Last reviewed:
What is AI-generated content labelling? Definition and Article 50 obligations
AI-generated content labelling refers to the signals, machine-readable (invisible watermark, metadata) or human-readable (label, notice), showing that an image, video, audio or text was produced or altered by AI. Article 50 of the AI Act has required it since 2 August 2026: providers technically mark the outputs of their generative systems, and organisations publishing deepfakes or certain public-interest texts must disclose it.
Article 50 of the AI Act (Regulation (EU) 2024/1689) splits the obligation between two actors. The provider of a system generating synthetic content (OpenAI, Google, Mistral AI, a software vendor) must mark its outputs in a machine-readable, detectable format (paragraph 2). The deployer, meaning the organisation using the tool, must disclose deepfakes (image, audio or video resembling real people, places or events that could appear authentic) as well as texts published to inform the public on matters of public interest, unless reviewed by a human under editorial responsibility (paragraph 4). For evidently artistic, satirical or fictional works, the disclosure can be discreet. These rules have applied since 2 August 2026. The Digital Omnibus (Regulation 2026/1744) gives only providers of systems already on the market before 2 August 2026 until 2 December 2026 for technical marking. On 10 June 2026 the Commission published a Code of Practice on marking and labelling AI-generated content, voluntary to sign, which it considers an adequate tool to demonstrate compliance. It sets out a layered approach (metadata, watermarking, fingerprinting or logging) and an EU icon for visible labels. Commission guidelines on Article 50, adopted in summer 2026, clarify which cases are covered. On the technical side, the open C2PA standard (Content Credentials), whose steering committee includes Adobe, Google, Microsoft, Meta and OpenAI, provides a provenance metadata format.
Concrete example
Illustrative case. A fashion retailer in Bordeaux generates the images for its product pages with an image tool. Those images carry a marking applied by the provider: the retailer only has to avoid stripping it, for example through exports that remove metadata. For its campaign abroad, it also publishes a video in which its founder's voice, cloned by AI, presents the collection in English. The video imitates a real person: it is a deepfake, to be disclosed visibly at first exposure. The style advice blog, drafted with AI then reviewed and signed by the editorial manager, needs no notice.
Comparison
| Situation | Responsible party | Requirement | Since |
|---|---|---|---|
| Chatbot, conversational assistant | Provider | Inform people they are interacting with AI | 2 August 2026 |
| Generated image, audio, video or text | Provider of the generative system | Machine-readable, detectable marking | 2 August 2026; 2 December 2026 for systems already on the market |
| Deepfake | Deployer publishing it | Visible notice that the content is generated or manipulated | 2 August 2026 |
| Public-interest text published without editorial control | Deployer publishing it | Notice that the text is AI-generated | 2 August 2026 |
| Emotion recognition, biometric categorisation | Deployer | Inform the people exposed | 2 August 2026 |
FAQ
What does Article 50 of the AI Act say?
It sets transparency obligations. Chatbots must disclose that they are AI. Providers of generative systems must mark their outputs in a machine-readable way. Organisations using emotion recognition or biometric categorisation must inform the people concerned. Those publishing deepfakes or public-interest texts generated without editorial control must disclose it. These rules have applied since 2 August 2026.
What is the Code of Practice on marking and labelling?
A document published by the Commission on 10 June 2026, drawn up with providers, publishers and experts. Signing is voluntary, but the Commission considers it an adequate way to demonstrate compliance with Article 50. It has one section for providers (marking and detection) and one for deployers (labelling deepfakes and certain texts).
How does AI watermarking work?
A watermark is an invisible signal embedded in an image, audio or video at generation time, which detection tools can recognise. It is usually combined with provenance metadata (C2PA format) and sometimes with a fingerprint logged by the provider. No single technique is foolproof: metadata is easily lost on export, hence the recommended layered approach.
Do I need to disclose a text written with AI?
In most cases, no. The obligation covers texts published to inform the public on matters of public interest, and it falls away if the text was reviewed by a human and published under the editorial responsibility of a person or organisation. A reviewed sales email or product page is not covered.
Which dates matter?
2 August 2026 for Article 50 as a whole. 2 December 2026 for technical marking by generative systems already on the market before 2 August 2026, a delay granted by the Digital Omnibus. Content generated before 2 August 2026 does not need to be marked retroactively.
What is C2PA?
The Coalition for Content Provenance and Authenticity publishes an open standard that attaches verifiable information about a file's origin and edits, called Content Credentials. Its steering committee includes Adobe, Amazon, the BBC, Google, Meta, Microsoft, OpenAI and Sony.
See also
Further reading
C2PA (Coalition for Content Provenance and Authenticity), open content provenance standard
Sources
- Regulation (EU) 2024/1689 on artificial intelligence (AI Act), Article 50. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), marking deadline of 2 December 2026. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32026R1744
- Commission publishes Code of Practice on marking and labelling AI-generated content, European Commission, 10 June 2026. https://digital-strategy.ec.europa.eu/en/news/commission-publishes-code-practice-marking-and-labelling-ai-generated-content
- Guidelines on transparency obligations for providers and deployers of certain AI systems, European Commission, 2026. https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations