AI Act chatbot: what your app must tell users

Since 2 August 2026, an AI chatbot or assistant in your app must tell people they are dealing with an AI, at the latest when the conversation starts, unless that is already obvious. The rule is Article 50 of the EU AI Act. It falls on the company that offers the assistant under its own name, even when the model behind it comes from OpenAI, Anthropic or Google.

If the app also generates images, audio, video or text with AI, those outputs must carry a machine-readable mark and be detectable as AI-generated. For AI systems placed on the market before 2 August 2026, that part applies from 2 December 2026. The duty to tell people they are talking to an AI has no such grace period.

This article is for companies with an AI feature in their own app: a support chatbot, a voice assistant, an agent that acts for the user, or generated content. It follows the European Commission's guidelines on Article 50 of 20 July 2026, linked at the end. It's general information, not legal advice.

What Article 50 requires

In practice these are three separate checks: the notice users see, the technical marking of generated outputs, and the labels on content you publish. Decide who owns each before you ask a supplier what it has built.

Article 50(5) adds how: the information must be given "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure", and it must meet the applicable accessibility requirements.

The AI Omnibus, the amendment to the AI Act in force since 27 July 2026, postponed the rules for high-risk AI systems, not these. Its only change to the dates of Article 50 is the December transition for marking.

Are you the provider when the model is someone else's?

If the AI feature runs under your company's name, usually yes. The AI Act defines a provider as whoever "develops an AI system [...] or that has an AI system [...] developed and places it on the market or puts the AI system into service under its own name or trademark". A deployer, by contrast, is whoever uses an AI system under its authority.

The Commission's guidelines give the example that fits most apps: a company that offers a chatbot, an image generator or an AI agent under its own name is the provider responsible for Article 50(1) and (2), whether the feature is free or paid. So is a company that builds a chatbot in-house and runs it for its own use under its own name.

Two consequences follow:

• A supplier building it doesn't move the duty. "Has [...] developed" is part of the definition. If an agency builds the assistant and it runs in your app under your brand, you're the provider.
• The model vendor can help, but the duty stays with you. Article 50 doesn't explicitly apply to general-purpose AI models. The Commission encourages model vendors to build in measures that make compliance easier for the providers who use their models, and lets you rely on a vendor's marking "without prejudice to the responsibility of the provider of the AI system to demonstrate compliance".

Which features need the AI notice

The Commission names four conditions: an AI system, intended to interact, directly, with people. Covered examples include chatbots and voice assistants in customer support, e-commerce, finance, healthcare or complaints handling, AI hotlines, AI avatars, and AI agents that make bookings, handle correspondence or buy on a user's behalf. An agent must also say on whose behalf it acts.

Outside the notice in Article 50(1):

• Automated answers that aren't AI, such as rule-based quick replies or an out-of-office message.
• Features without a conversation: recommendations, search that doesn't generate content, autocomplete, translation or transcription tools.
• AI behind a person: a support agent who uses an AI tool to draft replies, as long as people really review and send them. If an app mixes AI answers with human ones, the AI answers need the disclosure.

How to tell users

The AI Act prescribes no wording or design. The guidelines call for a clear notice at least once at the start of each interactive session, and give examples:

• a first message in which the chatbot says it is an AI,
• a label such as "You are interacting with an AI system" near the input field, possibly kept visible as a badge,
• in a voice assistant, a spoken sentence at the start, such as "This is an AI-powered assistant".

The guidelines also list what isn't enough on its own:

• a mention only in the terms and conditions or the documentation,
• an invisible marker in metadata,
• a vague label such as "assistant", or a human name and photo that suggest a person,
• one general line for a whole service, such as "Services on this website use AI",
• a technical description such as "this system uses LLMs".

One prominent notice before the first interaction is likely to be enough in most cases. In riskier contexts, the Commission says reminders during the conversation are likely to be necessary: financial advice, insurance, legal and health advice, complaints handling, and conversations with children, older people or people with disabilities. In every case, the assistant must say that it is an AI when someone asks.

The notice must be accessible too. If your app falls under the European Accessibility Act, its requirements apply to the notice as well; see Does the European Accessibility Act apply to your app?

An illustrative example, not a real app: an energy supplier's assistant opens with "Hi, I'm the virtual assistant. I'm an AI, not a person. I can help with bills and meter readings, and for anything else I'll pass you to our team." An "AI" label stays beside the reply box for the whole conversation.

The "obvious" exception is narrow

No notice is needed where the AI is obvious to "a natural person who is reasonably well-informed, observant and circumspect". The Commission says the exception must be read restrictively, and that the provider has to assess and demonstrate it. If the general public or vulnerable people can reach the assistant, the exception can't be relied on.

The guidelines' examples of obvious cases are narrow: a coding assistant for professional developers, an internal assistant for trained staff who know they are using AI, a tool for trained health professionals. For an assistant in a customer app, assume you need the notice.

Generated images, text and audio

If your app generates content with AI, for example product images, summaries, letters or a synthetic voice, the provider must make sure each covered output carries a machine-readable mark and that there is a way to detect it. A mark without a means of detection isn't enough. The mark can come from the model: you may rely on the vendor's solution if it meets Article 50(2), but showing that it does remains your job.

Outside the rule:

• short outputs such as single words, captions, alt text and interface labels,
• source code, and machine-to-machine output that no person sees,
• standard editing: grammar and spelling correction, light stylistic polishing, AI translation, formatting, minor photo corrections,
• in narrow cases, purely technical output that stays inside the company, is seen only by a limited group of staff and isn't meant to leave it.

The deadline: a generative AI system placed on the market before 2 August 2026 must mark its outputs from 2 December 2026. For an app, we read that as a feature your users already had before that date. A system placed on the market since then must mark them from the start. Content generated before 2 August doesn't have to be marked retroactively.

The Commission has also published a voluntary code of practice on the transparency of AI-generated content, in its final version since 10 June 2026. The Commission and the AI Board assessed it as adequate, so companies that sign it can rely on its measures to show they meet the marking and labelling duties.

Texts and deepfakes you publish

Article 50(4) concerns what a company publishes, not what users generate for themselves in the app. Two cases:

• Deepfakes, for example a realistic AI video or voice of a person in a campaign, must be disclosed at the latest when people first see them.
• AI-generated text published to inform the public on matters of public interest must be labelled, unless it has gone through human review or editorial control and a person or company holds editorial responsibility for the publication. The Commission says a spell-check or a formal check isn't a review.

The Commission offers optional EU icons for labelling AI-generated content. The labelling duty itself isn't optional.

Who enforces it and what a breach can cost

National market surveillance authorities enforce Article 50, on their own initiative or after a complaint, which anyone may file. Fines reach EUR 15 million or 3% of worldwide annual turnover, whichever is higher. For small and medium-sized enterprises, the lower of the two applies.

In Czechia, the Ministry of Industry and Trade has coordinated the AI Act since 28 May 2025. We couldn't confirm from official sources which Czech authority will enforce Article 50, or the status of the Czech implementing law. On 12 November 2026, the ministry runs a webinar on Article 50 and the Commission's guidelines together with the Ministry of Culture and the Czech Telecommunication Office.

What to do now

1. List every AI feature in your apps and on your website: chat, voice, agents, generated images or texts, summaries. Include features a supplier added.
2. Mark where you're the provider. If the feature runs under your name, you usually are, whoever built it.
3. Open each conversation as a new user would. Is there a clear notice before or in the first answer, outside the terms and conditions? Does the assistant have a human name or photo?
4. In sensitive flows, such as money, insurance, health, complaints and apps for children or seniors, add reminders and test that the assistant says it is an AI when asked.
5. Ask your model vendor in writing how marking and detection work, and record whether that covers your app's outputs. For a generative feature your users had before 2 August, the deadline is 2 December 2026.
6. Label what you publish, or record who reviewed it and who is responsible for it.
7. Write down your reasoning, especially if you rely on the "obvious" exception. The provider is the one who has to demonstrate it.

Who holds the accounts, keys and instructions of an AI feature is covered in Who owns the AI in your app? If you're planning an AI assistant in your app, or want a second look at how an existing one introduces itself, the mobile app development page describes how we work. A short description through the contact page is enough to start.

This article is general information, not legal advice. It reflects the AI Act as amended by the AI Omnibus and the Commission's guidelines as at 4 November 2026. Whether and how Article 50 applies to your feature is a question for a lawyer.

Sources

Checked on 4 November 2026.

• AI Act Service Desk, European Commission: Article 50: Transparency obligations for providers and deployers of certain AI systems
• AI Act Service Desk, European Commission: Article 3: Definitions
• AI Act Service Desk, European Commission: Article 99: Penalties
• AI Act Service Desk, European Commission: Article 113: Entry into force and application
• European Commission, 20 July 2026: Guidelines on the transparency obligations for providers and deployers of AI systems
• European Commission: Transparency obligations under Article 50 of the AI Act
• European Commission: Code of Practice on transparency of AI-generated content
• European Commission: EU icons for labelling AI-generated content
• European Commission, 27 July 2026: AI Omnibus enters into force
• Ministry of Industry and Trade: Umělá inteligence
• Ministry of Industry and Trade, 3 November 2026: Jak označovat obsah vytvořený pomocí AI? MPO pořádá webinář