Applicable to: AI providers.
Date of application: 2 august 2026.
Objective: to ensure that synthetic audio, image, video or text content, generated or manipulated by AI systems (including General Purpose Artificial Intelligence, GPAI) is machine-readable and detectable.
Regulation (EU) No 1689/2024 on Artificial Intelligence (AI Act) of 13 June 2024 introduced a series of transparency obligations within the European Union for suppliers and deployers of AI systems, even where such systems pose a limited risk (the lowest risk level assigned to certain AI systems under the risk-based approach adopted by the AI Act).
In this article, we will analyse the obligation covered by Article 50.2 of the AI Act, under which it is provided for:
Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated. Providers shall ensure their technical solutions are effective, interoperable, robust and reliable as far as this is technically feasible, taking into account the specificities and limitations of various types of content, the costs of implementation and the generally acknowledged state of the art, as may be reflected in relevant technical standards. This obligation shall not apply to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof, or where authorised by law to detect, prevent, investigate or prosecute criminal offences.
The elements that characterise this obligation are:
1) the circumstance that the AI system is intended to generate synthetic audio, image, video or text content audio, immagine, video o testuali sintetici
2) that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated
3) that the technical solutions adopted by the providers are effective, interoperable, robust and reliable.
The information under this obligation shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure.
Regarding the technical solutions to be adopted, the AI Act has clarified that the effectiveness and reliability of marking methods must be evaluated taking into account available techniques or a combination of such techniques (for example: watermarks, metadata identifications, cryptographic methods or other techniques, as may be appropriate; recital 133).
In this regard, the European Commission is drafting a Code of practice on transparency of AI generated contents (the final version of which should be approved within the month of June 2026).
The draft Code published in march 2026 states that no single marking approach is sufficient on its own to guarantee effectiveness and reliability and therefore proposes a multi-layered marking approach, based on a set of sub-measures.
The first sub-measure, applicable where contents are generated or exported in a data format that supports adding information as a part of the metadata, involves recording and embedding through the metadata an indication of whether the content is AI-generated or AI-manipulated.
If, however, the output is generated in a format that does not host metadata, an option that allows the download of a digitally signed manifest containing a certified version of the output generated or manipulated by their AI system must be implemented to certify the artificially generated or manipulated origin of the text content.
The second sub-measure involves robust and interwoven within the content marking techniques (such as an imperceptible watermarking directly intervowen within the content in a manner that is difficult for it to be separated from the content itself).
Where appropriate, to address deficiences in the two previous sub-measures, a third optional supplementary sub-measure may be adopted, consisting in fingerprinting or logging mechanisms of the AI-generated or AI-manipulated contents (for example, direct logging might be appropriate for text outputs, whilst perceptual hashing or other fingerprinting may be preferable for visual outputs).
The AI Act has, finally, clarified that this obligation does not apply to AI system:
- that perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof;
- authorised by law to detect, prevent, investigate or prosecute criminal offences.
Articoli Correlati