EU AI Act · Article 50 in force · marking deadline 2 December 2026
EU AI Act · Article 50(4) · Official icons

The official EU AI-content icons

From the Code of Practice, 10 June 2026Free to use, hash-verifiedLast reviewed July 2026

The European Commission's Code of Practice on Transparency of AI-Generated Content (10 June 2026) publishes a standard set of icons for labeling AI involvement in published content, a visual complement to Article 50(4)'s requirement that deepfakes and AI-generated content shown to the public be labeled. They're free to use, not trademarked against you, and using the recognized standard icon is a stronger signal than an improvised badge of your own.

Icons are for visible labeling (Art. 50(4)), not machine-readable marking (Art. 50(2)). These icons are something a human reader sees. They don't substitute for the embedded, machine-readable C2PA/provenance metadata Article 50(2) separately requires of generative-system providers.

The three icons

Official EU 'AI' icon

AI

General AI-involvement marker, for content where AI played a role, without specifying purely generated vs. edited.

Official EU 'AI GENERATED' icon

AI GENERATED

For content created entirely by an AI system: images, video, audio, or text with no substantive human-authored starting point.

Official EU 'AI MODIFIED' icon

AI MODIFIED

For real content that's been meaningfully edited or altered by AI, the deepfake-adjacent case where a genuine photo/video/audio has been AI-manipulated.

The black file downloads as LABEL_AI MOFIFIED_black.svg. That misspelling is in the Commission's own archive, and we keep the filename exactly as published so the file stays byte-identical to the official release. Rename it freely on your side.

Each icon ships in black and white, and in solid or transparent-background versions; pick whichever pair reads clearly against your own design. Files above are served directly from our copy of the Commission's official release; see the hash-verification note below.

When Article 50(4) calls for a label

Article 50(4) requires visible labeling in two situations: deepfakes (AI-generated or manipulated image, audio, or video content that resembles existing people, places, or events and would falsely appear authentic) and AI-generated or AI-manipulated text published to inform the public on matters of public interest, unless the text has undergone human review and a natural or legal person holds editorial responsibility for it. Routine marketing images or an AI-drafted internal memo aren't the target; content that could pass as real, shown publicly, is.

How to use them correctly

Source and integrity. These files are the Commission's own official release (Code of Practice on Transparency of AI-Generated Content, 10 June 2026, Annex I), downloaded 6 July 2026 and served here unmodified, byte-for-byte, with SHA-256 hashes kept on file. Always fine to fetch directly from the European Commission's Digital Strategy site instead if you'd rather source them yourself.

Where the label goes, format by format

Article 50(5) sets the placement standard for every one of these labels: the information has to reach people at the latest at the time of first interaction or exposure, in a clear and distinguishable manner. The Commission's guidelines of 20 July 2026 spell out the floor: a mention in the terms and conditions, a bot name, or a robot icon alone is normally not enough. What counts as first exposure differs by format, so where the icon goes differs too.

Images

First exposure is the first render, wherever that happens. A corner overlay on the image itself, or a caption immediately adjacent, is perceivable at that moment; a note further down the page is not. And images travel without their page: social cards, embeds, image-search results. An on-image overlay is the only visible label that travels with them, which is what the transparent-background variants above are for.

Video

First exposure is the opening seconds of playback, so the label belongs on screen when playback starts, kept up long enough to read. A line in the description below the player only reaches viewers who scroll, which is not something Article 50(5) lets you assume. For content likely to be clipped and reshared, a small persistent overlay survives the clip; a one-time intro card doesn't.

Audio

The icons are visual and audio's first exposure isn't, so the icon can't carry this format alone. An audible announcement at the start of playback covers the exposure moment itself; the icon then labels the visible surfaces around it: the embedded player, the episode page, the cover artwork. The two reach different listeners, so treat them as a pair, not alternatives.

Text

For AI-generated text in scope of Article 50(4), published to inform the public on matters of public interest, the label belongs where reading starts: at the top of the article, by the headline or byline. A footnote after the body discloses at last exposure, not first. And recall the carve-out described above: where the text has undergone human review and a natural or legal person holds editorial responsibility for it, the 50(4) text duty doesn't apply in the first place.

Common questions

Do I have to use these exact icons?

No. Article 50(4) requires the disclosure, not this particular artwork: a clear text label can satisfy the visible layer on its own. The icons come from the Commission's Code of Practice on Transparency of AI-Generated Content (10 June 2026) as the standard set, and the recognized standard is a stronger, more legible signal than a badge of your own design. The artwork is optional; making the AI involvement perceivable is not.

If I add the icon, is the machine-readable side covered too?

No. The icon is pixels; it carries no metadata. Article 50(2) separately requires providers of generative systems to mark output in a machine-readable format, which in practice means C2PA Content Credentials or IPTC provenance metadata embedded in the file itself. Two things follow: adding an icon does nothing for Article 50(2), and it's worth checking the published files even if your generator embeds marking at creation time, because image pipelines and CDNs often strip metadata when they resize or re-encode.

Is there a grace period before any of this applies?

No general one. The transparency obligations of Article 50 (Regulation (EU) 2024/1689) apply since 2 August 2026. The single deferral is narrow: the Article 50(2) machine-readable marking duty, for generative systems placed on the market before 2 August 2026, applies from 2 December 2026. Chatbot disclosure under 50(1) and the visible labels this page covers under 50(4) already apply. What the 2 December 2026 date does and doesn't cover →

Didn't the Digital Omnibus delay the AI Act?

Part of it. The Digital Omnibus (Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026, in force 27 July 2026) pushed back the high-risk regime's deadlines. It left Article 50 untouched, so the transparency duties, including the labeling these icons exist for, continue to apply since 2 August 2026. The full Article 50 guide →

What can a missing label cost?

Article 99(4)(g) sets the ceiling for transparency violations: fines up to €15 million or 3% of total worldwide annual turnover, whichever is higher. For SMEs and start-ups, the lower of the two applies (Art. 99(6)). Those are maximums; what a given case actually draws is decided by the enforcing authority within that frame.

Using these icons is the visible half of Article 50(4). Check the rest. DisclosureProof scans your site for both halves (visible labels and machine-readable marking), free. Run the free scan →