The European Commission has published a set of optional AI-generated content labels that creators, publishers and other deployers of generative AI systems may use to mark certain content generated or manipulated with artificial intelligence.

The icons are part of Section 2 of the new Code of Practice on marking and labelling AI-generated content. They are freely available in PNG and SVG formats and can be used without attribution to the European Commission or the AI Office.

Their use is optional. The disclosure obligations under Article 50 of the EU AI Act, however, are legal requirements and will apply from 2 August 2026.

Desk with notebook and coffee illustrating a free AI training course

Image source: Envato

What content needs to be labelled?

The disclosure requirements do not apply to every use of AI.

Under Article 50(4), deployers must disclose the artificial nature of:

Deepfakes involving AI-generated or manipulated images, audio or video that could falsely appear authentic, and AI-generated or manipulated text published to inform the public about matters of public interest when it has not undergone human review or editorial control and no person or organisation holds editorial responsibility.

The icons are intended to help people recognise clearly that this content has been artificially generated or manipulated.

AI Usage Notice: In preparing this article, AI tools were used with careful human oversight and editing. We believe in transparency regarding the use of AI in our work.
AI Usage Notice: In preparing this article, AI tools were used with careful human oversight and editing. We believe in transparency regarding the use of AI in our work.

Three icons for different uses

The EU has developed three types of icons.

Basic icon

General AI involvement

May be used when AI was involved in creating or modifying relevant content, including when it appears alongside a more specific text label.

Fully AI-Generated

Created entirely by AI

Intended for content created entirely by AI, without human-created elements or editorial control apart from prompting. Examples include fully AI-generated deepfake videos, AI-generated news summaries, music or visual content.

Partially AI-Modified

Human content altered with AI

Intended for pre-existing human-made content that has been altered using AI. This could include replacing a person's face in an authentic photograph or adding AI-generated furniture to an image of an empty apartment.

Each icon is available in four visual versions: black, white, black with 50% transparency and white with 50% transparency.

Looking for support around AI?

We (AImpactful 🙂) work with newsrooms, NGOs, institutions, teams, and individuals who need workshops, advisory support, or content production.

Text labels can improve clarity

The icons underwent user testing during their development.

According to the European Commission, their performance improved when the basic icon was accompanied by a clear text label, such as "modified".

This means that an icon alone may not always provide enough information. A short explanation can help people understand whether content was fully generated, partially modified or altered in a specific way.

Labels should use plain language and avoid jargon, confusing wording and abbreviations other than "AI".

Where should the icon appear?

The Commission has set clear rules for visibility and placement:

  • The icon should be clearly visible and distinguishable when a person first encounters the content
  • It should normally be embedded directly into the deepfake or published text, unless an equivalent alternative is available, such as a user interface overlay
  • The icon should remain visible when the content is downloaded or reshared
  • Icons should be displayed at a clearly visible size and, where possible, made readable by assistive technologies through alt text or ARIA labels
  • When a disclosure appears only temporarily, it should remain on screen long enough to be read and understood

Exceptions and limited disclosure requirements

The AI Act includes several exceptions:

Editorial oversight AI-generated text published to inform the public on matters of public interest does not need to be labelled when it has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication.
Law enforcement The disclosure obligation does not apply when the relevant use is authorised by law for the detection, prevention, investigation or prosecution of criminal offences.
Creative works For evidently artistic, creative, satirical, fictional or analogous works and programmes involving deepfake content, disclosure is still required, but it may be presented in a way that does not interfere with the display or enjoyment of the work.

Icons do not guarantee legal compliance

Using the EU icons does not establish compliance with the AI Act by itself.

Deployers remain responsible for ensuring that their disclosures meet the requirements of Article 50. Organisations that sign the Code of Practice must also implement its relevant measures, including its placement and presentation specifications.

Non-signatories may use the icons freely, but doing so does not mean that they have joined or formally committed to the Code.

The icons offer a common visual language for AI transparency. Their usefulness will depend on whether people can quickly understand what the symbol means and how AI was used in creating or modifying the content.

Read AI Briefings