Pick the Article 50 icon before publish
Choose generated, modified, or the basic EU icon before you publish. Paste lines live in the disclosure template.
Pick the Article 50 icon before you publish. Fully AI-generated, partially AI-modified, or the basic icon with a short text label: those are the three official marks. The clip is already made. The radio left is which icon (and which paste line) is true. Publishing and then hunting for a label in the caption field is how first exposure comes and goes unmarked.
Article 50 of the EU AI Act has applied since 2 August 2026. Commission guidelines were adopted 20 July 2026. Deployers who publish deepfakes — content resembling real people, places or events that would appear authentic — must put a visible or audible label at first exposure. Fines can reach €15 million or 3% of worldwide turnover. This page is not the law. It is the weekly pick. The copy-paste labels and placement notes live in the Article 50 disclosure template. The icons themselves are on the Commission page: EU icons for labelling AI-generated content.
Three icons, one truth per file
Download the Commission files. Do not redraw them. Four visual variations exist for each icon: black, white, black at 50 percent transparency, white at 50 percent transparency. Using the icons is optional. Labelling is not. The icon does not, by itself, establish legal compliance.
Fully AI-generated. The entire deepfake (or the public-interest text) is generated by AI, with no human-created elements and no human editorial control beyond prompting. A talking head that was never in a room. A news summary nobody edited. A Veo take of a politician who was not filmed. If that is this week’s file, this is the icon, plus a line like “This video is AI-generated.”
Partially AI-modified. Pre-existing, human-made content was partially modified with AI and the result is still a deepfake or public-interest text. A real apartment furnished with generated furniture. A real face swapped. A real street with a generated extra. If you shot the room and Omni-edited the sofa, you are probably here, not on fully-generated.
Basic icon. Use when AI was involved in creating a deepfake or published text, especially when you pair the mark with a custom text label or an interactive second layer. Commission example: a deepfake with the text “voices generated with” plus the basic icon. User-testing improved when the basic icon sat with words. Pair it. Do not ship the mark alone and call it done.
Standard assistive editing is exempt: colour, crop, denoise, captions on a real recording. A colour-graded interview of a real person is not a deepfake because you used a timeline. A photoreal person who was never filmed is not exempt because you added captions.
Artistic, creative, satirical, or fictional work may label in a way that does not spoil the work. The load-bearing word is “evidently.” A UGC-style ad built to look like a real customer is a poor fit for a quiet end card.
Provider marking (machine-readable C2PA / watermarks) is a different artefact, with a grace period to 2 December 2026 for systems already on the market before 2 August 2026. If you only prompt a hosted generator and publish, you are still the deployer of the label the viewer sees. A provenance tick is not your on-screen icon.
Pick before upload, not in the caption box
First exposure means the viewer does not have to expand a description, wait until the last second, or catch a two-frame flash. Burn the paste line into the file, or speak it on the first seconds, and use the platform toggle. Platforms re-encode. Manifests die. Pixels travel.
This week, before anyone hits publish:
- Is the file a realistic person, place, or event that would appear authentic to an EU viewer? If no — clearly a cartoon, clearly a game, clearly not a person — you may still disclose as a house style, but the deepfake radio may not be the Act’s. If yes, continue.
- Was the whole picture generated, or was a real recording modified? Generated → fully-generated icon + line. Modified → modified icon + line. Voice-only synthetic on real picture → basic icon plus “voices generated with AI” is the shape the Commission already used.
- Burn the line. Place the icon where no chrome covers it. Leave it up long enough to read.
- Then set TikTok’s AI label, YouTube’s altered-content disclosure, Meta’s self-disclosure. Those are extra. They are not the icon pick.
Content generated and published before 2 August 2026 is not retroactive. Generated before and published after is covered. If this week you are sitting on July Veo takes, the publication date is the one that counts.
Personal, non-professional use sits outside the Act. Regular economic benefit — a paid creator, a brand, an agency — is professional use. If you are reading this as a production checklist, assume you are the deployer.
Do not invent a fourth icon. Do not ship a custom SVG and call it the EU mark. Do not use a platform AI badge as a substitute. The Commission published three.
What the radio is not
It is not “pick the quietest icon so the CTR holds.” The label is not a ranking penalty on TikTok, YouTube, or Meta; unoriginal templated content is. Pick the true icon. Then worry about whether the clip is actually yours.
It is not the TikTok Creativity Program toggle. That toggle keeps you eligible. Article 50 is a different test: audience and content, not the app.
It is not a licence to skip advertising disclosure or likeness releases. A labelled fake testimonial is still a fake testimonial.
It is not a caption job. Auto-captions will not transcribe a legal overlay you forgot to add. The paste line is text overlay. The speech transcript is separate.
Veo 3.1 does not label the output for you. The model page is not a disclosure. You publish; you pick the icon.
The Monday publish list
Every file going out this week to an audience that includes the EU:
| File | True state | Icon + line |
|---|---|---|
| Fully generated photoreal person | No camera original | Fully AI-generated + “This video is AI-generated.” |
| Real shoot, generated extra / restyle | Human plate, AI change | Partially AI-modified + “This video was modified with AI.” |
| Real picture, synthetic voice | Voice is the deepfake-shaped part | Basic icon + “The voice in this video is AI-generated.” |
| Real interview, captions and grade | Assistive editing | No Article 50 deepfake icon required for that reason |
If you need the overlay step in the same studio that made the clip, sign up and treat the icon pick as picture lock, not as an afterthought at the scheduler. First exposure is the first frame.
FAQ
Do I have to use the EU icons, or is the paste line enough?
Labelling is required for covered content; the icons are optional. User-testing favoured icon plus plain words. Use the Commission files if you use an icon. Do not redraw. The template has the sentences.
We only post to US accounts. Skip the radio?
If EU viewers will still see the clip, the Act follows the audience, not your HQ. If you are sure they will not, you still have platform AI labels. This radio is the EU icon/line. Do not skip platform toggles because you skipped the icon.
Can I put the icon only in the description?
Not if that means it is missed at first exposure. Mobile descriptions are a tap away. Burn the line or speak it. Description is extra, not the mark.