Transparency of AI-Generated Content

On 8 July 2026, the European Commission concluded, by way of an opinion, that the Code of Practice on Transparency of AI-Generated Content (Code) adequately covers the obligations laid down in Article 50(2), (4) and (5), of Regulation (EU) 2024/1689 (AI Act) and facilitates their effective implementation, an assessment endorsed by the European Artificial Intelligence Board on the following day.

The Code, published on 10 June 2026, thus constitutes an appropriate voluntary instrument recognised throughout the Union, on which providers and deployers of in-scope generative artificial intelligence systems may rely to demonstrate compliance with the marking and labelling obligations applicable to artificially generated or manipulated content from 2 August 2026.

Operators wishing to be included in the initial list of signatories must submit the accession form by 22 July 2026, at 18:00 CEST (Central European Summer Time).

1. The European Union's position

Article 50 of the AI Act establishes a horizontal set of transparency obligations addressed to providers and deployers of certain artificial intelligence (AI) systems, with a view to mitigating risks of deception and manipulation and preserving the integrity of the information ecosystem.

These obligations include:

• The marking, in a machine-readable format, of audio, image, video or text content generated or manipulated by AI systems, so that it is detectable as artificially generated or manipulated (paragraph 2).
• Disclosure by deployers that certain image, audio or video content constitutes a deepfake (paragraph 4).
• Disclosure that certain text has been artificially generated or manipulated where it is published for the purpose of informing the public on matters of public interest, unless it has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication.
• The provision of information to natural persons in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure (paragraph 5).

These obligations apply from 2 August 2026¹ and are legally binding irrespective of adherence to the Code.

Pursuant to Article 50(7) of the AI Act, the AI Office encouraged and facilitated the drawing-up of codes of practice intended to support the effective implementation of the obligations relating to the detection and labelling of artificially generated or manipulated content.
The Code was drawn up by independent experts through a multi-stakeholder process facilitated by the AI Office. The process involved, among others, providers of generative AI systems, developers of marking and detection techniques, associations representing deployers, civil-society organisations, academics and organisations specialising in transparency.

Following the Commission Opinion of 8 July and the AI Board’s adequacy assessment of 9 July, the Code was recognised as an appropriate voluntary instrument at Union level on which in-scope operators may rely to demonstrate compliance with the obligations laid down in Article 50(2), (4) and (5), of the AI Act, irrespective of their place of establishment or operation and of the competent market surveillance authority.

The Commission is also expected to publish interpretative guidelines on the scope of Article 50 (Guidelines), which will complement the Code by clarifying the scope of the legal obligations, including matters not covered by the measures set out in the Code.

2. Key features of the code

2.1. Nature and legal value

The Code does not replace the AI Act or the Commission Guidelines concerning Article 50, nor does it affect the other obligations laid down in the Regulation, including those applicable to high-risk AI systems and general-purpose AI models. The latter are covered by a separate code of practice published previously.

It is a practical framework, recognised at Union level, for the consistent, practicable and proportionate implementation of the transparency obligations.

Adherence to the Code is voluntary. Its value lies in its potential to operate as a mitigating factor for enforcement purposes and as evidence of compliance, allowing signatories to rely on the measures set out therein to demonstrate compliance with the legal obligations, reduce administrative burden and benefit from uniform legal certainty across all Member States. However, adherence does not constitute conclusive and/or irrefutable evidence of compliance – that assessment remains for the competent authorities to make in light of the circumstances of each case.

By contrast, operators choosing alternative compliance routes will need to demonstrate the adequacy of those measures on a case-by-case basis before the various authorities, with the foreseeable increase in information requests and individual scrutiny.

2.2. Structure: two sections, two groups of addressees

The Code is organised into two self-contained sections reflecting the architecture of Article 50 of the AI Act, which may be signed separately. It is not, however, possible to adhere only to individual commitments within a section:

• Section 1 – applicable to Providers (Article 50(2) of the AI Act). This section describes the commitments and measures intended to ensure that audio, image, video or text content generated or manipulated by an AI system is marked in a machine-readable format and detectable as artificially generated or manipulated. The technical solutions used must be effective, interoperable, robust and reliable, insofar as technically feasible, taking account of the specificities and limitations of the different types of content, implementation costs and the generally acknowledged state of the art, including relevant technical standards.
• Section 2 – applicable to Deployers (Article 50(4) and (5), of the AI Act). This section describes commitments concerning the disclosure and labelling of: (i) artificially generated or manipulated image, audio or video content constituting deepfakes, i.e. content resembling existing persons, objects, places, entities or events and which would falsely appear to be authentic; and (ii) text artificially generated or manipulated by AI and published for the purpose of informing the public on matters of public interest, unless subject to human review and editorial responsibility.

An organisation acting simultaneously as provider and deployer in relation to in-scope systems should consider signing both sections.

2.3. Supporting tools

The European Union has made available a set of icons that deployers may use to identify certain AI-generated content, thereby promoting harmonised visual recognition throughout the Union.

Signatories will also participate in Signatory Taskforces, cooperation structures intended to facilitate the sharing of practices and the further implementation of marking and labelling techniques in a field undergoing rapid technological development, including watermarking techniques, provenance metadata and detection mechanisms.

3. What economic operations should do

In view of a timetable which remains essentially unchanged², economic operators should now:

• Map and classify the generative AI systems developed, integrated or used within the organisation, including general-purpose systems embedded in products and workflows, and classify the organisation’s role in relation to each system (provider, deployer or both), since the applicable obligations and the relevant section of the Code depend on that classification.
• Decide by 22 July 2026 whether to adhere to the Code and whether to sign one or both sections, depending on the role performed, in order to be included in the initial list of signatories. The form must be signed by a senior officer authorised to bind the organisation. Adherence may take place at a later date, but inclusion in the list of signatories will await a subsequent review, the date of which has not yet been determined.
• Providers should implement marking and detection, i.e. adopt technical solutions for machine-readable marking (provenance metadata, watermarking and equivalent techniques aligned with the state of the art and emerging standards), ensure their robustness and interoperability, and review terms of use and acceptable-use policies to prevent the removal of markings. For systems already placed on the market before 2 August 2026, providers benefit from the transitional period until 2 December 2026 to complete implementation.
• Deployers should label content and organise editorial workflows, i.e. (i) identify use cases liable to generate deepfakes and design clear and visible disclosure mechanisms, taking into consideration the supporting tools made available by the European Union; and (ii) for text content on matters of public interest, document the human-review and editorial-responsibility processes that make it possible to benefit from the statutory exemption.
• Document and assign accountability. Compile evidence of compliance, including an audit trail, suitable for submission to market surveillance authorities; integrate the Article 50 obligations into internal AI governance; contractually cascade the requirements to suppliers and technology providers; and train the relevant teams.
• Track the Guidelines. Monitor publication of the final version of the Commission Guidelines on Article 50 of the AI Act, expected before 2 August 2026, which will clarify the scope of the obligations (AI interaction, machine-readable marking, deepfakes and public-interest text) and the means of demonstrating compliance.

4. Key dates

The Technology team will continue to closely monitor legislative and regulatory developments applicable to the artificial intelligence sector and remains fully available to address any further questions.

_______________________