Summary

• The European Commission has published a voluntary code of practice for marking and labelling AI-generated content.
• Providers and deployers can use the code to support compliance with AI Act Article 50 transparency obligations.
• The code turns AI disclosure into an operational issue for platforms, publishers, vendors, and enterprise AI users.

The European Commission has published a voluntary code of practice for marking and labelling AI-generated content, giving AI providers and deployers a clearer route through one of the AI Act’s most visible transparency requirements.

The code supports compliance with Article 50 of the AI Act, which covers transparency obligations for providers and deployers of generative AI systems. Those obligations apply from 2 August 2026 and deal with risks including deception, manipulation, deepfakes, synthetic content, and certain AI-generated public-interest publications.

Two sections define the shape of the code. The first covers providers and the marking and detection of AI-generated and manipulated content, while the second covers deployers and the labelling of deepfakes and AI-generated or manipulated text. The EU has also created a set of icons that deployers may use when labelling AI-generated material.

The code does not replace the AI Act or the Commission’s formal guidelines on Article 50. It is intended to act as a practical framework that organisations can rely on if it receives a positive adequacy assessment from the Commission and the AI Board. Providers and deployers that do not follow it will still have to show that their own measures are adequate, with market surveillance authorities assessing those approaches individually.

That makes the code commercially useful even though adherence is voluntary. Large AI providers, enterprise software vendors, public bodies, media organisations, online platforms, marketing departments, and other deployers will need a defensible way to show how synthetic content is marked, detected, disclosed, and explained to users.

The legal duty is straightforward enough: users should not be tricked into believing synthetic content is human-produced or authentic when it is not. Implementation is much harder. Labelling rules have to work across text, image, audio, video, chatbots, synthetic documents, automated publications, recommender systems, and content flows where AI-generated material may be edited, republished, or mixed with human work.

Procurement teams will need to ask whether generative AI systems support marking and detection in ways that clients can actually use. Product teams will need to decide how disclosure appears to users without damaging accessibility or burying the warning in interface clutter. Governance teams will need policies for when AI content is labelled, who owns the disclosure, and how exceptions are handled.

Publishers and platforms face a particularly awkward line around public-interest text. The AI Act does not treat every AI-assisted draft as a synthetic publication requiring disclosure, especially where human review and editorial responsibility are present. Organisations using generative systems in communications, content production, customer support, public-service messaging, or market research will still need internal rules that can survive audit and everyday workflow pressure.

Detection remains imperfect, which limits any neat compliance story. Watermarking, metadata, provenance standards, and content credentials can all help, but content can be copied, stripped of metadata, screen-recorded, compressed, translated, or deliberately altered. A practical code can create a useful baseline without pretending that labels alone solve synthetic media risk.

European AI regulation is now moving from legislative text into operational machinery. The organisations best placed for compliance will not simply be those with the largest legal teams, but those able to translate transparency duties into product controls, audit trails, customer guidance, and user-facing disclosure that survives normal use.

The Commission’s Code of Practice on Transparency of AI-Generated Content now gives that process a reference point. It also gives regulators a benchmark against which to judge whether voluntary commitments match the requirements of a law that is about to become part of live AI operations.