EU Code of Practice on Transparency of AI-generated Content

The Code of Practice on Transparency of AI-generated Content is a voluntary instrument that helps providers and deployers of generative AI systems demonstrate compliance with the marking, detection and labelling obligations in Article 50(2), (4) and (5) of the EU AI Act. It was drawn up by independent experts in a multi-stakeholder process facilitated by the EU AI Office, published in final form on 10 June 2026, and confirmed by the European Commission and the AI Board as an adequate voluntary tool. Adherence is voluntary; the underlying Article 50 transparency requirements are legal obligations, applicable from 2 August 2026 (Source: digital-strategy.ec.europa.eu).

Full title: Code of Practice on Transparency of AI-generated Content Enacting body: European Commission AI Office, via independent expert working groups Legal status: Voluntary; assessed as an adequate tool by the Commission and the AI Board Scope: Providers and deployers of generative AI systems within the scope of Article 50; Section 1 is additionally open to providers of marking and detection solutions and to providers of AI models

It is the transparency counterpart to the GPAI Code of Practice, which operationalises Articles 53 and 55 for general-purpose AI model providers. The two instruments cover different obligations and have separate signatory lists.

Status and timeline

Article 50 addresses risks of deception and manipulation and applies to providers and deployers of generative AI systems, complementing the separate regimes for high-risk AI systems and for general-purpose AI models (Source: digital-strategy.ec.europa.eu).

The drafting process ran from autumn 2025 to mid-2026. The Commission launched a consultation and call for expressions of interest in September 2025, ran eligibility checks and selected chairs and vice-chairs in October 2025, and held a kick-off plenary on 5 November 2025. The first working-group meetings followed on 17–18 November 2025 and the first draft was published on 17 December 2025. A second drafting round opened with working-group meetings on 12 and 14 January 2026 and workshops on 21–22 January 2026; the second draft appeared on 3 March 2026, followed by working-group meetings and workshops from 16 to 27 March 2026. Draft guidelines on the implementation of the Article 50 obligations were published for consultation on 8 May 2026. The closing plenary and publication of the final code took place on 10 June 2026 (Source: digital-strategy.ec.europa.eu). OpenAI announced its support for the code the following day (Source: ec.europa.eu; openai.com).

Organisations wishing to appear on the initial signatory list published before the AI Act's 2 August 2026 date of application had to submit completed forms by 27 July 2026 at 18:00 CEST (Source: digital-strategy.ec.europa.eu). The code remains open for signature after that date, and the signatory list is updated on an ongoing basis.

The drafting considered feedback from the Commission's consultation on transparency requirements for certain AI systems and three expert studies on technical solutions to mark and detect AI-generated content, commissioned by the AI Office (Source: digital-strategy.ec.europa.eu).

Structure

The code has two sections, mirroring the split in Article 50 between providers and deployers, and was drafted by two correspondingly structured working groups, each led by independent chairs and vice-chairs.

Section 1 — Providers. Rules for marking and detection of AI-generated and manipulated content. Working group 1 addressed the requirement that outputs of AI systems — audio, image, video and text — be marked in a machine-readable format and be detectable as artificially generated or manipulated, and that the technical solutions employed be effective, interoperable, robust and reliable as far as is technically feasible. That feasibility standard takes into account the specificities and limitations of different content types, the costs of implementation, and the generally acknowledged state of the art as reflected in relevant technical standards (Source: digital-strategy.ec.europa.eu).

Section 2 — Deployers. Rules for labelling of deepfakes and of AI-generated and manipulated text. Working group 2 addressed the requirement that deployers disclose content that is artificially generated or manipulated and constitutes a deepfake — defined as image, audio or video content resembling existing persons, objects, places, entities or events that would falsely appear to a person to be authentic or truthful — and that they disclose AI-generated or manipulated text publications informing the public on matters of public interest, unless the publication has undergone human review and is subject to editorial responsibility (Source: digital-strategy.ec.europa.eu).

Both working groups also considered cross-cutting questions, including the horizontal requirements governing information provided to natural persons under Article 50(5), and cooperated with actors across the value chain (Source: digital-strategy.ec.europa.eu).

A Tech Policy Press explainer describes the code as introducing a more flexible disclosure regime for artistic, creative, satirical, fictional and similar works, and as implementing the AI Act's exemption for AI-generated or AI-assisted published text that has undergone human review or editorial control (Source: techpolicy.press).

The code is accompanied by two related instruments: a set of EU icons that deployers of generative AI systems may use to label AI-generated content, and guidelines on the scope of the transparency obligations laid down in Article 50 (Source: digital-strategy.ec.europa.eu).

Effect of signing

Following the positive adequacy assessment, providers and deployers that sign the code can rely on its measures to demonstrate compliance with the AI Act's rules on labelling and detection of AI-generated content, deepfakes and certain text publications. The Commission describes this as reducing administrative burden and giving signatories predictability, legal certainty and consistent treatment across member states. Providers and deployers that choose to comply through other means must demonstrate that those measures are adequate, assessed individually by the different national market surveillance authorities (Source: digital-strategy.ec.europa.eu). The law firm Cooley characterises the same arrangement as giving signatories a degree of presumption of conformity and a more favourable enforcement posture, with non-signatories facing closer scrutiny (Source: cooley.com).

The Commission states that the code does not replace the AI Act or the Commission's Article 50 guidelines, and presents it as an EU-wide recognised practical framework for signatories to demonstrate compliance (Source: digital-strategy.ec.europa.eu).

Signatories are invited to join two Signatory Taskforces, due to launch in September 2026, whose stated objectives are to share industry practices, provide feedback on the implementation of transparency measures, and advance the state of the art and the effective implementation of the code (Source: digital-strategy.ec.europa.eu).

Signatories

By the end of July 2026, about 190 organisations across sectors including IT, telecommunications, education and retail had signed the code, ahead of the marking obligations taking effect on 2 August 2026. As of the list's 5 August 2026 update, Section 1 had 82 signatories and Section 2 had 152; organisations may sign either or both. The Commission notes that about half of the signatories are small and recent companies. Although the legal obligation in Section 1 applies only to providers of AI systems, that section may also be signed by providers of marking and detection solutions and by providers of AI models that could help generative-AI providers implement the code's measures (Source: digital-strategy.ec.europa.eu).

The Commission's own examples of prominent AI companies among Section 1 signatories are Aleph Alpha, Anthropic, Black Forest Labs, Cohere, Google, Meta, Microsoft, Mistral, OpenAI and Synthesia. Its Section 2 examples are Bulgari, Fastweb, Getty Images, Iberdrola, Lenovo and Lufthansa (Source: digital-strategy.ec.europa.eu). Reporting published on 31 July 2026 put the figure at more than 180 organisations, including Google and Meta (Source: theguardian.com).

Implementation

Anthropic confirmed on 11 August 2026, through an updated support page, that it will watermark text and files generated by its models in order to comply with the code. All Anthropic models released after 2 August 2026 carry the watermarking automatically, with the C2PA open standard used for files, and the company said it will extend support to older models. Anthropic said the watermark is applied at the model level, travels with text that is copied and pasted, "may persist through some editing," and applies across the Claude platform API, Claude, Claude Code, Claude Cowork and Claude Tag (Source: techcrunch.com). See Data Provenance, C2PA, and Watermarking and AI Content Provenance.

Enforcement context

The Article 50 obligations sit within the AI Act's general enforcement architecture rather than being created by the code. Synthetic text, images, video and audio designed to look authentic must be visibly marked as AI-generated and carry a digital watermark, with systems already on the market given four additional months to comply. Non-compliance carries fines of up to €15 million or 3% of worldwide global turnover, a lower band than the Article 99 maxima applying to prohibited practices. Exemptions cover personal content and "evidently artistic," satirical and fictional works. Green MEP Sergey Lagodinsky called the rules "a matter of democracy protection," while Boniface de Champris, AI policy lead at CCIA Europe, said the Commission's July guidance expanded the deepfake definition beyond the 2024 text so that "almost everything gets labelled" (Source: theguardian.com).

Relationships

• depends-on: EU AI Act (Regulation 2024/1689) — the code operationalises Article 50(2), (4) and (5) and has no independent legal force.
• related: EU General-Purpose AI Code of Practice (2025) — the parallel voluntary code for Articles 53 and 55, with a separate signatory list and a different addressee class.
• supports: AI Content Provenance, Data Provenance, C2PA, and Watermarking — the code specifies machine-readable marking and detection measures, and C2PA is the standard signatories have named for files.
• supports: Synthetic Media / Deepfakes, AI and Misinformation — deepfake labelling under Section 2.
• related: EU AI Office — facilitator of the drafting process; supervision of the underlying obligations rests with it and the national market surveillance authorities.
• related: Anthropic, OpenAI, Google DeepMind, Meta AI, Microsoft, Mistral AI, Cohere, Aleph Alpha — Section 1 signatories named by the Commission.

Open questions

• Whether the two Signatory Taskforces, due to launch in September 2026, publish their outputs or operate as closed industry fora is not yet disclosed.
• How national market surveillance authorities will assess adequacy for non-signatories, and whether their assessments converge, is untested as of August 2026.
• Whether the watermarking approaches adopted by signatories are interoperable across providers, as Section 1 requires "as far as technically feasible," has not been independently evaluated.

Provenance note: The code's own text (PDF) has not been ingested; citations rest on the Commission's canonical pages and are marked (Source: URL). Upgrade them to a sources/ wikilink once the primary text is ingested.