From 2 August 2026, Article 50 of the EU AI Act requires that people interacting with an AI system, a website chatbot, a voice agent on the phone, an AI assistant in an app, are told they are dealing with AI, unless it is obvious from the context. If your Irish business runs a customer-facing chatbot, review the provider’s implementation and the actual customer journey, and should be checked on every channel where the bot appears.
What Article 50 requires
Article 50(1) places the design obligation on providers: AI systems intended to interact directly with people must be designed so that the people concerned are informed they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant person from the circumstances. The interaction-disclosure provision differs than the high-risk regime, no conformity assessments, no logging mandates, but its scope and exceptions must be checked, whatever the risk level. A product-questions chatbot that is otherwise minimal risk still needs the disclosure.
Which of your systems are caught
- Web chatbots, customer service, sales and FAQ bots on your site.
- Voice agents, AI answering or making phone calls, taking bookings, routing queries. Voice deserves particular care, because a caller has fewer contextual clues than a website visitor.
- Messaging-channel bots , AI responding to customers on WhatsApp, Instagram or SMS, where the customer may assume a human is typing.
- In-app assistants , AI helpers embedded in your product or customer portal.
The “obvious from context” carve-out is narrower than it sounds. A tool literally branded as an AI assistant may qualify; a chat widget that says “Chat with us” and answers in a conversational tone almost certainly does not. The safe approach is to disclose and stop relying on the exception.
What good disclosure looks like
- Label the entry point, “AI assistant” on the widget or menu, not a vague “Chat with us”.
- State it in the opening message: “You’re chatting with our AI assistant.” For voice agents, say it at the start of the call.
- Keep it in plain language, the test is whether an ordinary customer understands, not whether a lawyer would.
- Offer the human route where one exists, which is good practice and defuses most complaints.
- Walk every journey where the bot appears, site, app, messaging, phone, and record the check. The disclosure must be present on each.
Separate content-marking and disclosure duties
Article 50 generally applies from 2 August 2026. Paragraph 2 concerns providers making synthetic outputs technically detectable, with exceptions. Paragraph 4 concerns deployers disclosing deepfakes and certain AI-generated public-interest text. The text provision includes an exception for human review or editorial control with editorial responsibility. This is not a blanket rule to label every AI-assisted marketing document.
Keep a record of the disclosure decision
Record the chatbot’s purpose, provider, customer journey and disclosure decision. Check the current AI Act guidance and the supplier’s implementation, including applicable exceptions. A register helps retain that evidence; a software entry does not by itself establish compliance.
Legal references checked 4 October 2026: AI Omnibus in force; Article 4 guidance; Current application dates; Article 50 transparency; High-risk classification; Irish Act and AI Office. Check the provision and any transition that applies to your particular system.
One specific transition matters: Article 111(4) gives providers of synthetic-content systems placed on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2). This is not a general postponement of deployer disclosures. Article 111 also contains separate existing-system high-risk transitions.