Code of Practice on Transparency of AI-Generated Content Finalised

The Code of Practice on Transparency of AI-Generated Content is a voluntary compliance framework developed by independent experts through a multi-stakeholder process facilitated by the AI Office, covering the marking and labelling obligations in Article 50(2), (4), and (5). The drafting process moved through several stages: a first draft was published on December 17, 2025, followed by a second draft on March 5, 2026 that incorporated feedback gathered from hundreds of participants (industry, academia, civil society, EU member states via the AI Board, and members of the European Parliament) through a survey, stakeholder meetings, and workshops held in January 2026. The final version was published in June 2026.

On July 20, 2026, the European Commission and the AI Board formally confirmed the Code as adequate for demonstrating compliance with Article 50’s transparency obligations; a separate development from the Commission’s final Guidelines on Article 50, published the same day, which is a distinct document covering legal interpretation rather than practical implementation.

The final Code is structured in two sections: Section 1 addresses providers of generative AI systems and their marking/detection obligations under Article 50(2); Section 2 addresses deployers who generate or manipulate deepfakes, or publish AI-generated text on matters of public interest, under Article 50(4). Compared to the first draft, the final version made several previously mandatory measures optional, expanded flexibility around marking methods, introduced a standardised EU labelling icon, and set a concrete interoperability deadline of February 2, 2027.

As mentioned, signing the Code is voluntary. By the end of July 2026, approximately 190 companies and organisations had signed on.

Worth being precise about what this Code does and doesn’t cover: it’s addressed primarily to providers, meaning the companies building the generative AI systems, on how to mark their own output. It doesn’t relieve a deployer’s separate duty under Article 50(4) to ensure genuine human review and editorial responsibility before publishing AI-assisted content. A provider signing this Code says nothing about whether a publisher using that provider’s tools has met its own, separate obligation.

Sources: