Labelling AI content under the AI Act applies from 2 August 2026: tool providers must mark outputs in a machine-readable way, and whoever publishes a deepfake must disclose it. The Digital Omnibus adopted in 2026 did not postpone these obligations, apart from a short transition period for technical marking. Here's who has to do what, and what it means for an image or a spot.
What Article 50 of the AI Act says about labelling AI content
Article 50 of Regulation (EU) 2024/1689 (text on EUR-Lex) sets out the transparency obligations. For marketing, two paragraphs matter:
| Paragraph | Who | Obligation |
|---|---|---|
| Art. 50(2) | Providers of systems generating synthetic audio, images, video or text | Mark outputs in a machine-readable format, detectable as artificially generated or manipulated |
| Art. 50(4), first subparagraph | Deployers, meaning anyone using an AI system in a professional capacity, who generate or manipulate a deepfake | Disclose that the content has been artificially generated or manipulated |
| Art. 50(4), second subparagraph | Deployers publishing AI-generated text to inform the public on matters of public interest | Disclose it, unless there is human review with editorial responsibility |
| Art. 50(5) | All of the above | Clear and distinguishable information, at the latest at first exposure to the content |
By deepfake the AI Act means AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic (Article 3(60)). For evidently artistic, creative, satirical or fictional works, the obligation is limited to a disclosure that doesn't hamper the display or enjoyment of the work (AI Act Service Desk source).
When it applies, and what the Digital Omnibus changed
Article 50 obligations apply from 2 August 2026. The Digital Omnibus on AI, Regulation (EU) 2026/1744 published in the EU Official Journal in July 2026 (text on EUR-Lex), mainly postponed the rules on high-risk systems. On transparency it introduced one exception: providers of generative systems placed on the market before 2 August 2026 have until 2 December 2026 to comply with the Article 50(2) marking. The Commission's original proposal had a longer period, shortened in the final text (Council of the EU source).
The point for a brand: the duty to disclose deepfakes (Article 50(4)) was not postponed. The transition period covers tool providers' technical marking, not whoever publishes the content.
Fines for breaching Article 50 can reach €15 million or 3% of worldwide annual turnover, whichever is higher (Article 99 of the AI Act, EUR-Lex source).
The Commission's code of practice and guidelines
In June 2026 the Commission published the code of practice on marking and labelling AI-generated content (European Commission source). It's voluntary and has two parts: one for providers, on marking and detection, and one for deployers, on labelling deepfakes and certain public-interest text. The legal obligations stay binding whether or not you sign it (Commission FAQ). The Commission has also published guidelines on Article 50 that clarify its scope (AI Act Service Desk resources).
"A clear label doesn't weaken a campaign. It removes the doubt."
What it means for a brand's images and spots
The practical question is: does my campaign contain a deepfake? Some typical cases, with the cautious choice:
- A real person recreated with AI (testimonial, founder, digital twin). The clearest case: it must be disclosed, on top of requiring the person's consent.
- Invented avatars or models in photorealistic scenes. They don't resemble an existing person, but the scene can look like a real shoot. Cautious choice: disclose it.
- A real product in a recreated real place. It can fall within the definition, since it concerns existing objects and places. Cautious choice: disclose it.
- Clearly fictional visuals (a flying product, surreal worlds, an illustrated style). The rule for creative works allows a light disclosure that doesn't spoil the content.
- Assisted retouching (clean-up, colour correction). The AI Act exempts from the marking obligation standard editing functions that don't substantially alter the content.
How to label in practice: a short, readable line such as "Image generated with AI" or "Video made with artificial intelligence", visible from first exposure: as an overlay at the start of a video, in a post caption, next to the image on a page. Keep the files' provenance metadata (such as C2PA Content Credentials) instead of stripping it on export.
Meta and TikTok labels
Platforms have their own rules, on top of the law. TikTok requires creators to label AI content with realistic images, audio or video, and automatically labels content carrying C2PA Content Credentials (TikTok source). Meta shows an "AI info" label on ads created or significantly edited with generative AI tools, its own or third-party, in the "About this ad" section and sometimes next to "Sponsored" (Meta source).
The platform label doesn't replace yours: it's a separate system that may not trigger. Build the disclosure into the creative itself.
How we handle it in our projects
At DigitalX Studio we label AI content where the AI Act requires it and agree it with the client in the brief: where the line goes, in which formats, in which languages. Digital avatars are invented people or consensual digital twins; in AI UGC video ads disclosure is part of the format, not a last-minute add-on. For rights and copyright see AI image copyright.
Frequently asked questions
When do you have to label AI-generated content in the EU?
From 2 August 2026, when Article 50 of the AI Act applies. The Digital Omnibus gave only tool providers already on the market until 2 December 2026 for technical marking.
Does a brand have to disclose that an ad was made with AI?
Yes, if the content is a deepfake: images, audio or video that resemble real people, objects, places or events and could appear authentic. For clearly creative content, a disclosure that doesn't hamper enjoyment is enough.
Did the Digital Omnibus postpone Article 50?
No. It mainly postponed the high-risk rules. For Article 50 there is only a transition period until 2 December 2026 for marking by generative systems already on the market.
How do you label an AI video on Instagram or TikTok?
With a readable line in the video or caption, visible at first exposure, plus the platform label. TikTok requires labels on realistic AI content; Meta shows an "AI info" label on ads made or edited with generative AI.
What are the fines for not labelling AI content?
For breaches of Article 50 the AI Act provides fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher.