On 2 August 2026 the transparency obligations of the EU AI Act take effect. Anyone serving customers through a chatbot, a voicebot or AI-assisted reply suggestions must disclose from that date that artificial intelligence is involved. For contact centres this is not an academic debate but a task list with a deadline.
What Article 50 requires
The rule is plain: people must know when they are talking to a machine — unless that is obvious anyway. For customer service this means four things.
- Notice at first contact. Visible in the chat window or audible in the voicebot greeting. Not in the terms and conditions, not in the privacy policy.
- Labelling of synthetic content. AI-generated speech, text and images must be marked as machine-generated.
- A way through to a person. Customers must be able to reach a human being — and that route has to work in practice.
- Emotion recognition needs consent. Anyone analysing the mood in a caller’s voice requires the caller’s explicit agreement.
How large the fines actually are
Two figures circulate here, and the distinction matters. For breaches of the transparency obligations in Article 50, Article 99 provides for fines of up to 15 million euros or 3 per cent of worldwide annual turnover. The frequently quoted 35 million euros or 7 per cent apply to prohibited AI practices under Article 5 — a different offence altogether.
Both are serious enough. But anyone arguing for a budget internally should quote the right number.
Service AI is rarely a high-risk system
Many companies fear that a chatbot automatically drops them into the high-risk category, with conformity assessment and registration duties. As a rule that is not the case. Ordinary customer service AI — chatbots, voicebots, reply suggestions, conversation summaries — falls under limited risk and therefore only under the transparency obligations.
Two areas are exceptions and deserve a closer look: systems that assess creditworthiness or eligibility for essential services, and systems used in recruitment. Neither is typical of a contact centre, but both can appear on its edges.
The postponement does not apply here
For contact centres nothing changes: the transparency obligations in Article 50 are unaffected by the postponement under discussion. 2 August 2026 stands.
Five steps to take now
- Take stock. Where does AI touch your customer contact? Chatbot, IVR with speech recognition, summaries in the ticketing system, automatic email classification — put it all on one list.
- Write the notices. Short, comprehensible, in the right place. For example: “You are chatting with our AI assistant. To speak to a colleague, simply type human.”
- Check the escalation path. Does the handover to a person work — including the conversation context, so the customer does not start from scratch?
- Train the team. The AI literacy obligation in Article 4 has applied since February 2025. Your people should know what the systems in use can do and where their limits lie.
- Document it. A simple register listing system, purpose, provider and the notice used. That is enough for a supervisory enquiry and it makes the next review easier.
None of this is a major project. But it takes time, and the deadline is fixed.
This article gives a general overview and does not replace legal advice. How the obligations apply in a particular case depends on the systems in use.
