AI Labelling Obligation: What Companies Must Do Now
By Dennis Weidner · 12. November 2026
The AI labelling obligation is set out in Article 50 of the EU AI Act and has applied since 2 August 2026. It requires three things: people must be told when they are talking to an AI; AI-generated images, video, audio and text must be marked in a machine-readable way; and deployers who generate or manipulate deepfakes with AI must disclose them. For generative systems already on the market before 2 August 2026, the technical marking applies from 2 December 2026. This guide shows which duty applies to your company, with sample wording and a checklist. As of November 2026.
What the AI labelling obligation covers
The legal basis is Regulation (EU) 2024/1689, the AI Act. Its Article 50, "Transparency obligations for providers and deployers of certain AI systems", applies regardless of whether a system is classified as high-risk. It therefore covers the website chatbot, the image generator in marketing and the phone assistant in customer service.
When the obligation applies
Article 50 has applied since 2 August 2026. The Digital Omnibus Regulation (EU) 2026/1744 of 8 July 2026, published on 24 July 2026, changed one point: providers of systems generating synthetic audio, image, video or text that were placed on the market before 2 August 2026 must implement the machine-readable marking under Article 50(2) by 2 December 2026. All other transparency duties, including deepfake disclosure, apply without delay.
| Date | What applies |
|---|---|
| 2 August 2026 | Notice when people interact with AI, information on emotion recognition, disclosure of deepfakes and of AI text on matters of public interest |
| 2 August 2026 | Machine-readable marking for generative systems placed on the market from that date |
| 2 December 2026 | Machine-readable marking also for generative systems already on the market before 2 August 2026 |
Providers and deployers
A provider develops an AI system, or has it developed, and places it on the market or puts it into service under its own name. A deployer uses an AI system under its authority in a professional context. Most companies are deployers. They become providers once they build, or commission, their own system and run it under their own name, such as a customer service assistant on their website, even if a large vendor's language model runs underneath.
The four cases at a glance
| Case | Who is obliged | What to do | Article |
|---|---|---|---|
| AI interacts directly with people | Provider | Inform people they are interacting with an AI unless obvious | Art. 50(1) |
| AI generates image, video, audio or text | Provider | Mark outputs in a machine-readable format as artificially generated | Art. 50(2) |
| Emotion recognition or biometric categorisation | Deployer | Inform the people exposed | Art. 50(3) |
| Deepfake or AI text on matters of public interest | Deployer | Disclose that the content is artificially generated or manipulated | Art. 50(4) |
The information must be clear and provided at the latest at the first interaction or exposure (Art. 50(5)).
Chatbots and phone assistants
If you run your own assistant, customers should learn at the start that they are talking to an AI, for example: "Hello, I am the digital assistant of Example Ltd and I work with artificial intelligence. If you would rather talk to a person, just type 'agent'." A route to a human is not mandatory but noticeably improves acceptance. See automating customer service with AI.
AI images, video and deepfakes in marketing
Not every AI image is a deepfake. The regulation means content that resembles real people, objects, places or events and would falsely appear authentic. Photorealistic AI images in ads, on websites and in social media should carry a visible note such as "illustrative image, AI-generated". Do not strip the machine-readable markers many tools write into files. The European Commission has published voluntary EU icons for labelling AI-generated content that make visible labels consistent.
Examples: when AI-generated content must be labelled
| Example | Label needed? | Why |
|---|---|---|
| Your own chatbot answers customer questions on your website | yes, notice at the start | direct interaction, Art. 50(1) |
| Photorealistic AI image of a real person in an ad | yes | deepfake, Art. 50(4) |
| Synthetic voice of the CEO in a corporate video | yes | deepfake |
| AI illustration in a drawn style for a blog post | no, voluntary note possible | obviously artificial |
| Product description in a shop, reviewed by editors | no | not public interest, editorial control |
| Unreviewed AI text on a matter of public interest | yes | Art. 50(4), second subparagraph |
AI text on your website
Deployers must disclose AI-generated text published to inform the public on matters of public interest, unless it has undergone human review or editorial control and someone holds editorial responsibility. Product copy and blog posts with responsible editors are usually not covered. Clear internal rules are also part of the AI literacy measures Article 4 asks of providers and deployers; see our AI training for teams.
Fines and supervision
Breaches of the transparency duties can be fined up to 15 million euros or 3 percent of worldwide annual turnover, whichever is higher (Art. 99(4)); for SMEs the lower amount applies (Art. 99(6)). In Germany the Federal Network Agency (Bundesnetzagentur) is the main market surveillance authority under the national AI act of 22 July 2026, with BaFin remaining responsible for the financial sector. GDPR continues to apply; see GDPR-compliant AI agents.
Checklist
- List every AI system that talks to customers or generates content, with an owner.
- Clarify the role per system: provider or deployer.
- Add a notice at the start of every chat or call handled by an assistant.
- Build machine-readable marking into your own generative solutions, for legacy systems by 2 December 2026.
- Label photorealistic AI images and synthetic video visibly and keep the metadata.
- Assign and document editorial responsibility for published text.
- Put the rules in writing and train the team.
Want to know which duties your tools trigger? We are happy to go through your list in a short intro call. This guide does not replace legal advice in an individual case.
Transparency note: this article was produced with the help of artificial intelligence.