
Yes, an employer can give their team a voice assistant. In France and under EU law, four rules apply: a specific work purpose, employees informed before the tool goes live, no permanent or systematic listening, and no inference of emotions. Everything else follows: data limited to what is needed, a documented legal basis, an impact assessment when the risk is high. This article focuses on France and the EU; other countries have their own rules. It gives general guidance, not legal advice: for your own situation, talk to your data protection officer or a lawyer.
Tuesday, 3:10 p.m., in the office behind the dish pit
Sabine runs two restaurants in Lyon. Lunch service is over and the dishwasher is still going. She has seen a voice assistant demo and likes the idea: her commis could ask for a cleaning procedure without putting down the sponge. Then Bastien, her chef de partie, leans in the doorway: "A mic on us all day? Are you going to listen to what we say in the kitchen?"
Sabine has no answer. She does not know what she is allowed to do, what she needs to put in writing, or for whom. The characters and the scene are fictional, but Bastien's question is the one every employee asks when a microphone shows up.
Why a microphone worries people more than a screen
A microphone feels like eavesdropping
You look at a screen when you choose to. A microphone seems to pick up what was never meant for it. Even if the tool does nothing of the sort, the fear is real, and it is legitimate: it is exactly what the law protects.
The law looks at use, not at the device
No text bans "microphones". The rules govern what is collected, why, for how long, and what employees know about it. The same device can be lawful or not depending on how it is set up and used.
Consent will not cover you
Getting staff to sign an "I agree" form feels safe. It is not a good basis. The European Data Protection Board, in its Guidelines 05/2020 on consent, republished by the CNIL, says employees are only very rarely in a position to freely give, refuse or withdraw consent, given their dependence on the employer. The CNIL's reference framework on HR data only accepts consent in exceptional cases.
Emotions are an EU red line
Inferring from someone's voice that they are stressed, tired or angry is not a setting to switch off: it has been banned at work since 2 February 2025.
What the texts say, in plain terms
The GDPR requires purpose, minimisation and notice
The General Data Protection Regulation (Regulation (EU) 2016/679, Article 5) sets the basics. The purpose must be specified and explicit, for instance "giving employees access to the procedures, tasks and checklists for their station". Data minimisation means collecting only what serves that purpose. Each purpose needs a legal basis (Article 6): in practice, legitimate interest or performance of the employment contract, chosen and documented. Informing people is mandatory (Article 13). Special categories of data (health, religion, opinions and so on) are prohibited in principle (Article 9).
The French Labour Code demands transparency and proportionality
Two articles cover the essentials in France. Article L1222-4: no information concerning an employee personally may be collected by a device that has not been brought to their attention beforehand. Article L1121-1: no one may restrict people's rights and freedoms in ways not justified by the nature of the task or proportionate to the aim. In French companies with at least 50 employees, the employee representative body (the CSE) must also be informed and consulted: see our dedicated article.
The CNIL rules out permanent or systematic listening
In its guidance on listening to and recording calls in the workplace (in French), the CNIL states that an employer may not set up permanent or systematic listening or recording, unless a specific law provides for it (for certain emergency services, for example). The guidance is about phone calls, but the logic applies to any microphone: occasional, justified listening that staff know about can be defended; background listening cannot.
An impact assessment when the risk is high
Article 35 of the GDPR requires a data protection impact assessment (DPIA) when processing is likely to result in a high risk. The CNIL publishes a list of processing that requires one (in French), including processing designed to constantly monitor employees' activity. Outside that list, the CNIL recommends a DPIA once two risk criteria are met. Employees are among the data subjects considered potentially vulnerable, and using a new technology is another criterion. A workplace voice assistant often ticks both boxes.
The EU AI Act bans emotion inference
Regulation (EU) 2024/1689 on artificial intelligence, Article 5, has banned since 2 February 2025 the use of AI systems to infer a person's emotions in the workplace, except for medical or safety reasons. Analysing tone, stress or mood in an employee's voice is off the table across the EU.
The CNIL white paper on voice assistants
In September 2020, the CNIL published "À votre écoute" ("Listening to you"), its white paper on the ethical, technical and legal issues raised by voice assistants. It focuses mainly on consumer assistants, but one point carries over to work: its tests found activations when no command had been addressed to the device. Hence the value of a tool that only responds when spoken to, in a session the person opens themselves.
The 8-point employer checklist
- Write the purpose in one sentence. What the tool lets the employee do, not what it lets the manager see.
- List the data processed. Voice for the length of the request, the text of the question, the answer, the identity of the signed-in person. Drop anything that does not serve the purpose.
- Choose and document the legal basis. Not consent. Record the processing in your register of processing activities.
- Make sure there is no background listening. The tool activates when the person asks, never continuously, and not in front of guests.
- Rule out any emotion or behaviour analysis, and any scoring of employees based on their questions. Get the vendor to commit to it in writing.
- Carry out the impact assessment if the criteria are met, which is the usual case, and have your DPO review it if you have one.
- Inform every employee before go-live, in writing and in plain words: purpose, data, retention period, recipients, rights, contact. In a French company with at least 50 employees, consult the CSE before deciding.
- Offer an alternative. Everything the tool does by voice should also be possible on screen, for people who would rather not talk to a machine.
What is acceptable and what is not
| Use | Position | Why |
|---|---|---|
| Asking for a procedure or your next task, in a session the employee opens | Defensible | Work purpose, voluntary activation, employee informed |
| A microphone left on to "catch problems" | Ruled out | Permanent or systematic listening, disproportionate |
| Detecting stress or irritation in someone's voice | Banned in the EU | AI Act, Article 5, since 2 February 2025 |
| Counting an employee's questions to judge their skill level | Avoid | Change of purpose, and the fear it creates kills adoption |
| Recording what an employee says about their health | Avoid | Special category data (GDPR, Article 9), unrelated to the purpose |
| Going live without telling staff first | Unlawful in France | Labour Code, L1222-4 |
Common mistakes
- Collecting signed consent and stopping there. It replaces neither the legal basis, nor the notice, nor proportionality.
- Pitching the tool as a way to "keep a closer eye on the team". You have just described a monitoring tool, with the obligations that come with it.
- Forgetting the dining room. A microphone near guests picks up third parties who asked for nothing.
- Copying the vendor's privacy policy. The employee notice is yours: you are the employer and usually the data controller.
- Treating the DPIA as paperwork. It is there to spot real risks, such as an employee dictating information about a colleague.
What to measure
- The share of employees informed in writing before their first use: the target is 100%.
- Employees' questions about the tool, received and answered, in the month after launch: they tell you what worries people.
- The date of the last review of the register and the impact assessment, to be redone whenever the use changes.
How San is designed
Tsuno San, the intelligent voice assistant for frontline teams, was designed with these rules in mind. San Voice, the voice interaction, is available as a demo; San Audio, the hands-free experience, is being explored and tested.
- A session the person opens. Each person opens their session with the location code and their PIN. San responds when spoken to.
- Work use. San gives access to the procedures, tasks, checklists and routine checks in the Tsuno environment.
- The person's permissions, nothing more. San only sees and does what the person can already see and do in Tsuno. Every action takes two steps: San prepares and summarises, the person confirms, and it is logged.
- No judgement, no emotions. The AI does not score or rank employees, and San evaluates no one. It does not try to infer anyone's mood or state.
- No special category data. Tsuno refuses to record an employee's health, pregnancy, disability, religion or opinions, and offers to log dated, observed facts instead.
We do not summarise hosting, providers or the app's data retention here: see the Trust and data page. For San's audio and text in real use, the page does not set a retention period yet: ask us in a meeting. San itself is presented on the Tsuno San page.
Further reading
Start with the pillar page, Tsuno San, the work companion for frontline teams. If your French company has 50 or more employees, read should you consult employee representatives before using a voice assistant?. On the line between support and surveillance, see why management software can quickly feel like surveillance and why AI should not judge your employees. The pillar page AI, voice and privacy at work connects this legal framework to questions of retention, access rights and trust.
Key takeaway
A microphone at work is acceptable when it serves the employee, activates when they ask, and is never used to judge them or read their emotions. Write down the purpose, inform people first, document it, and choose a tool that makes these rules visible.
Frequently asked questions
Can an employer in France give staff a voice assistant?
Yes, if the tool serves a specific work purpose, employees are informed before it goes live, data collection is limited to what is needed and the setup is proportionate. The French data protection authority, the CNIL, does not accept permanent or systematic listening to employees unless a specific law provides for it.
Do employees have to consent to a workplace voice assistant under the GDPR?
As a rule, no, and consent is a poor legal basis at work. The CNIL and the European Data Protection Board consider that employees can very rarely give free consent because of their dependence on the employer. The employer chooses another basis, often legitimate interest or performance of the employment contract, and documents it.
Can a voice assistant detect stress or mood in employees' voices?
Not in the EU. Since 2 February 2025, the EU AI Act has banned AI systems that infer a person's emotions in the workplace, except for medical or safety reasons.
Is a data protection impact assessment required for a workplace voice assistant?
It is mandatory when processing is likely to result in a high risk to people. The CNIL recommends one as soon as two risk criteria are met, and employees count as potentially vulnerable data subjects. For a new technology used by staff, an assessment is therefore very often needed.
What should the notice given to employees include?
The purpose of the tool, the data processed, the legal basis, who receives the data, how long it is kept, employees' rights and how to exercise them, and the data protection officer's contact if there is one. It must be given before the tool goes live.