The European Commission has published a set of EU icons that creators, publishers and other deployers of generative AI systems may use to label AI-generated content, in line with the AI Act's transparency rules. The icons are freely available and form an integral part of Section 2 of the Code of Practice on marking and labelling of AI-generated content. They are meant to help people recognise, clearly and distinguishably, that content has been artificially generated or manipulated.
What has to be labelled
Not all AI-generated content needs a label. The icons support compliance with Article 50(4) of the AI Act, which requires deployers to disclose two categories:
- Deepfakes: AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful.
- Text on matters of public interest: AI-generated or manipulated text published to inform the public that did not undergo human review or editorial control and for which no legal or natural person has assumed editorial responsibility.
Three limits apply. For evidently artistic, creative, satirical, fictional or analogous works, disclosure only has to be made in an appropriate manner that does not hamper the display or enjoyment of the work. The obligation does not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences. And it does not apply to AI-generated text that has gone through human review or editorial control, with a person holding editorial responsibility.
The three icons
The icons come in four colour variations (black, white, and each at 50% transparency), as SVG and PNG files.
- Basic icon: for any case where AI was involved in creating a deepfake or published text, or where a custom text label or an interactive second layer is used. Example: a deepfake video with the text "voices generated with" followed by the icon.
- Fully AI-generated: for deepfakes or text generated entirely by AI, with no human-created elements or human editorial control apart from prompting. Examples: fully AI-generated deepfake videos of politicians or fictional events, AI-composed music or art (which may enjoy limited disclosure for artistic work) and AI-generated news summaries.
- Partially AI-modified: for pre-existing, human-made content that AI partly modified into a deepfake or into text on public-interest matters. Examples: a face in an authentic photograph swapped for a politician's, or an authentic photo of an empty apartment furnished with AI.
User testing informed the design, and performance improved across all measures when the basic icon was accompanied by a text label such as "modified."
How to display them
The icon should be clearly perceivable and distinguishable at the latest at first exposure to the content, with no overlay elements in the way. It should be embedded directly in the deepfake or text, except for creative works, unless an equivalent alternative such as a user-interface overlay is available, and it must stay visible when content is reshared or downloaded. Users are encouraged to make it a clearly visible size, keep any label in plain language, support alt text or ARIA labels for assistive technologies, keep time-limited disclosures on screen long enough to be read, and make any second interactive layer navigable with assistive tools.
What the icons do not do
Use of the icons is optional, but the Article 50 labelling requirements are not, and using them does not establish legal compliance by itself, since deployers remain responsible for meeting Article 50. They are free to use without attribution to the Commission or the AI Office. Signatories of the Code of Practice must follow its placement specifications, while non-signatories' use should not be read as signaling adherence to the code.
For how the transparency rules fit the AI Act timetable, see the Omnibus deferral, and for a lab's own response, Claude's watermarking of its output.