EU AI Act for Call Centers: AI Agents and Staff Monitoring

What does the EU AI Act change for call centers?

A call center uses AI in two directions. One faces customers: AI voice agents, routing, call summaries. The other faces staff: quality scoring, coaching, monitoring. The AI Act is gentle on the first and strict on the second.

Customer-facing AI mostly needs honesty: callers must be told they are talking to an AI (Article 50). Staff-facing AI is where the ban and the high-risk list sit, because the Act treats the workplace as a place where people cannot easily say no.

Can a call center use AI to track its agents' emotions?

No. Article 5(1)(f) prohibits "the use of AI systems to infer emotions of a natural person in the areas of workplace and education institutions, except where the use of the AI system is intended to be put in place or into the market for medical or safety reasons". It has applied since 2 February 2025.

The European Commission's guidelines on prohibited practices give a call center as one of their examples. Germany's AI Act regulator, the Bundesnetzagentur, lists it in its summary: "The use of webcams and voice recognition systems by call centres to track their employees' emotions, such as anger" (Bundesnetzagentur).

In practice, that rules out features such as agent stress scores, anger alerts on the agent's voice, or "agent mood" dashboards built from voice analysis. A banned practice can cost up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher (Article 99(3)).

Two limits of the ban, according to the Commission's guidelines as summarized by the Future of Privacy Forum: emotions inferred from written text alone are outside it, because the ban needs biometric data such as voice or face; and it covers the employees, not the customers on the line. Text analysis used to score staff can still be high-risk, as the scoring section below shows.

Can a call center analyze customers' emotions?

It is not banned, as long as the same tool does not also track the agent's emotions. But it is regulated. Emotion recognition systems are on the high-risk list in Annex III point 1(c), with full duties from 2 December 2027. And Article 50(3) already requires, from 2 August 2026, that the people exposed to an emotion recognition system are told it is running.

So a real-time "customer sounds frustrated" prompt is possible, but it needs a notice to the caller now and high-risk controls later. Ask the vendor whether their sentiment feature reads the voice or only the text: the answer changes the rules that apply.

Which call center AI tools are banned, high-risk or simply disclosed?

What must an AI voice agent tell callers?

Article 50(1) requires AI systems that talk to people to be designed so that people know they are dealing with an AI, unless it is obvious. Article 50(5) says the information must be clear and given at the latest at the first interaction. On the phone, that is the first sentence.

A simple opening: "Hello, this is the AI assistant of [Company]. I can help with your order, or put you through to a colleague." It names what the caller is talking to and how to reach a person. Use the same opening for outbound calls, and answer honestly whenever a caller asks if they are talking to a human.

If you run calls for clients under their brand, agree in writing who sets the greeting, so the disclosure is never dropped when a client edits the script. Breaching Article 50 can cost up to EUR 15 million or 3% of worldwide annual turnover; for SMEs the lower of the two applies (Article 99(4) and (6)).

Is AI quality scoring of human agents high-risk?

Yes. Annex III point 4(b) lists AI used "to allocate tasks based on individual behaviour or personal traits or characteristics or to monitor and evaluate the performance and behaviour of persons" in work relationships. Automated QA scorecards, ranking agents, flagging "low performers" and coaching that grades each agent all fit that description.

There is no easy way out through Article 6(3). It lets some Annex III systems escape the high-risk label when they only do narrow or preparatory tasks, but it also says an Annex III system is always high-risk when it profiles people. Evaluating a named agent's work performance is profiling.

Scoring calls in aggregate, for example "how many calls mention a refund this week", without rating individual staff, is a different use. Keep the two apart in your tools and your documents.

What must a call center do for high-risk AI?

When you use a high-risk system built by a vendor, you are its "deployer". These duties from Article 26 apply from 2 December 2027.

What are the key dates for call centers?

The ban did not move: the July 2026 amendment delayed the high-risk duties, not the prohibitions. The AI Act also reaches call centers outside the EU when the output is used in the EU (Article 2), so offshore teams serving EU customers are in scope too.

How does an AI phone agent from AInora fit?

AInora builds AI voice agents that take inbound calls and make outbound ones: first-line answers, bookings, order questions and reminders. It is a customer-facing agent, not a tool for scoring your staff. How it lines up with the rules above:

• The agent says it is an AI assistant in its first sentence, in the words you choose.
• It hands the call to a person when the caller asks, or when your rules say so.
• Every call gets a transcript and a summary your team can review.
• You decide what it may and may not ask or say.
• It speaks many languages, so the same disclosure holds on every line.

Your counsel decides the rules; the agent follows them. To hear one on your own calls, fill out the short form and we build a demo agent from your website. For the bigger picture, see AI voice agents for European contact centers.