Labelling AI-generated content: who marks, who labels, and by when
The AI Act has two duties about generated content, and they fall on different people. The company whose system generates the content has to mark it in a machine-readable way. The company that publishes a deep fake has to say so. Most confusion comes from mixing the two.
Two duties, two people
- Providers mark. Under Article 50(2), the provider of a system that generates synthetic audio, image, video or text has to make its outputs machine-readable as artificially generated or manipulated.
- Deployers label. Under Article 50(4), the one who uses such a system to make a deep fake, or to publish text on matters of public interest, has to disclose it to the people who see it.
The Commission's FAQ adds the point that trips people up: a deployer cannot rely on the provider's machine-readable mark to meet its own duty to disclose. The mark is for machines; the label is for people.
Marking: what providers owe
Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.
The technical solution has to be effective, interoperable, robust and reliable as far as technically feasible. The duty does not apply where the system only performs an assistive function for standard editing, or does not substantially alter the input data or its meaning.
The Commission's FAQ also lists outputs outside Article 50(2): short sequences of numbers, symbols or letters, source code, machine-to-machine outputs and closed-loop industrial outputs.
The dates
- 2 August 2026. Article 50 applies, under the AI Act's general date of application in Article 113.
- 2 December 2026. Providers whose generative systems were placed on the market before 2 August 2026 have until this date to comply with Article 50(2). The deadline was added by Regulation (EU) 2026/1744, as a new Article 111(4), in force since 27 July 2026.
- No look back. The Commission's FAQ says content generated before 2 August 2026 does not have to be labelled after the fact.
The four-month grace period covers the marking duty only. The deployer's duty to disclose deep fakes has applied since 2 August 2026 without one.
Deep fakes: what deployers owe
'deep fake' means AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful;
A deployer whose AI system generates or manipulates such content has to disclose that it was artificially generated or manipulated. The Commission's FAQ reads the definition as three conditions that must all be met: the content resembles something, that something exists, and it would falsely appear authentic.
So a clearly drawn illustration of an invented product is not a deep fake. A realistic photo of your real shop with customers who were never there may be one.
For an evidently artistic, creative, satirical or fictional work, the duty shrinks to disclosing that generated content is there, in a way that does not spoil the work.
Generated text
Text is covered more narrowly than people think. The second subparagraph of Article 50(4) applies to generated or manipulated text that is published with the purpose of informing the public on matters of public interest.
It falls away when the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication. A newsroom that edits and signs off AI drafts is in that exception; a site publishing machine-written news with nobody responsible is not.
The code of practice
The Commission published a final Code of Practice on transparency of AI-generated content on 10 June 2026. It has a section for providers on marking and detection, and one for deployers on labelling. Signing it is voluntary; the Article 50 duties are not.
On 8 July 2026 the Commission concluded that the code adequately covers Article 50(2), (4) and (5). Its opinion adds that adherence to the code does not constitute conclusive evidence of compliance.
The Commission also adopted guidelines on Article 50 on 20 July 2026. The FAQ that summarises them is the quickest way in.
In practice, for a business
- Make a list of where AI generates what you publish: product photos, ad images, voice-overs, videos, articles.
- Ask each tool's maker how its output is marked, and keep the answer. If you build your own generator, the marking duty is yours.
- Check that your export or upload step does not strip the mark from the file.
- Label anything realistic that shows real people, places or events as generated.
- Keep the list current: a new tool is a new line on it.
Our marking check samples up to five images on a public page and looks for C2PA and other provenance metadata. Finding it shows the image carries a mark. Not finding it on an ordinary photo is not a violation, and the check says so: whether an image is AI-generated is not something its bytes alone can prove.
Who enforces it
National market surveillance authorities enforce most of it, with the AI Office for systems built on a general-purpose AI model by the same provider and for very large online platforms and search engines. The fine for breaking Article 50 is the same as for chatbots: up to EUR 15 000 000 or 3% of total worldwide annual turnover, whichever is higher, and the lower of the two for SMEs.
Questions people ask
- Do I have to label AI-generated images?
- If you are the provider of the system that generates them, they must be marked in a machine-readable way under Article 50(2). If you publish a realistic image that resembles real people, places or events and would falsely appear authentic, you must disclose it is generated under Article 50(4).
- When does machine-readable marking become mandatory?
- Article 50 applies from 2 August 2026. Providers of generative systems placed on the market before that date have until 2 December 2026 under Article 111(4), added by Regulation (EU) 2026/1744.
- Does AI-written marketing text need a label?
- Article 50(4) covers text published to inform the public on matters of public interest, and not when a person reviewed it and someone holds editorial responsibility. Product copy is not obviously in that category; ask a lawyer about your case.
- Is signing the code of practice enough to comply?
- No. The Commission found the code adequate for Article 50(2), (4) and (5), but its opinion says adherence does not constitute conclusive evidence of compliance.
Sources
- Regulation (EU) 2024/1689 (AI Act), Articles 3, 50, 99 and 113, EUR-Lex
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), new Article 111(4), EUR-Lex
- European Commission, FAQ on transparency obligations under Article 50
- European Commission, Code of Practice on transparency of AI-generated content
- European Commission, opinion on the adequacy of the code of practice, 9 July 2026
- European Commission, Guidelines on transparency obligations for providers and deployers of AI systems
All sources read again on November 5, 2026. This guide explains the text of the law and the Commission's published guidance. It is not legal advice, and a lawyer should look at your own case.
