EU AI icons have a second layer

The EU basic icon can carry a second interactive layer: more information behind a mark the viewer can open. That is not a burned-in "AI generated" line on TikTok, and it is not YouTube's Shorts overlay. The Commission designed a two-layer object for deployers who want a small first mark plus details on request. Most social players will not host that second layer for you. Plan the first exposure as if the extra layer never opens.

Article 50 has applied since 2 August 2026. Guidelines 20 July 2026. Fines can reach €15 million or 3%. The icons and the second-layer note live on EU icons for labelling AI-generated content. Using the icons is optional. Labelling is not.

This page is the interactive layer. Generated versus modified is EU icons: generated vs modified. Paste copy is the Article 50 disclosure template.

What the Commission published

Three icons: fully AI-generated, partially AI-modified, and a basic icon. The basic icon is for when AI was involved in creating a deepfake or published text, especially when a custom text label or an interactive second layer is implemented. Commission example at the first layer: a deepfake with the text "voices generated with" plus the basic icon.

User-testing improved when the basic icon sat with words. Pair it. A silent mark is the weak version of this system.

On the second layer, the Commission's display summary says: if additional information is provided through a second interactive layer, the icon should clearly indicate that further information is available, and the second-layer content shall be navigable using assistive technologies. The first layer still has to be clearly perceivable and distinguishable at first exposure, placed where no intervening overlay elements exist, and (except for creative works) directly embedded unless an equivalent user-interface overlay is available. The icon must remain visible when content is reshared or downloaded.

Read that as two jobs:

First layer: a person knows, immediately, that this is generated or manipulated. Words plus icon. No extra tap required for that fact.
Second layer: optional detail — which part is synthetic, which model family, a link to a method note — behind a control that announces itself and works with a screen reader.

The second layer does not postpone the first. A tap-to-learn-it-was-AI is the failure mode Article 50(5) is written against: information at the latest at first interaction or exposure. If the only honest sentence is behind the tap, you did not ship a first layer. You shipped a FAQ.

Signatories of the Code of Practice have fuller placement specs in section 2 of that code. Non-signatories may still use the icons. Using them is not a claim that you signed. The icon does not, by itself, establish legal compliance. You are usually the deployer of the file you publish.

Why TikTok's burned line is a different object

TikTok wants a realistic AIGC label through the built-in control in the posting flow. It can also auto-apply from detection and from Content Credentials. Creators often add a burned "AI generated" line as house style or as an Article 50 first-exposure sentence.

That burned line is a single-layer, in-picture, always-on object. It travels with the file when someone screen-records. It does not open. It does not carry a method note. It is not the Commission's interactive second layer. It can still be a good first-exposure line.

What it is not:

Not a substitute for TikTok's built-in toggle. The Creativity Program rule is the native label. A burned sentence is extra.
Not a YouTube Shorts overlay. YouTube draws its own "AI" mark on photoreal Shorts. Your burned line will sit next to that overlay and can be covered by it.
Not an EU second layer. There is no Commission-standard accordion inside TikTok's player that your icon can open.

If you need more detail than "this video is AI-generated," put a short true sentence on the first layer ("The voice in this video is AI-generated") and put the essay on a landing page, a description field, or a site you control. Description fields are easy to miss on a feed. Do not make them the only layer.

YouTube's below-player row on long-form is closer to a platform-hosted second glance — still not the EU interactive layer, still not your Article 50 first exposure if the Act applies. Meta's AI Info hub is Meta's object. Do not call any of these "the Commission second layer."

How to ship a two-layer disclose in a real pipeline

On a site or a player you control. You can actually build the second layer: icon plus "more info," an accessible panel, alt text / ARIA, keyboard navigation. First layer still has the plain sentence. Second layer can name generated versus modified, voice versus picture, and a contact. Download the Commission files. Do not redraw them.

On TikTok, Reels, Shorts. You do not control the player chrome. Ship a complete first layer in the file: official icon if you use one, plus words, placed where chrome and the YouTube overlay will not cover it. Set the native AI content label. Use the description for optional detail, not for the only disclosure. Check a phone.

On a paid unit. First exposure is the ad. Interactive layers that only exist on the brand site do not help the person who never left the feed. Burn the line. If NY §396-b also applies, conspicuous disclose in the ad is its own test.

Artistic work. The Act allows a quieter disclose that does not hamper enjoyment. A second layer in the credits can be that quieter shape when the work is evidently a work. Ads and UGC testimonials are not that shape.

Provider marks (SynthID, C2PA) are not a second layer. They are detector-facing. A viewer does not open them. Grace to 2 December 2026 is the provider marking calendar, not a reason to hide the first layer.

A UGC video generator will not implement the Commission's accordion. Veo 3.1 will not either. If you want the first-layer line on the same timeline as the take, sign up and burn a complete sentence. Save the interactive panel for destinations that can actually open one.

Accessibility is part of the Commission's summary, not a flourish: visible size, plain language, avoid jargon except "AI," readable by assistive tech, on-screen long enough to be understood if it is timed. A 12-frame flash with a "tap for more" that does not work in the For You player is the opposite of that summary.

Default for 2026 social

Assume the second layer will not open. Make the first layer enough. Use the basic icon plus words when the file is a mixed or voice-only deepfake. Use fully-generated or modified when that is the true radio. Put extra method notes on a page you own.

Do not tell a client "we used the EU interactive icon" when you shipped a TikTok with a sparkle. Those are different objects. The sparkle is TikTok. The second layer is a Commission pattern for a player that can host it.

First exposure is still the first frame.

FAQ

Can the second layer be the only disclosure?

No. The icon has to indicate that more information is available, which means the first layer already has to work as a disclosure plus a pointer. A buried panel is not first exposure.

Does TikTok's "AI-generated" badge count as the EU second layer?

No. It is TikTok's native label. It can help a person perceive that the clip is AI. It is not the Commission's interactive layer, and it is not automatically Article 50-compliant wording or placement.

Do I have to build a second layer to use the basic icon?

No. The basic icon is also for pairing with a custom text label. Words plus icon on the first layer is the tested pattern. The second layer is additional information when you have a player that can host it.