AI Act Article 50 is the EU rule that makes AI honest about itself. If your chatbot or voice assistant could be mistaken for a human, users must be told they are talking to AI. If you use AI to create realistic images, video, audio or certain public texts, people must be able to tell they are AI-made. These rules have applied since 2 August 2026.
Article 50 applies to all AI systems covered by it, not only to “high-risk” AI. So even a simple website chatbot or an image generator used for marketing can be caught. Below we explain what it requires and what to do.
What AI Act Article 50 requires: the five duties
Article 50 contains five separate duties. Two fall on the provider, which means the company that develops an AI system and puts it on the market under its own name. Three fall on the deployer, which means the business that uses an AI system in its own work, for example a shop that adds a chatbot to its website.
Many businesses are both: if you build and offer your own chatbot, you are usually its provider; if you just switch on someone else’s tool, you are usually a deployer.
| Duty | Who | What it means in practice |
|---|---|---|
| 50(1) Chatbots and other interactive AI | Provider | Build the system so people are told they are talking to AI |
| 50(2) Marking AI-generated content | Provider | Add a hidden, machine-readable mark to AI-made audio, images, video and text |
| 50(3) Emotion recognition and biometric categorisation | Deployer | Tell people when such a system is being used on them |
| 50(4) Deepfakes | Deployer | Clearly disclose that realistic image, audio or video content is AI-generated or manipulated |
| 50(4) AI text on public-interest topics | Deployer | Disclose AI-written text published to inform the public, unless a person has reviewed it and takes editorial responsibility |
Article 50(5) adds a general rule on how to give this information. It must be clear and easy to notice. It must be given at the latest at the first interaction or the first time the person sees the content. It must also be accessible to people with disabilities.
Does your chatbot need an AI label?
Article 50(1) covers AI systems that are meant to interact directly with people. Customer-service chatbots and phone voice bots are clear examples. A recommendation engine that works quietly in the background, without talking to anyone, is probably outside this duty.
The “obvious” exception
You do not need a notice if it is already obvious that the user is dealing with AI. The test is what a reasonably well-informed, observant and careful person would understand in the circumstances. This test comes from EU consumer law.
A bot clearly presented as a bot may pass this test. But if your audience includes older people, children or people with disabilities, what seems “obvious” to you may not be obvious to them.
How much do you need to say?
The duty is to tell people that they are talking to AI. You do not have to explain how the system works. A short line at the start of the chat is usually enough, for example: “You are chatting with an AI assistant. You can ask for a human at any time.”
The provider must build the notice into the product. As a deployer, check that it actually appears on your website or app.
Marking AI-generated content under Article 50(2)
Providers of AI systems that generate synthetic audio, images, video or text must make sure the output is marked in a machine-readable format and detectable as AI-generated or manipulated. The mark is meant mainly for detection software and platforms, not as a visible label.
The technical solution must be effective, interoperable, robust and reliable, as far as is technically feasible. In practice, providers often combine several methods:
- Metadata: information stored inside the file. It is easy to add but easily lost, for example when someone takes a screenshot.
- Watermarks: signals built into the content itself, which are harder to remove.
- Provenance tools, logging and fingerprints: ways to check later where content came from.
There is an exception for AI that only helps with standard editing, or that does not substantially change the input or its meaning. Spell-checkers, simple translation or noise reduction are typical examples.
Timing. New generative AI systems placed on the market from 2 August 2026 must comply straight away. Providers whose systems were already on the market before that date have until 2 December 2026 to meet the marking duty, under Article 111(4) as amended by the Digital Omnibus on AI.
Deepfakes, emotion recognition and AI-written news
Deepfakes
A deepfake is AI-generated or manipulated image, audio or video content that looks like real people, objects, places or events and could falsely seem authentic. If your business creates or publishes a deepfake, you must clearly disclose that it is artificial. Attach the disclosure to the content itself, not only to a caption that may be lost when it is shared.
For content that is clearly artistic, creative, satirical or fictional, the duty is lighter. You still need to disclose that AI was used, but in a way that does not spoil the display or enjoyment of the work.
Emotion recognition and biometric categorisation
If you use a system that reads people’s emotions or sorts them by face, voice or body features, you must inform the people exposed to it about how it operates. You must also follow data protection law, such as the GDPR. Remember that the AI Act separately bans emotion recognition in workplaces and schools, except for medical or safety reasons.
AI-written text on public-interest topics
If you publish AI-generated text to inform the public on matters of public interest, such as news, health or politics, you must disclose that it was AI-generated. The exception is where a human has reviewed or edited the text and a person or company holds editorial responsibility for it. A quick spelling check is unlikely to be enough.
Who is outside Article 50, and what are the fines?
People who use AI for purely personal, non-professional activities are not “deployers”, so the deployer duties do not apply to them. A business using AI for marketing, sales or customer support is not in this category. Some systems authorised by law for fighting crime are also exempt.
Breaching AI Act Article 50 can lead to fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups, the cap is whichever of the two is lower.
Note that a missing label does not, by itself, make content illegal for platforms to host; that depends on other laws, such as defamation rules.
Practical examples
Example 1: A Dubai online shop selling to customers in Germany. The shop adds an AI chatbot, built on a third-party model, to answer delivery and returns questions. It has a human name and photo. Because the bot talks directly to EU customers and looks human, it should show a clear notice at the start of every chat, such as “I am an AI assistant”. Its supplier contract should say who handles the notice.
Example 2: A London marketing agency making an AI video. The agency uses an AI tool to create a realistic video of a “customer” praising a product. The person does not exist, but looks real. This is likely to be a deepfake, so the video must clearly disclose that it is AI-generated. Presenting an invented person as a real customer could also break consumer law.
Example 3: A software company with its own image generator. Its tool has been sold to EU users since 2025, so its outputs must carry a robust machine-readable mark by 2 December 2026.
What to do next
- List your AI features. Note every chatbot, voice agent and content generator you use or sell.
- Decide your role for each one. Are you the provider, the deployer, or both? This tells you which Article 50 duties apply.
- Write and test your notices. Show a short AI notice at the first interaction.
- Check content marking. Ask your AI tool suppliers how they mark outputs. If you are a provider with an existing system, plan for the 2 December 2026 deadline.
- Set an editorial policy. If you publish AI-assisted content on public-interest topics, record who reviews it and who is responsible.
- Update your contracts and terms. Make sure supplier contracts and your website terms and e-commerce documents reflect how you use AI.
- Get advice where needed. Our AI Act compliance lawyers can review your chatbots and content workflows, and our AI and intellectual property team can help with ownership of AI-generated material. For the bigger picture, read EU AI Act explained for businesses or, for Gulf-based firms, how the EU AI Act affects UAE companies. To discuss your case, contact us.
This article is general information, not legal advice for a specific case.
Frequently asked questions
Does every chatbot have to say it is AI?
In most cases, yes. Under AI Act Article 50, the provider must design the chatbot so users are told they are talking to AI. The only exception is when this is already obvious to a reasonably well-informed, observant and careful person. If your users include older people or other vulnerable groups, do not rely on that exception.
When did AI Act Article 50 start to apply?
From 2 August 2026. Providers of generative AI systems that were already on the market before that date have until 2 December 2026 to add the machine-readable marking required by Article 50(2).
Do I need to label AI-generated images on my website?
It depends on the image. The tool provider must add a hidden, machine-readable mark. As the business publishing the image, you must clearly disclose deepfakes, meaning realistic content that looks like real people, places or events. Generic images that are obviously artificial may not be deepfakes, but check each case.
Does Article 50 apply to companies outside the EU?
It can. The AI Act can reach businesses outside the EU when they offer AI systems in the EU or when the output of their AI is used in the EU. A Dubai or UK business with EU customers should check its chatbots and AI content.
What is the fine for breaching Article 50?
Up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups, the cap is whichever of the two is lower.
Sources
- European Commission AI Act Service Desk: Article 50
- European Commission AI Act Service Desk: Article 99 (penalties)
- European Commission AI Act Service Desk: Article 111 (systems already on the market)
- European Commission AI Act Service Desk: Article 113 (entry into force and application)
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex
- Nicolaj Feltes, ‘Article 50 AI Act: Do the Transparency Provisions Improve Upon the Commission’s Draft?’, 16 (2025) JIPITEC 222
Insights & answers
Latest from PAD