The Commission publishes the Code of Good Practice on the labelling and identification of content generated with AI

On 10 June 2026, the European Commission published the Code of Practice on the Transparency of Content Generated by Artificial Intelligence (hereinafter referred to as the "Code"), in accordance with the obligations set out in Article 50 of Regulation (EU) 2024/1689 of the European Parliament and of the Council, of 13 June 2024, which establishes harmonized rules on artificial intelligence (hereinafter referred to as the "AI Regulation").

In this respect, the Code addresses operators who voluntarily adhere to it and who, depending on their role in the AI value chain, are subject to the transparency obligations set out in Article 50 of the AI Regulation, although it also allows for the voluntary adherence of other relevant stakeholders, such as providers of generative AI models or external providers of labelling and detection solutions, in order to facilitate compliance by providers integrating these technologies into their own systems.

In any case, adherence to the Code does not itself constitute conclusive evidence of compliance with the AI Regulation, although it serves as a reference tool to demonstrate this compliance to the competent market surveillance authorities.

To this end, the Code is structured into two sections:

• Firstly, Section 1, addressed to providers of generative AI systems, aims to serve as a guide to demonstrate compliance with the obligations set out in Article 50, paragraphs 2 and 5, of the AI Regulation.
• Secondly, Section 2, addressed to those responsible for the implementation of generative AI systems, also aims to serve as a reference tool to demonstrate compliance with the obligations set out in Article 50, paragraphs 4 and 5, of the AI Regulation.

The main commitments and measures outlined in each section of the Code are as follows:

• Section 1, relating to providers of generative AI systems, is structured around four main commitments:
  • Labelling of content generated or manipulated by AI: The first commitment involves the implementation of labelling solutions that allow for the identification, in a human-readable format, that the content has been generated or manipulated by AI. The Code is based on the premise that, in most cases, a single labelling technique will not be sufficient to meet the requirements of effectiveness, interoperability, robustness, and reliability demanded by the AI Regulation. Therefore, it promotes a layered approach, combining, when appropriate, digitally signed metadata and watermarks. Additionally, adherents must make efforts to preserve existing marks and avoid their intentional removal or alteration, including restrictions in their terms of use or in the contractual documentation.
  • Detection of markings: The second commitment focuses on ensuring that signatories not only mark the content but also provide mechanisms to detect these markings. These detection solutions must be available to users, those responsible for implementation, third-party integrators, competent authorities, and other legitimate stakeholders. Detection results must be provided clearly, understandably, and accessibly, so that individuals exposed to the content can understand whether it has been generated or manipulated by AI and, where applicable, the technique on which this conclusion is based.
  • Quality of labelling and detection solutions: The third commitment requires that labelling and detection solutions be effective, reliable, robust, and interoperable, as long as it is technically feasible. In particular, these solutions must enable the clear identification of whether content has been generated or manipulated by AI, function correctly in different contexts and types of content, withstand common alterations or attempts at manipulation, and work towards greater compatibility between systems and providers. Furthermore, the Code encourages adherents to contribute to the development of standards, best practices, and technical solutions that progressively improve the state of the art in this field.
  • Testing, verification, and compliance: The fourth commitment requires signatories to establish and maintain internal processes to document, test, verify, and oversee compliance of their labelling and detection solutions in accordance with Article 50, paragraphs 2 and 5, of the AI Regulation. In particular, they must document how they implement the measures of the Code, conduct tests prior to bringing the system to the market or putting it into service, and periodically throughout its life cycle, train relevant staff, and remedy any deficiencies found. Furthermore, signatories must cooperate with market surveillance authorities by providing, where applicable, the necessary documentation and access to their marking and detection solutions, without prejudice to the protection of trade secrets and confidential information.
• Section 2establishes four main commitments for those responsible for implementation:
  • Disclosure of deepfakes and published texts:The first commitment requires those responsible for implementation to provide clear, distinguishable, and accessible information whenever a deepfake (ultra-spoofing) or a published text for informative purposes has been generated or manipulated using AI. To this end, the Code foresees the use of the European AI icon, or an equivalent label, which must be prominently displayed from the moment the user is exposed to the content. The European AI icon is provided in three different forms:
    • "GENERATED BY AI", intended to identify deepfakes or published texts fully generated by AI;
    • "MODIFIED WITH AI", intended to identify deepfakes or published texts that have been partially manipulated or modified with AI;
    • "AI", as a basic icon that allows, where applicable, the addition of an interactive layer with more information or an alternative text label. In visual content, the labelling must be easily perceptible; and, when a visual indication is not possible, an audible alert or another appropriate mechanism must be used.
  • Internal compliance processes: The second commitment requires signatories to have internal processes, awareness measures, and review mechanisms in place to ensure the correct application of the marking obligations set out in Article 50, paragraphs 4 and 5, of the AI Regulation. In particular, they must document how they implement the European AI icon or an equivalent label, train or raise awareness among the relevant personnel, and establish mechanisms to review and correct any omission or error in the labelling. They must also respond to justified notifications of incorrectly labelled or unlabelled content and cooperate with the competent authorities when necessary.
  • Disclosure in artistic, creative, or similar works: The third commitment regulates cases where a deepfake is part of an artistic, creative, satirical, fictional, or similar work or programme. In these cases, signatories must indicate that the content has been generated or manipulated by AI, but in a manner that does not affect the normal enjoyment, viewing, or exploitation of the work. Disclosure must be clear, distinguishable, and accessible, and can be made through the European AI icon or an equivalent label, positioned appropriately based on the type of work and presentation context, such as in credits, descriptive notes, digital interfaces, entries, brochures, or packaging.
  • Human review and editorial control of published texts: The fourth commitment refers to published texts generated or manipulated by AI. Signatories who are media service providers and are already subject to editorial standards or regulatory, co-regulatory, or self-regulatory frameworks may rely on the exemption from the disclosure obligation provided for in Article 50.4, second paragraph, of the AI Regulation, applying their existing human review and editorial control procedures. All other adherents, including those who do not have these procedures, must establish, adapt, or maintain appropriate policies for human review or editorial oversight prior to publication, ensuring that an individual or entity takes editorial responsibility for the publication.

As next steps, the Code remains open for signature by providers and those responsible for the implementation of AI systems. Once approved as an appropriate instrument, signatories will be able to rely on it to demonstrate compliance with the transparency obligations of the AI Regulation applicable from 2 August 2026.

All this, without prejudice to the fact that certain obligations regarding the labelling of content generated or manipulated by AI will not be binding until December 2026, due to their postponement following the formal adoption of the Digital Omnibus Regulation on AI.

Moreover, the Commission plans to complement the Code with practical guidelines to clarify and develop those aspects not covered by the Code itself.

Informative note from the TMT department of ECIJA Madrid.