By profession
AI in human resources and payroll
In short
- Employee and candidate data is personal data, sometimes sensitive, such as anything touching on health.
- The CNIL, France’s data protection authority, is clear: only the people in charge of personnel management may see employees’ information.
- AI is still useful for writing a job ad, a template letter or a procedure, with no names.
Job ads, welcome booklets, procedures, template letters: generative AI saves time in human resources. But an HR department handles very personal data: identity, pay, health, sanctions, applications. Here is how to keep the first without exposing the second.
The CNIL also points out that an employee can obtain, simply by asking and without giving a reason, a copy of the data about them. So write about them as if they could read you.
The rules that apply
-
CNIL, “Les règles pour la gestion du personnel” (rules for personnel management)
Only the people in charge of personnel management may see employees’ information; the employer must ensure that only authorised people have access to it. No information about an employee may be collected by a system they were not told about beforehand.
-
Data concerning health, opinions, trade union membership or sex life belong to the categories whose processing is prohibited, except for the exceptions the text provides.
-
Code du travail (Labour Code), article L1121-1
No one may restrict people’s rights and individual freedoms in ways not justified by the nature of the task to be done nor proportionate to the aim pursued.
-
Code du travail, article L1221-6
Information asked of a candidate may only serve to assess their ability to do the job offered; it must have a direct and necessary link with that job.
Three everyday situations
Sorting applications
You have received eighty CVs for a position and think of uploading them to an AI assistant to rank them.
What is the problem
CVs contain personal data of candidates who do not know their application is going to a third party. The service provider is not an authorised person in your HR department.
The right move
Ask the AI for an assessment grid drawn from the job description, with criteria linked to the job, then apply it yourself. If your company wants a screening tool, have the person in charge of the GDPR check how it is covered: contract and information for candidates.
Writing a letter about sick leave
An employee has been on sick leave for several weeks. You want an AI to help you write a letter about their return, pasting in the history of their absences.
What is the problem
The history identifies the employee and reveals information about their health, a category of data the GDPR gives special protection.
The right move
Ask for a template return-to-work letter, with no name, no precise dates and no medical reason. Then fill in the template in your HR software.
Preparing a disciplinary meeting
You are preparing a meeting before a sanction and want the AI to structure the alleged facts, taken from emails and reports.
What is the problem
The facts name people, are sometimes disputed, and also concern colleagues. They would leave the company, to people who are not authorised to know them.
The right move
Ask for an outline of a pre-sanction meeting and the steps of the procedure, then check them in the Labour Code and your collective agreement. Write up the facts yourself, in your own tools.
What you can still do without special precautions
- Writing a job ad from a job description.
- Preparing a welcome booklet or an internal procedure.
- Having a payroll or employment law rule explained, then checking it on an official source.
- Building a training plan or a blank annual review questionnaire.
Try it with a document from this profession
The demo shows, on a fictional document from this profession, what would be sent to the AI, its pseudonymised version, then two answers side by side. The documents are in French. Open the demo.
Key points
- Only authorised people see employee data; an outside AI service is not one of them.
- Health information is especially protected data.
- Ask for templates and grids; apply them yourself to the real files.
- No information about an employee is collected by a system they were not told about.
Sources
- CNIL, Les règles pour la gestion du personnel (17 August 2023) (in French), consulted .
- CNIL, Règlement européen sur la protection des données, chapitre 2 (article 9) (in French), consulted .
- Légifrance, Code du travail, article L1121-1 (version in force since 1 May 2008) (in French), consulted .
- Légifrance, Code du travail, article L1221-6 (version in force since 1 May 2008) (in French), consulted .
This page is for information only. It is not legal advice and does not replace a professional’s view of your situation. Rules change: each source is dated with the day it was consulted.