work towards a minimum level of interoperability among detection solutions for watermarking that reduce burden and costs for the Signatories, while ensuring a more seamless access for end-users. More specifically, Signatories will implement an interoperability solution for their detection mechanisms by 2 February 2027 by implementing one or more of the following:

i) a publicly available interoperable industry standard access method that enables detection queries to be routed to relevant detection mechanisms, and detection results to be interpreted by shared or provider-agnostic detection solutions interacting with endpoints implementing such API, including detection solutions that support public signposts but do not necessarily require them. Signatories who implement this approach shall not create unnecessary traffic to other services where such traffic can be avoided by directly reading out a known interoperability solution (like a signpost);

ii) a publicly readable signpost or other interoperable mechanism in the AI-generated or manipulated content that will signal to the public which detection solution to use without having to run all the detection solutions of providers of AI systems. Signatories will take reasonable steps to ensure that the detection mechanism(s) associated to their signpost marking technique(s) or other interoperability solution are made available to the Commission and relevant stakeholders so that they can be integrated in their detection workflows;

iii) a shared provider detection solution by a consortium of Signatories that is accessible and inclusive for any other Signatory to join, including SMEs and SMCs, subject to appropriate safeguards for the protection of confidential technical information. Any such aggregated detection solution shall be agnostic to the Signatories participating in the consortium, capable of detecting outputs generated or manipulated by their generative AI systems, and able to work in an interoperable manner with other detection solutions;

iv) another interoperability solution that achieves interoperability between Signatories comparable to the solutions above. The interoperability solution will be implemented in a manner that is not detrimental to the robustness of other embedded watermarks and that includes safeguards against misuse risks creation.

(d) To advance towards full interoperability, contingent upon their resources and capacities, Signatories will cooperate in good faith with the Commission and relevant stakeholders towards the development of improved and state of the art interoperability solutions and interoperability standards. In particular, Signatories are encouraged to participate in the task force described in Measure 3.5. Signatories are also encouraged to join and/or support international and European standardisation bodies, as well as consortia initiatives focused on the development of open content marking and detection standards that help to operationalise the measures envisaged in this Section of the Code. In these subsequent stages, Signatories commit to implement those standards in a timely manner as they emerge, to the extent technically feasible and needed for the application use case of their AI systems, in order to continue to be compliant with the requirement for interoperability under Article 50(2) AI Act.