Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act — C(2026) 5054 final, Annex

(94) The marking and detection obligations under Article 50(2) AI Act focuses on how the content has been created and its artificial origin, not on who created the content. Any marking and detection solutions employed by providers of AI systems must therefore be compliant with applicable Union data protection law, including data protection principles and obligations, such as the obligation of data protection by design and by default, the data minimisation and the security and confidentiality principles. For example, for the purpose of marking and detection the information related to the creator of the content should not be processed and any other personal data processed should be deleted as soon as the purpose of detection is fulfilled, security and confidentiality must be ensured, etc.

(95) Recital 136 AI Act further highlights the particular relevance of the transparency obligations under Article 50(1), (2) and (4) AI Act to facilitate the effective implementation of the DSA. This applies in particular as regards the obligations of providers of very large online platforms (‘VLOPs’) or very large online search engines (‘VLOSEs’) to diligently identify, analyse, assess and mitigate systemic risks that may arise from the dissemination of content that has been artificially generated or manipulated, in particular the risk of the actual or foreseeable negative effects on

democratic processes, civic discourse and electoral processes, including through disinformation (Articles 34 and 35 DSA). Machine-readable marks may enable those providers to detect content that is AI-generated or manipulated and to provide appropriate labelling and disclosures, thus diminishing the risk of AI-generated disinformation across their services.

(96) The obligations to mark and label content generated or manipulated by AI systems under Article 50(2) and (4) AI Act are without prejudice to the obligation in Article 16(6) DSA for providers of hosting services to process notices on illegal content received pursuant to Article 16(1) DSA (29). In particular, marking or labelling applied to AI-generated or manipulated content should not influence the assessment and the decision on the illegality of the specific content under other regulatory frameworks. That assessment should be performed solely with reference to the rules governing the legality of content in accordance with those frameworks (30). For example, if a labelled deep fake image is flagged as potentially child sexual abuse material (CSAM) or infringing on trademark or copyright law, then the application of a label does not affect the potential illegality of the content under those regulatory frameworks. That assessment should be conducted solely on the basis of the respective, applicable laws (e.g. criminal law, trademark or copyright rules).

(97) Conversely, the machine-readable marking embedded in the content pursuant to Article 50(2) AI Act can help providers of VLOPs and VLOSEs to fulfil their obligations under Article 34 and 35 DSA to identify, analyse and assess systemic risks that may arise from the dissemination of content that has been artificially generated or manipulated, in particular the risk of the actual or foreseeable negative effects on democratic processes, civic discourse and electoral processes, including through disinformation, (31) and to put in place reasonable, proportionate and effective measures to mitigate those risks, tailored to the specific systemic risks identified. Providers of VLOPs and VLOSEs may also qualify as providers of AI systems if they meet the relevant conditions under the AI Act where they integrate AI systems into their services (see also Section 6.1.5).

(98) Providers of online platforms, online search engines and other relevant actors distributing content along the value chain are in particular encouraged to preserve and enable identification of the marks implemented by providers pursuant to Article 50(2) AI Act, so that the natural persons exposed to the content can be informed about its origin.