01
Context: there is no blanket labelling duty for every AI-assisted task
The AI Act is Regulation (EU) 2024/1689. It follows a risk-based approach and, alongside rules for prohibited and high-risk systems, contains specific transparency obligations. Article 50 is particularly relevant to the identification of AI-generated or manipulated content.
Article 50 distinguishes between providers of AI systems and persons or organisations deploying them. Providers must design certain systems so that interactions and synthetic outputs can be identified. Deployers and publishers must visibly inform people in particular situations, including deepfakes and certain text concerning matters of public interest.
Not every research aid, spelling correction, translation or editorial assistance automatically triggers the same duty. The assessment depends on the circumstances. Voluntary disclosure may nevertheless be valuable because it sets expectations, makes editorial responsibility visible and can build trust beyond the legal minimum.
02
What Article 50 covers
The provision contains several distinct transparency obligations. For websites and editorial content, provider-side technical marking and deployer-side visible disclosure are especially relevant.
Article 50(1)
Direct interaction with AI
Providers of interactive AI systems must generally inform people that they are interacting with an AI system, unless that is already obvious in the circumstances.
Article 50(2)
Machine-readable marking
Providers of generative AI systems must generally mark outputs in a machine-readable format and make them technically detectable. This provider obligation is distinct from a visible notice placed on a website.
Article 50(3)
Emotion and biometric categorisation
Deployers of emotion-recognition or biometric-categorisation systems must generally inform exposed persons, unless a statutory exception applies.
Article 50(4)
Deepfakes and certain text
Deepfakes must be disclosed as artificially generated or manipulated. The same generally applies to AI-generated or manipulated text intended to inform the public on matters of public interest, unless a relevant exception applies.
03
What does this mean for articles and websites?
Editorial services benefit from a differentiated assessment. The AI Act does not treat every form of assistance identically, while broader voluntary transparency remains possible.
AI for research, structure or language editing
Using AI only for research support, ideation, spelling correction or as an editorial tool does not automatically create a visible labelling duty for the entire article. A voluntary statement can still make the actual workflow understandable.
AI-generated public-interest text
Where text is published to inform the public on matters of public interest, Article 50(4) may require disclosure. Under the regulation, the obligation does not apply where substantive human review or editorial control took place and a natural or legal person holds editorial responsibility.
AI-generated images and cover images
A fully AI-generated image is not automatically a deepfake in the legal sense. If it depicts persons, objects, places or events in a way that may falsely appear authentic, disclosure may be required. Voluntary image disclosure may still be appropriate outside those cases.
Deepfake audio and video
AI-generated or manipulated audio, image or video content constituting a deepfake must generally be disclosed. For evidently artistic, creative, satirical or fictional works, disclosure may be adapted so that it does not unduly hamper display or enjoyment of the work.
04
How disclosure should be presented
Article 50(5) requires information to be clear, distinguishable and accessible no later than the first interaction or exposure. Presentation should be appropriate to the medium and context.
Visibility
Clear and distinguishable
The notice should not be hidden in general terms. It should be recognisable as information about AI use and clearly connected to the relevant content.
Timing
At first exposure
People should receive the information no later than when they first interact with the system or encounter the affected content.
Accessibility
Accessible and understandable
Text, symbols and technical integrations should respect applicable accessibility requirements and remain understandable without specialist knowledge.
Technology
Visible and machine-readable
A visible notice informs people. Machine-readable metadata, provenance information and technical marks can supplement it, but do not always replace visible disclosure.
05
How to understand this standard
AI Usage Disclosure is a technical and editorial tool. It records AI use by content component: for example, text without AI, an AI-generated cover image, AI-assisted research and expert human review. One URL combines the readable explanation, structured table and machine-readable JSON-LD manifest.
The standard is intentionally broader than the minimum legal cases. Organisations can publish voluntary transparency for research assistance or generated illustrations that are not deepfakes. The software does not determine whether a legal obligation applies in an individual case.
- Component-level information for text, images, research, translation, audio, video and code
- Separate fields for AI contribution, purpose, human review and evidence basis
- A combined article view with narrative summary, table and one shareable link
- Machine-readable JSON-LD and self-hostable SVG badges
- No certification, legal advice or automated finding of legal compliance
06
Official sources and further information
Legal assessment should always be based on the current legal text, official guidelines and, where appropriate, qualified advice. The following sources are provided by institutions of the European Union.