European Commission publishes Code of Practice for labeling AI-generated content

What the code covers

The European Commission has published a Code of Practice on marking and labeling AI-generated content, according to a report on dev.to. The document is meant to help organizations that deploy generative AI systems meet the transparency obligations attached to the EU AI Act, and it treats two sides of the same problem: how AI-generated material should be labeled so audiences can recognize it, and how it can be detected once it is circulating.

Deepfakes receive particular attention, but the scope extends to synthetic images, audio, video and text. The guidance is aimed at deployers, meaning companies that put generative tools in front of customers rather than the developers building the underlying models. The dev.to report also notes that the code references EU icons that can be attached to AI-generated output and sets out how disclosure should work when content has been generated or altered in the public interest.

Commission materials cited by dev.to place the code's applicability around the third quarter of 2026, which makes it a near-term concern for businesses already publishing AI-drafted marketing copy, generated campaign imagery or chatbot responses.

What deployers should be doing

The question the code poses is not whether a business uses AI internally, but whether the people on the other end of a website, feed or chat window can tell when output came from a machine. A convincingly altered video that could make viewers believe an event took place presents a different transparency problem from an internal draft that never leaves the company, and the code does not force every use case into a single template.

The dev.to article outlines a workable sequence for teams adapting to the guidance: map every point where AI-assisted output reaches customers or the public, covering images, video, audio, text and chatbot interactions; separate freshly generated material from manipulation of existing recordings; establish who decides when a label appears and in what form; embed disclosure into templates and publishing workflows rather than leaving it to individual judgment at the moment of release; and review what labeling and detection capabilities suppliers and platforms already support.

Chatbots get a specific mention in the report's reasoning: identifying a system as artificial before a customer starts typing sets honest expectations for the conversation that follows.

Part of a wider August package

The labeling code did not arrive in isolation. According to dev.to, it was published alongside other EU activity in August 2026 that points toward continued attention on online safety and consumer protection. The Better Internet for Kids network's #NotOnOurFeed campaign runs from 24 August to 25 September 2026 and covers prevention, protection, reporting and support around cyberbullying, including AI-enabled abuse.

Separately, the Commission opened a call for tenders for a study on how online marketplace design influences user behavior, linked to the Digital Services Act and Very Large Online Marketplaces. The report emphasizes that the tender is not a new rule and draws no conclusions about specific design choices, but it signals that the Commission is gathering evidence on how interfaces steer consumers. Executive Vice-President Henna Virkkunen's participation at Gamescom was also confirmed in the same period, placing the games industry within the EU's broader digital policy agenda.

Why it matters

The EU AI Act's transparency requirements stop being abstract the moment a regulator publishes concrete guidance on how to satisfy them. This code gives deployers a shared standard for disclosure, which matters for three reasons: customers get consistent signals about what they are viewing or interacting with, businesses reduce the legal guesswork around AI-generated output, and platforms gain a common vocabulary, down to the icons, for marking synthetic content. For any organization running public-facing generative AI today, the practical work is mapping where machine-made content reaches an audience and making labeling an ordinary, repeatable step in the pipeline rather than an afterthought.