AI Meeting Notetakers in Europe: What the Law Expects

An AI notetaker takes one click to switch on and produces something genuinely useful: a transcript, a summary and a list of actions, without anyone typing. It also records people's voices, which are personal data, and creates a written record of what was said that can travel further than anyone intended.

Most of the legal questions can be answered with a short announcement and a few settings.

Recording a meeting is processing personal data
Under GDPR a voice recording and its transcript are personal data. That means you must tell participants about the recording, rely on a lawful basis, collect no more than you need and keep it no longer than you need.

French criminal law adds a separate rule: recording words spoken in private or confidential circumstances without the consent of the people speaking is an offence under Article 226-1 of the Penal Code. An ordinary business meeting that has been announced is a different situation, but that rule is the reason never to record people secretly.

Information is not the same as consent
The GDPR does not require the consent of every participant before an AI tool processes a meeting. What it does require is information, given before the recording starts: that the meeting is being recorded and transcribed, why, who will have access, how long it will be kept and how to object. The lawful basis can then be consent, or legitimate interests where you can show the recording is needed and that no lighter option, such as taking notes, would do.

With employees, consent is a weak basis, because the relationship of subordination makes it hard to say it was freely given. Employers usually rely on legitimate interests and document the reasoning. Before introducing a system that records or monitors staff, employees must be informed in advance and, where there is a staff committee, it is consulted.
An announcement that works
Put a notice in the invitation and repeat it at the start of the meeting. Something like this works: This meeting will be recorded and transcribed by an AI tool so that we can produce minutes. The transcript goes to attendees and is deleted after 30 days. Tell us now if you would rather we did not record.
If someone objects, stop the recording and take notes by hand. External participants deserve the notice most, since they have not agreed to your internal policies.
Settings that matter
Do not let the tool join every calendar event automatically. Turn it on for the meetings where it is needed.

Keep the transcript and drop the audio where you can, or set a short retention period for the audio.
Do not share transcripts automatically with everyone invited. Send them to the people who need them.

Limit who can search past transcripts.
Switch off any feature that scores participants, measures sentiment or infers emotions. Since February 2025 the AI Act has banned systems that infer emotions in the workplace, and a meeting tool that does so crosses that line.

Do not reuse recordings to evaluate people's performance, which is a different purpose from producing minutes.
Meetings you should not record
Disciplinary and dismissal meetings, conversations about someone's health, meetings covered by legal professional privilege and discussions with employee representatives need a human note-taker, not a bot. The harm from a misplaced transcript in those situations is high and the benefit is low.

What to check in the vendor
Read the data processing agreement and the sub-processor list. Ask where recordings and transcripts are processed and stored, whether the vendor uses your content to train its models, how deletion works and how quickly, and whether you can export or erase everything at the end. A tool that cannot answer in writing is not ready for business meetings.

A lighter alternative
If the aim is minutes, you may not need a recording at all. Taking your own notes, pasting them into an assistant and asking it to turn them into minutes with decisions and actions gives you most of the benefit and avoids voice data entirely. Mirage Cloud generates documents from descriptions and includes a generalist agent for this kind of task. Whichever tool you use, check the draft against what was actually decided before you circulate it.