Applicable since 2 August 2026
Article 50 transparency duties apply from this date. Information must be clear, unambiguous, and accessible by the first interaction or exposure.
Free tool · EU AI Act
Do you need to label your AI content?
Answer a few questions about an image, video, audio clip, text, or chatbot. Get a traceable first assessment, a suitable label, and practical placement guidance.
No account and no upload. Your choices stay on your device. Legal-source baseline: August 2026.
Check AI disclosure in a few questions
Assess one specific item at a time. Content type, publication context, realistic appearance, and human control all matter.
What Article 50 actually requires
There is no blanket duty to visibly label every use of AI. The rules distinguish AI-system providers from deployers publishing specific kinds of content.
Disclose deepfakes visibly
This covers generated or manipulated images, video, and audio that resemble real or plausibly real people, places, objects, or events and could appear authentic.
Text only in defined cases
The duty covers published AI text on matters of public interest. Substantive human control plus assumed editorial responsibility creates an exception.
Machine-readable marking is a separate duty
Paragraph 2's machine-readable marking of synthetic output generally applies to providers of generative AI systems. A visible notice and metadata are not the same thing.
Frequently asked questions
Do I have to label every AI-written text?
No. Visible disclosure mainly concerns published AI text on matters of public interest where sufficient human review and editorial responsibility are absent.
Is a note in the imprint or profile bio enough?
No. The disclosure must be clear and unambiguous by the first exposure to the relevant content. A hidden general notice does not meet that goal.
Are the official EU icons mandatory?
No. The EU icons are voluntary and do not establish compliance on their own. Clear, understandable text can also be suitable.
What about creative or satirical works?
A deepfake in an obviously creative, artistic, satirical, or fictional work still requires disclosure, but it may be presented without impairing the display or enjoyment of the work.
Does AI colour correction require a label?
Usually not when it is only assistive standard editing and does not substantially change the input or its meaning. Generative replacement or realistic content creation may be different.