Can the AI listen in on a phone call? Announcement, notes, recording

Who it's for

Anyone who uses an AI on the phone or wants to: to answer calls, to take call notes or to write summaries.

What you'll be able to do

Know what the announcement covers, what it doesn't cover, and which questions you should settle before you start.

As of

October 2026

Why the question is coming up now

On 8 September 2026, Telekom announced that business customers on its business mobile plans will be able to have calls summarised by AI from 5 October 2026. The feature is called “Voice AI Notes Mobile”; it writes a summary and turns it into tasks and calendar entries. It is aimed at small and medium-sized businesses.¹

One condition in the same announcement is worth noting: the feature may only be used with the prior consent of the people on the call. Once the summary has been sent, the content is deleted, according to Telekom.¹ Even a large provider makes consent a condition before the AI takes notes.

Art. 50 AI Act

2 Aug 2026²

applies since this day

Telekom

5 Oct 2026¹

launch of AI call notes

Data protection

66%³

named as an obstacle by companies that use AI

The concern is widespread: in a Bitkom survey, 66 per cent of the companies already using AI name data protection as an obstacle.³ On the phone it becomes tangible, because there is a person speaking who may not know who is listening.

The announcement: what the AI Act requires

Article 50 of the AI Act has applied since 2 August 2026. Under paragraph 1, people have to be told that they are talking to an AI.² The amending Regulation (EU) 2026/1744, known as the Digital Omnibus, pushed back the obligations for high-risk systems, but not the duty under paragraph 1.²

For the phone, in the assessment of the software vendor CATAMA, that means: a short announcement at the start of the call meets the duty.⁴ Exactly which notices are enough and which aren't is covered in the post Labelling AI: what chatbots, AI images and AI text have to disclose.

The announcement says who is speaking, not what happens to the call. So if you have built it in, you are not yet done with data protection.

Announcement, notes, recording: three different things

On the phone, an AI can do three things that sound alike and are treated differently in law. The comparison shows what happens in each case and what the sources say about it.

The statements on recording come from CATAMA, a software vendor, in a post about AI phone assistants in workshops. It is not an authority and not a law firm.⁴ I repeat them here because they are concrete enough to be useful in practice, not as my own reading of the Criminal Code.

If you only announce, but take notes or record, you have dealt with just the first of the three points. Which basis for the notes fits your business, consent or something else, depends on the individual case. That is a question for an adviser, not for this post.

Who processes the data: processing on your behalf

If the AI phone assistant runs at a service provider and not on your own systems, CATAMA names two things: a data processing agreement under Article 28 GDPR and a contractual exclusion from model training.⁴

A data processing agreement, DPA for short, is a contract under Article 28 GDPR. You generally need one when a service provider processes personal data on your behalf. Exclusion from model training means the provider may not use your calls to train its AI further.

If a service provider writes the note, then according to CATAMA a contract with you goes with it. How to collect such contracts from all your service providers without losing track is covered in the post on DPAs and vendor onboarding.

And if the AI says something wrong?

The announcement doesn't protect you from a wrong answer either. The Higher Regional Court of Hamm treated the answers of a chatbot on a website as a commercial act of its operator, because the operator has sufficient influence over the system and sets it in motion.⁵ The Hamm judgment rests on German unfair competition law, not on the AI Act.⁵

The judgment concerns a chatbot on a website, not the phone. I would still assume that you have to answer in the same way for what the AI promises on the phone. That is my assessment, not a court decision. What the judgment is about and what to check before launch because of it is covered in the post When your chatbot gets it wrong, you said it.

Before you start

The sources above can be turned into a list to go through once before you start. It is my recommendation, not a list from one of the sources and not legal advice.