The transparency deadline has passed. What Article 50 actually changes for a contact centre.

Article 50 of the EU AI Act came into force on 2 August 2026. The European Commission published its final guidelines on 20 July, so the interpretation is now settled enough to plan against. If you run a contact centre that touches EU customers, this is the piece of the AI Act that lands on your queue first.

There is no grace period on this one. The Act was adopted in 2024. Transparency was always scheduled for August 2026. That date has now arrived.

What Article 50 actually says

Article 50 covers four transparency duties. Two of them are directly relevant to any contact centre that uses AI or that receives calls from customer-side AI.

First, any AI system that interacts with a natural person must make clear to that person that they are dealing with an AI, unless it is obvious from context. That applies to your Mode 2 IVR, chatbot and voice agent, and to any customer-side AI that reaches you in Mode 3.

Second, AI-generated audio, image, video or text must be labelled in a machine-readable way. Deep fakes and AI-generated content on matters of public interest must be disclosed to the person receiving them. Your outbound comms, escalation notes and auto-generated case summaries are in scope if they end up in front of a customer.

The other two limbs (emotion recognition and biometric categorisation) may or may not apply to you depending on your stack. If your voice platform reads customer sentiment or classifies callers by voice biometrics, get advice.

What this changes on the queue

Three things become non-negotiable from now.

Disclosure at first contact. Your IVR and your voice agent have to tell the caller they are interacting with an AI. “Hi, this is Ava, a virtual assistant for [brand]” is the pattern. It has to be prompt, unambiguous, and given before the caller has committed to the interaction. “Obvious from context” is not a defence for a natural-sounding voice.

Disclosure on the way back too. If the customer's AI is calling you, your human agents need a way to recognise that and to log it. Right now, most contact centres have no field for “caller was an AI acting for a named consumer”. That has to change, because the transparency obligation runs both ways once the caller declares.

Machine-readable labels on AI-generated content. Anything an AI drafts on your behalf and sends outward (email replies, chat responses, case summaries seen by the customer) needs to carry a machine-readable marker. Practically, that means your CCaaS, your ticketing platform, and your outbound comms stack all need a field for “this content was AI-generated”, populated automatically, and preserved when the content is forwarded.

What a contact centre leader needs on the record

You do not need a compliance project. You need six things written down, in one place, ready to hand to an auditor.

One: a list of every AI system on your queue, on your channels, and in your case-handling stack. Two: the exact disclosure wording used at the start of every AI interaction, per channel. Three: how AI-generated content is marked and preserved through your systems. Four: how your agents handle inbound customer-side AI calls, and how they log them. Five: your fallback rule when an AI on either side cannot complete the interaction. Six: who owns each of those decisions on your operating chart.

That list is the Service Handshake for a contact centre, cut down to what Article 50 asks you to declare. It is not extra work on top of Dual CX; it is the same work, in the shape the regulator now expects.

The read across to Modes 3 and 4

Article 50 tacitly assumes a Mode 2 world: your AI, their human. But the moment a customer-side AI shows up on your line (Mode 3), the same transparency logic runs the other way. If the caller AI does not declare, your operation should. If your operation does not declare, the caller AI should. Neither of you gets to hide behind “obvious from context” once both sides can carry AI.

That is why we built the Service Handshake around declared authority, disclosure and fallback. It is what Article 50 was reaching for, extended to the case the Act does not yet name.

If you can hand an auditor those six declarations today, Article 50 is a form-filling exercise. If you can't, it is a redesign. Better to do the redesign now, once, than to retrofit it under a complaint.