The EU AI Act in customer service: transparency duties, explained briefly

The EU's AI Act, Regulation (EU) 2024/1689, has been in force since 1 August 2024. Its obligations apply in stages. Under Article 113, the transparency duties in Article 50 have applied since 2 August 2026.

They affect almost every customer service operation that uses AI in conversation. This post summarises what Article 50 requires, who is affected and how the duties can be reflected in a contact centre. It is not legal advice.

#Providers and deployers: who carries which duty

The regulation distinguishes two roles. Providers develop an AI system or place it on the market under their own name. Deployers use it under their own authority, such as an insurer running a voicebot on its hotline.

In customer service, both roles meet. Some duties sit with the platform provider, others with the organisation deploying the bot. Many can only be met together: the provider supplies the capability, the deployer configures and uses it.

#The four duties in Article 50

Paragraph 5 sets out the how. The information has to be clear and distinguishable, given at the latest at the first interaction, and meet the applicable accessibility requirements.

#Bots and voicebots: disclosure in conversation

For chatbots, voicebots and AI agents, paragraph 1 is the core. Anyone talking to them must learn that they are interacting with an AI. In customer service, it is unwise to rely on the "obvious" exception: a well-made voicebot is precisely not obvious.

Formally, paragraph 1 is addressed to the provider, who has to design the system accordingly. In practice, deployers write the texts. Disclosing AI without breaking the conversation shows what a disclosure sounds like when it does not get in the way.

Service EN· active

Queues

Voice3 waiting

Chat5 waiting

Email2 waiting

Messaging1 waiting

Conversation

AI agent, disclosedHandover from AI agent · context complete

I received the decision letter, but the amount does not match my application.

I have the case in front of me. One moment, I am checking the calculation.

Sentiment: calmVoice

Context

Case

2026-0412

History

3 · 14 d

Univerx Assist

Suggested reply

The letter deviates from the application. I will propose a correction and send you the confirmation by message.

InsertAccept

Summary

Request: deviation in the letter. Documents on file. Next step: correction, confirmation by message.

#Generated content: marking in the data format

Paragraph 2 covers synthetic content, which includes a voicebot's generated speech and generated text. It has to be identifiable as artificially generated in a machine-readable format.

For systems already on the market before 2 August 2026, amending Regulation (EU) 2026/1744 allows until 2 December 2026.

Systems that only assist with standard editing, or that do not substantially alter the input, are exempt. For reply suggestions that an employee reviews and sends herself, the classification has to be assessed case by case.

Either way, clean documentation of roles matters: who generated the text, who reviewed it, who sent it?

#Sentiment analysis: a line worth knowing

Conversation analytics often evaluate sentiment. That is not automatically emotion recognition in the regulation's sense, which is tied to biometric data. But if emotions are inferred from voice characteristics, for example, the information duty in paragraph 3 applies.

For employees, a stricter line applies. Since 2 February 2025, Article 5(1)(f) prohibits AI systems that infer the emotions of people in the workplace, except for medical or safety reasons. Analytics about employees therefore deserve particular care.

#When it is about more than transparency

Most service bots are not high-risk systems. That changes when AI helps decide on essential services. Annex III lists, among others, creditworthiness assessment and risk assessment and pricing in life and health insurance.

Eligibility checks for public assistance benefits are listed too. Banks, insurers and public authorities should therefore look closely at every use case. A bot that answers questions is something different from a system that helps decide on an application.

#AI literacy: a duty that already applies

Since 2 February 2025, Article 4 has required providers and deployers to take measures on their staff's AI literacy. Since the 2026 amendment, these are measures that support building that literacy; no particular level has to be guaranteed.

For contact centres, that means anyone working with AI assistance or looking after bots should understand how they work and where their limits are.

#Reflecting the duties in the product

The duties translate into concrete requirements for a platform. Univerx is designed for these requirements:

The Trust Center page on the EU AI Act covers the classification and the documents. The principles behind it are in Responsible AI, and the implementation in the platform in Europe & regulation.

The guide Running a contact centre in Europe gives an overview of all European requirements. The glossary explains the EU AI Act.