HR question · AI at work
Can employers monitor how employees use AI?
Once the policy is written, the question arrives on its own: how do you check it is followed? Can you look at usage logs, measure which tools are used, reconstruct one employee's exchanges with an AI assistant? The legal answer fits in one sentence: yes, if the monitoring is announced, proportionate and consulted on — never covert.
This page separates what an employer may do from what destroys the monitoring itself. The key point is counter-intuitive: a badly installed monitoring system does not protect the company, it destroys its own evidence.
The cumulative conditions, in the EU
For any monitoring of employees' activity, data protection law imposes conditions that all apply at once:
- Purpose and proportionality: the monitoring must serve a legitimate purpose — verifying the proper performance of the tasks assigned, securing property and systems — and not impose excessive interference on rights and freedoms. Monitoring the use of professional tools is a normal counterpart of the employment contract, but not in its excess: constant keystroke-level logging, for instance, cannot distinguish professional from personal use — the textbook disproportionate device.
- Information and consultation: employees must be informed before the monitoring starts, and in most EU jurisdictions their representatives must be consulted first — in France, any employee-monitoring device requires prior works council consultation in companies of 50 or more (Code du travail, art. L.2312-38). Case law is explicit: an employer must inform and consult the works council on any device that monitors employees' activity.
- A legal basis under the GDPR: employees' usage logs are personal data, and consent is not a valid basis in employment given the imbalance of power — the employer relies on legitimate interest, tested against proportionality.
What is allowed, what destroys the monitoring
| Practice | Verdict | Why |
|---|---|---|
| Announced aggregate usage logs, named in the policy | Allowed | Legitimate purpose, proportionate, announced, consulted |
| Reading a professional tool not identified as personal | Allowed, without the employee present | Files not identified as personal are not protected (French Cour de cassation, 16 May 2013) |
| A folder or mailbox marked "personal" | Forbidden | Correspondence is protected even on a professional device |
| Device installed without information or consultation | Evidence unlawful | French case law: without information and consultation, the evidence is unlawful |
| Keystroke-level logging of everything | Disproportionate | Cannot distinguish professional from personal use |
Since 2023, covert evidence is fragile — not banned
France's supreme labour court changed course in December 2023: unlawfully obtained evidence is no longer automatically excluded — a judge may admit it if its production is indispensable to the exercise of the right to proof and the interference is strictly proportionate. That does not rehabilitate covert monitoring. A declared, non-concealed device with informed employees produces evidence that is admissible without debate; a hidden device produces fragile evidence, admissible only exceptionally, at the judge's discretion — and a March 2025 ruling shows the compliant path winning: device declared, not concealed, employees informed, evidence accepted.
Frame the monitoring, this week
If the company wants to verify usage rather than trust it, four decisions are enough:
- Declare the monitoring in the policy: what is logged, for which purpose, who has access. Employees are informed before it starts.
- Consult the works council where applicable, before the device is switched on — not after the first dispute.
- Stay proportionate: aggregate logs and alerts on sensitive data are usually enough; reconstructing one employee's exchanges must remain the exception, tied to a specific incident.
- Run a DPIA if the monitoring becomes systematic: systematic monitoring of employees is a classic high-risk trigger under the GDPR.
Monitoring AI usage is allowed, but it is not decreed: it is announced, consulted on, kept proportionate and documented — all pieces an AI framework carries naturally. LUCIDIA, the company behind this site, is run end to end by AI agents on NanoCorp.
The Eternal Intern pack, for your job
You answer six questions about your job, your tools and your most annoying tasks; you receive a ZIP by email with eight skills and prompts written for your position, plus the setup guide for your own ChatGPT or Claude subscription. €19.99, one-time, nothing subscribed here. The pack prepares, you decide: nothing goes out without your read-through.
Read next
This page supports governance work; it is not legal advice. Consult your own counsel for your specific situation.