AI labelling obligation under Art. 50
Since 2 August 2026, the transparency obligations of the EU AI Act (Art. 50) have applied. For most companies, three cases matter: chatbots and voice assistants that identify themselves as AI, deepfakes that are disclosed as such, and AI-generated text on matters of public interest.
What does Art. 50 of the AI Act regulate?
Art. 50 divides the obligations between two roles: providers, who develop an AI system or put it into service under their own name, and deployers, who use it in a professional capacity. The five paragraphs at a glance:
- Paragraph 1, providers: AI systems that interact with people must be designed so that people know they are talking to an AI, unless this is obvious.
- Paragraph 2, providers: AI-generated or manipulated images, videos, audio and text must be marked in a machine-readable format and be detectable as artificially generated.
- Paragraph 3, deployers: Anyone who uses emotion recognition or biometric categorisation informs the persons concerned.
- Paragraph 4, deployers: Deepfakes, and AI-generated text intended to inform the public on matters of public interest, are disclosed as AI-generated. For text, this does not apply if a person reviews it editorially and takes responsibility for it.
- Paragraph 5, all: The notice must be clear and distinguishable, given at the latest at the first interaction or exposure, and accessible.
The Digital Omnibus (Regulation (EU) 2026/1744) has not changed these obligations. Only paragraph 7, on the Commission’s codes of practice, was reworded.
When does the labelling obligation apply?
Art. 50 has applied since 2 August 2026. For providers of generative AI systems placed on the market before that date, a transitional period until 2 December 2026 applies to the machine-readable marking under paragraph 2 (Art. 111(4)). There is no transitional period for deployers: a deepfake in advertising has had to be disclosed since August. According to the Commission, content created before 2 August 2026 does not have to be labelled retrospectively. All dates are on the deadlines page.
What does this mean for typical cases in a company?
| Case | What applies | Who is responsible |
|---|---|---|
| Chatbot on the website | a notice that an AI system is responding, unless this is obvious | the provider of the chatbot; anyone who has a bot developed under their own name can be a provider themselves |
| AI voice assistant on the phone | a notice that an AI system is speaking, ideally as an announcement at the start of the call | the provider of the assistant |
| Realistic AI image or video of a real person in advertising | deepfake: disclose that the content is AI-generated | your company as deployer |
| AI illustration that is recognisably not real | no disclosure under Art. 50(4) | machine-readable marking by the provider of the tool |
| AI-generated text on a matter of public interest | disclose, unless a person reviews it editorially and takes responsibility for it | your company as deployer |
| ChatGPT used only internally | as a rule no obligation under Art. 50, but AI literacy under Art. 4 | your company as deployer |
This classification is based on the Commission guidelines of 20 July 2026 and its questions and answers on Art. 50. You should have borderline cases, such as whether an advertising text concerns a matter of public interest, assessed individually.
How do you label correctly?
The Regulation does not prescribe a particular form, but it does require a clear notice at the right time. The following have proven effective:
- In the chat window, a visible sentence such as “You are chatting with an AI assistant” before the first reply appears.
- On the phone, an announcement at the start, as recommended in the Commission guidelines.
- For images and videos, a notice directly on the content, not just in the legal notice. For artistic or satirical works, a notice that does not interfere with the work is sufficient.
- An internal rule on who reviews content before publication and when a notice is needed. It belongs in your AI policy.
The Commission’s Code of Practice on labelling AI-generated content has been available since 10 June 2026 and offers optional symbols. It is voluntary.
What are the penalties for infringements?
Infringements of Art. 50 can be punished with fines of up to €15 million or 3 % of global annual turnover (Art. 99(4)). For small and medium-sized enterprises, the lower of the two amounts applies. The competent authorities are the national market surveillance authorities. More on the fine ranges on the fines page.
Frequently asked questions about AI labelling
Do I have to label every AI-generated image?
No. For deployers, the disclosure obligation applies to deepfakes, i.e. images, videos or audio recordings that resemble real persons, objects, places or events and would falsely appear to be authentic (Art. 3(60), Art. 50(4)). A recognisably drawn illustration is not a deepfake. Machine-readable marking of the files is a matter for the providers of the AI tools.
Does a blog article written with AI have to be labelled?
Only if the text is published to inform the public on matters of public interest and no person has reviewed it editorially and taken responsibility for it (Art. 50(4)). According to the Commission, a mere spelling check does not count as a review. Anyone who checks the content of texts before publication and takes responsibility for them does not have to label them under Art. 50.
Does Art. 50 also apply if we only use ChatGPT internally?
As a rule, no. The obligations concern people who interact with an AI system or are exposed to it, as well as published content. For internal use, however, the AI literacy obligation under Art. 4 applies, see AI literacy obligation.
Do I need the EU symbol for AI content?
No. The Commission’s Code of Practice on labelling AI-generated content is voluntary, and the symbols in it are merely an option. Anyone who does not sign the Code must show in another way that they meet the obligations. What matters is a clear notice at the point where people see the content.
Train your team now The EU AI Act at a glance
Note: this article is for general information only and does not constitute legal advice.