The European Commission has published free icons that publishers, creators and other deployers can use to identify certain AI-generated or AI-manipulated content. The icons and the Code of Practice that contains them are voluntary. The underlying duties in Article 50 of the EU AI Act are not—but they apply only to specified content and actors, not to every organisation that uses generative AI.
Article 50’s relevant transparency obligations have applied since 2 August 2026. For deployers, Article 50(4) covers two categories: image, audio or video that qualifies as a “deep fake”; and AI-generated or manipulated text published for the purpose of informing the public on matters of public interest. A deep fake is an AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful. The disclosure duty falls on the deployer of the AI system, and Article 50(5) requires clear, distinguishable information no later than first exposure.
The icons
The Commission provides three meaning-based icon variants: Basic, Fully AI-Generated and Partially AI-Modified. Each is available in four visual treatments—black, white, black at 50% transparency, and white at 50% transparency—and in SVG and PNG packages. The Basic icon can be used for in-scope material involving AI or with a custom text label or interactive second layer. One Commission example is a deepfake video labelled “voices generated with” alongside the icon.
The fully AI-generated icon is for material generated entirely by AI, without human-created elements or editorial control beyond prompting. Examples include synthetic videos of politicians or fictional events and AI-generated news summaries.
The partially AI-modified icon is for pre-existing human-made material altered with AI—for example, a politician’s face swapped into an authentic photograph or an empty flat digitally furnished. The Commission reports that user testing found performance improved across all measures when the Basic icon was paired with a text label such as "modified". The published summary does not provide the full study methodology or numerical results.
The icons appear in Annex I of the Code of Practice on Transparency of AI-Generated Content. The final code was published on 10 June 2026, following a multi-stakeholder process involving more than 187 participants and six independent experts. The Commission concluded on 8 July that the Code adequately covered the relevant Article 50 obligations; the AI Board adopted its assessment the following day. Adherence to the code is voluntary, as is use of the icon set. Icons may be used without attribution to the Commission or AI Office. Using an icon does not show that an organisation has signed the Code and does not by itself prove legal compliance. Deployers remain responsible for meeting the law; organisations that do not sign the Code must demonstrate compliance through other adequate means.
How the icons should be shown
The Commission’s summary says an icon should be clearly perceivable by the time a person first encounters the material, positioned so other overlays do not obscure it, and embedded in the content unless an equivalent alternative—such as a user-interface overlay—is available. It should remain visible when the material is reshared or downloaded. The detailed design and placement specifications appear in Section 2 of the Code; these are commitments for Code signatories.
The Act’s general legal standard is that required information be clear and distinguishable by first exposure. The icon itself is optional. This is not a requirement to label every item made with AI. For deepfakes forming part of an evidently artistic, creative, satirical, fictional or analogous work or programme, Article 50(4) limits the duty to disclose the generated or manipulated content in an appropriate way that does not hamper the work’s display or enjoyment. The duty also does not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences.
For public-interest text, the disclosure exception applies where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for publication. The Act does not require that responsible person or organisation to be publicly named. Article 50(2) imposes a separate obligation on providers of AI systems—including general-purpose AI systems—that generate synthetic audio, image, video or text. They must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, using technical solutions that meet the statutory standards as far as technically feasible.
The duty does not apply to the extent a system assists with standard editing, does not substantially alter the input or its meaning, or is authorised by law for the specified crime-related purposes. These are limited exceptions, not a blanket exemption for AI editing tools.
A transition to 2 December 2026 applies to providers of qualifying systems placed on the market before 2 August 2026. Under the Digital Omnibus on AI, those providers must take the necessary steps to comply with Article 50(2) by that December date. The transition concerns the provider marking duty; it does not postpone Article 50(4)’s deployer disclosures generally. Article 99 sets maximum fines for Article 50 breaches at €15 million or, for an undertaking, 3% of worldwide annual turnover for the preceding financial year, whichever is higher. These are ceilings, not automatic penalties. The regulation provides lower caps for SMEs and, following the 2026 amendment, small and mid-cap enterprises in this fine tier.
British organisations are not automatically outside the Act. Article 2 covers providers and deployers established or located in a third country where the AI system’s output is used in the Union, subject to the Act’s scope and exclusions.
For further guidance, contact the AI Act Service Desk.
