Labelling AI Content: The EU Code of Practice Explained

On 10 June 2026 the European Commission published its final Code of Practice on labelling AI-generated content (press release IP/26/1328). The Code is voluntary and sets out practical steps for meeting the transparency duties of the EU AI Act, which apply from 2 August 2026. For small and mid-sized businesses that use AI day to day, this is a concrete roadmap, not an abstract legal topic.

What applies from 2 August 2026

On that date, the transparency obligations under Article 50 of the AI Act take effect. They apply regardless of whether your AI is high-risk; the point is the open disclosure of AI interaction and AI content.

• Interactive AI such as chatbots: users must be able to recognise that they are talking to an AI, not a human.
• Deepfakes: AI-generated or AI-manipulated images, audio and video must be labelled as such.
• AI texts on matters of public interest: must be labelled when published without human or editorial oversight.
• Providers of generative AI: must mark their AI outputs in a machine-readable way, so they are technically detectable as artificially generated.

The new Code of Practice mainly operationalises two of these: machine-readable marking by providers (Article 50(2)) and the visible labelling of deepfakes and published texts by deployers (Article 50(4)). The plain chatbot disclosure is the related duty under Article 50(1) and also applies from 2 August 2026.

Provider or deployer: where do you stand?

The AI Act distinguishes two roles, and the Code is split into two sections accordingly. The decisive point for most companies: anyone who merely uses AI is a deployer, not a provider.

• Providers: develop or distribute generative AI systems and must ensure the machine-readable marking of outputs.
• Deployers: use AI, and must visibly label deepfakes and AI texts on public topics, and disclose the AI in chatbots.

What this means for mid-sized businesses

Most mid-sized companies are deployers. Even so, real obligations arise as soon as generative AI becomes visible to the outside world:

• A chatbot on your website or in customer service: a clear notice such as "You are chatting with an AI" meets the requirement.
• AI-generated images or videos in marketing and social media: check whether a deepfake label is needed.
• AI-created texts published without human review that concern public-interest topics: label them.
• Keep an internal record of where AI is used at all. That is the basis for any labelling.

How to label: the EU icon

For visible labelling, the Code of Practice provides an official EU icon that may be used freely and without attribution. There are two variants: "AI GENERATED" for fully AI-created content and "AI MODIFIED" for partly AI-altered content. This answers the question "How do I label?" in concrete terms.

• Images and videos: place the icon clearly visible, for example top right; in videos at the start and after ad breaks.
• Published texts: place the notice near the headline or at the start of the text.
• Audio-only with no screen: prepend a short, clearly understandable audible notice.
• The label must be clearly recognisable and accessible at the latest on first contact.

When you do not have to label

The Code sets out clear exceptions. Not every piece of AI content needs to be labelled:

• Published texts that have undergone human or editorial review and for which a person holds editorial responsibility.
• Artistic, creative, satirical or fictional works: here labelling is done so it does not impair the work, for example in accompanying text or credits.
• Content whose use is legally permitted for detecting, preventing or prosecuting criminal offences.

Voluntary, but with an upside

The Code is not mandatory. However, those who sign it can, once approved by the Commission and the AI Board, more easily demonstrate compliance with the relevant AI Act duties. In addition, the Commission has announced practical guidelines that will further clarify the scope. The Code was drawn up by six independent experts with the involvement of more than 180 stakeholder groups.

People in Europe have a right to know whether what they see, hear or read was created or altered by AI. Transparency is how we protect trust. (paraphrased: Henna Virkkunen, Executive Vice-President of the European Commission)

The bridge to ISO 42001

Labelling is not a standalone topic but part of a whole: those who know which AI systems run in their organisation, who is accountable for them and how their outputs are handled meet transparency duties almost in passing. That is exactly what an AI management system (AIMS) under ISO/IEC 42001 delivers. It makes the AI Act's duties structured and demonstrable. I build such systems so they work in day-to-day operation, and I check beforehand whether your labelling holds up.

If you are unsure whether you count as a provider or a deployer, and which labelling your AI requires, a short initial conversation is enough to classify your systems. Source: European Commission, press release IP/26/1328 of 10 June 2026.