Article 50 EU AI Act: Compliant vs Non-Compliant Disclosure Examples
One of the most common questions from businesses trying to get ready for the 2 August 2026 deadline is deceptively simple: what does a compliant disclosure actually look like? The EU AI Act Article 50 sets out the obligation clearly — inform users they are interacting with an AI system, clearly and in a timely manner — but leaves the exact wording open. That flexibility is useful, but it also means businesses make avoidable mistakes.
This article gives you concrete examples of disclosures that work, disclosures that don't, and the reasoning behind each.
The legal baseline: what Article 50 actually requires
Before looking at examples, it helps to be clear on what the standard is. Under Article 50(1) of the EU AI Act, deployers of AI systems designed to interact with natural persons must ensure those persons are informed they are interacting with an AI system. This information must be:
- Provided at the latest at the start of the first interaction
- Clear and distinguishable — meaning actually noticeable, not buried
- Timely — not after several exchanges have already happened
The May 2026 draft guidelines from the European Commission added useful detail: a disclosure that appears once in a header and is never shown again may not be sufficient for AI systems where users build ongoing relationships. The disclosure should be part of the interaction itself, not a legal notice the user scrolls past before the chat begins.
Chatbot first-interaction disclosure examples
✅ Compliant
Example A — Standard tone:
"This is an AI assistant. I can answer questions about [topic] but I am not a human. How can I help you?"
Why it works: identifies the system as AI, appears as the first message before any user input, and is written in plain language that a non-technical user will immediately understand.
Example B — Friendly tone:
"Hi there! I'm an AI assistant here to help with your questions. You're chatting with AI, not a human — but I'll do my best to help. What can I help you with today?"
Why it works: same core information, delivered in a warmer register. Tone is a choice; the legal requirement is the disclosure, not the formality of it.
Example C — Formal B2B tone:
"Welcome. Please note that you are now interacting with an AI-powered assistant. This system is not a human representative. For complex matters, you may request to speak with a member of our team."
Why it works: formal register appropriate for a B2B context, clear AI identification, and an explicit human escalation path — the last point is not legally required but is good practice.
❌ Non-compliant or high-risk
Example D — No disclosure at all:
"Hi! How can I help you today?"
Why it fails: gives no indication the system is AI. A user could reasonably assume they are speaking with a human. This is the baseline violation that Article 50 is designed to prevent.
Example E — Buried in terms:
[Chat widget opens. In tiny text at the bottom of the widget: "Powered by AI"] "How can I assist you?"
Why it fails: the disclosure is present in some sense, but it is not delivered at the start of the interaction, is not distinguishable from a generic UI label, and does not inform the user they are interacting with an AI system in the way the regulation means — as part of the conversation, before they engage.
Example F — After initial engagement:
[First message from user: "I need help with my order"] "Thanks for reaching out! Just so you know, I'm an AI assistant — how can I help with your order?"
Why it fails: the disclosure comes after the first interaction has already occurred. The regulation requires disclosure at the latest at the start of the first interaction, meaning before the user sends anything.
Example G — Ambiguous phrasing:
"You're speaking with our virtual support team."
Why it fails: "virtual support team" does not clearly communicate AI. A user unfamiliar with the term might assume it means a remote human team. The disclosure must be unambiguous about the AI nature of the system.
Persistent chat label examples
Article 50 does not only require a first-message disclosure. The draft guidelines note that for ongoing interactions, persistent labelling — a visible indicator throughout the conversation — supports compliance.
✅ Compliant persistent labels
- "This chat uses AI ⓘ" (shown in the chat header throughout the conversation)
- "AI Assistant · Disclo-verified" (shown as a subtitle under the chat name)
- A small badge visible in the corner of the chat window at all times
❌ Insufficient
- A notice that appears only on the first message and then disappears
- A hover tooltip that only appears if the user specifically looks for it
- AI disclosure only in the pre-chat survey but not in the chat interface itself
Voice agent first-turn script examples
For AI voice agents, the same principle applies — the first thing the user hears should identify the system as AI.
✅ Compliant
"Hello, you've reached [Company Name]'s AI assistant. I can help with account questions, orders, and general enquiries. For complex issues, I can transfer you to a team member. How can I help you today?"
Why it works: immediately identifies itself as AI, sets expectations, and offers a human escalation path.
❌ Non-compliant
"Hello, thank you for calling [Company Name]. How can I direct your call today?"
Why it fails: sounds indistinguishable from a human receptionist. Nothing in the opening identifies this as an AI system.
What makes a disclosure compliant: the checklist
Before you go live, check each AI touchpoint against these four criteria:
- Timing — does the disclosure appear before the user's first input, not after?
- Clarity — is the word "AI" or equivalent unambiguous language used? ("Virtual assistant" alone probably isn't enough.)
- Visibility — is it part of the conversation itself, not just a header or a footer?
- Persistence — for ongoing chat interactions, is it visible throughout, not just at the start?
Not sure whether your current setup qualifies? Run the free Scope Check to find out exactly which obligations apply to your site.
Answer 5 questions and get a free personal report with exact obligations and deadlines for your setup.
What the evidence requirement adds
Having compliant disclosure wording is step one. The second step — the one most businesses miss — is being able to prove that the disclosure was actually shown to users, on a specific date, in a specific form.
A screenshot taken today does not prove the disclosure was there in September. A configuration setting in your chat platform does not prove the notice was actually rendered on a user's screen. When a national competent authority requests evidence under Article 99, they need documentation of what was shown and when — not what your settings currently say.
This is what an evidence log is for. Disclo's hosted badge logs a disclosure event every time the notice is rendered on a visitor's screen, with a timestamp. The log is exportable on demand.
Get the disclosure wording and the evidence log handled today.
Founding price. Regular price €69 from 1 July 2026.
This is general guidance, not legal advice. Based on the official EU AI Act and the Article 50 draft guidelines published 8 May 2026. Consult a qualified EU technology lawyer for advice specific to your situation.
Stay ahead of the deadline.
Get notified the moment the final EU Article 50 guidelines are published — so you can act before your competitors do.