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

To complement the marking techniques specified in Commitment 1 and contingent upon their capacities, Signatories may include as part of their detection solution, as an additional optional measure, forensic detection mechanism to detect content generated or manipulated by their AI system or underlying model for which marking has been stripped. Where Signatories implement such a forensic detection mechanism, they will ensure that the privacy and security requirements as specified in Sub-measure 2.1.3 are satisfied and that the disclosure of the detection result is clear and accessible as specified in Measure 2.3, and make best efforts to ensure that their forensic detection mechanism aligns with the state of the art as specified in Commitment 3. At the time of publication of the Code, forensic detection mechanisms were not deemed mature enough to comply with the quality requirements set in Article 50(2) AI Act and in

Commitment 3. As long as no better forensic detection mechanism has emerged under the state of the art, Signatories who implement forensic detection mechanisms as an optional measure may restrict access to the detection results provided by such forensic detection mechanisms to verified expert users, in line with the exception for detection mechanisms for free-form text watermarking described in Sub-measure 2.1.2.