HomeAI ActTransparency obligations
Art. 50 AI Act Labelling AI: what has applied since August 2026.
The transparency obligations of the AI Act have applied since 2 August 2026. They concern chatbots, AI‑generated images, video and audio, and certain texts. Not everything created with AI has to be labelled. Here you can see who has to label what.
As of October 2026, checked against the text of the regulation. General information, not legal advice.
Four obligations in Art. 50
-
Chatbots and assistants (para. 1)
Anyone who interacts directly with an AI system must be told that it is AI, unless that is obvious. The obligation falls on the provider, who has to design the system accordingly.
-
Machine-readable marking (para. 2)
Generative AI must mark its image, video, audio and text output in a machine-readable way as artificially generated. Purely assistive functions in editing that do not substantially alter the content are excluded.
-
Emotion recognition and biometric categorisation (para. 3)
Deployers must inform the people concerned about the use.
-
Deepfakes and texts of public interest (para. 4)
Deployers must disclose that content has been artificially generated or manipulated. For art, satire or fiction, a notice that does not spoil the work is enough. For texts, the obligation lapses if a person reviews them and takes editorial responsibility.
The notice must be given clearly, unambiguously and accessibly at the latest at the first interaction or exposure (Art. 50(5)).
In practice What this means for companies.
Six cases that occur in many companies.
-
Chatbot on the website
It must identify itself as AI right at the start. If you had it developed specifically and operate it under your name, you bear this obligation yourself as provider.
-
Realistic AI images and videos
If they look like real people, places or events and could be taken for real, they are deepfakes (Art. 3(60)) and must be labelled. Recognisably artificial illustrations are generally not covered.
-
Advertising copy and product descriptions
The obligation for texts applies only to texts that inform the public on matters of public interest. Ordinary advertising and business texts are not covered by the wording.
-
Articles and press releases
If they deal with matters of public interest and are produced without human editorial control, they must be labelled. With review and clear responsibility, the obligation lapses.
-
Your own AI tools
If you offer a system that generates images or texts, it needs a machine-readable marking. Systems that were on the market before 2 August 2026 have until 2 December 2026 (Art. 111(4)).
-
Sentiment analysis in customer service
Customers must be informed about it (para. 3). Such systems also count as high-risk AI. For employees, emotion recognition is prohibited.
Typical cases for orientation. Whether an obligation applies depends on the individual case.
Get in touchHow to label
The Act prescribes no particular wording. The notice must be clear and unambiguous and appear at the latest at first contact. In practice short, visible sentences are usually enough:
- In the chat window: "You are writing with an AI assistant."
- Under an image or in a video: "Created with AI."
- For audio: a spoken notice at the start.
The machine-readable marking under para. 2 is a matter for the provider and is done technically, for example through metadata or watermarks. Other rules, for example from competition law, may require additional notices (Art. 50(6)).
Fines
Infringements of Art. 50 can be punished with fines of up to EUR 15 million or 3 % of worldwide annual turnover, whichever is higher. For SMEs the lower amount applies (Art. 99(4) and (6)).
How we support you
-
Labelling from the start
When we build chatbots or AI tools for you, we plan in the notice and the marking from the start. More on AI customer service, AI chatbots and process automation.
-
Clear rules in the team
In your AI policy we set out when your team labels AI content.
For borderline cases, for example in advertising or press, we recommend a legal review in each individual case.
Frequently asked questions about the labelling obligation
From when does the labelling obligation apply?
Since 2 August 2026. For generative AI systems that were already on the market earlier, machine-readable marking applies from 2 December 2026 (Art. 111(4)).
Do AI‑generated texts have to be labelled?
Only texts that are meant to inform the public on matters of public interest, and not even these if a person reviews them and takes editorial responsibility (Art. 50(4)).
Do AI images have to be labelled?
Realistic images, videos or audio content that can appear genuine must be disclosed as deepfakes. Independently of that, providers of generative AI mark their output in a machine-readable way.
Does the labelling obligation also apply to private individuals?
No. The obligations of deployers do not apply to people who use AI exclusively privately and not professionally (Art. 2(10)).
What penalty applies for missing labels?
Up to EUR 15 million or 3 % of worldwide annual turnover; for SMEs the lower amount applies (Art. 99(4) and (6)).
Would you like to know what you have to label?
Tell us where AI content is created in your company. We will reply personally, free of charge and without obligation.