10 min read

Works council and AI: consult before deploying

A works council (CSE) must be consulted before an AI tool is deployed in firms of 50+ employees: obligations, timelines, the file, recent rulings.

GouvernanceCSEConformité

If your company has 50 employees or more, you must consult the works council (CSE, comité social et économique, the body that represents employees) before deciding to introduce an AI tool that changes working conditions or makes it possible to monitor employees' activity. Plan for a detailed information file, a period of one to three months, and the CSE's opinion before deployment.

This article presents a method and is not legal advice. The law on AI at work is evolving: have your situation reviewed by an employment lawyer before deciding. It describes French law.

What are your obligations in a company of 50 employees or more?

You must consult the CSE before deploying a tool that changes work or makes it possible to monitor employees' activity (see the five questions below), and inform employees of any system that collects information about them. At this headcount, the CSE is mandatory, as are the internal rules (règlement intérieur, the document that sets the company's rules and disciplinary sanctions), and the consultation on new technologies applies.

The thresholds to know: 11 employees for a CSE to exist (article L. 2311-2 of the Labour Code), 50 employees for the consultation on new technologies to apply and for the internal rules to be mandatory. If you have several sites, there is one CSE per site and a central CSE at company level.

Three provisions of the Labour Code frame the consultation:

  • New technologies (article L. 2312-8). The CSE is informed and consulted on "the introduction of new technologies, any major change affecting health and safety or working conditions" (text, in French). This article does not mention AI: courts are the ones applying it to AI tools.
  • Monitoring of activity (article L. 2312-38). The CSE is informed and consulted, before the decision, on the means or techniques that make it possible to monitor employees' activity. It is also informed about automated personnel-management processing, meaning software that automatically processes data about employees (text, in French).
  • Internal rules (article L. 1321-4). If you add rules on AI use with sanctions to the internal rules, the CSE must give its opinion first. The rules cannot come into force before one month after the publicity formalities, and are sent to the labour inspector together with the CSE's opinion (text, in French). Internal rules are mandatory from 50 employees (article L. 1311-2).

Does your AI tool trigger the consultation?

It does if it significantly changes work or makes it possible to measure employees' activity. Courts look at the tool's effects, not at how much AI it contains.

Ask yourself these five questions. A single "yes" is enough to open the consultation:

  1. Does the tool change how employees perform their tasks, at the scale of a team or more?
  2. Does it record, measure or evaluate people's activity or performance?
  3. Does it process personnel-management data (scheduling, appraisal, recruitment, training)?
  4. Does your AI policy provide for sanctions, or is it part of the internal rules?
  5. Does the pilot (the trial phase) last several months, with employees actually using it to do their work?

A writing assistant open to all employees and a test of a few weeks reserved for volunteers are not in the same situation, but the line is blurry. If in doubt, inform the CSE before the pilot starts and ask your lawyer whether the consultation should open at that stage.

The Nanterre court (February 2025) refused to treat a pilot that employees had been using for several months as a mere test. It recalled a rule: once the consultation is open, the project cannot be implemented before the CSE's opinion.

What about ChatGPT, Claude or Copilot?

The tool itself does not matter, the use does. If the assistant is connected to your documents, given to a whole team and tracked by usage statistics, you should generally consult the CSE.

  • ChatGPT: in a case ruled on appeal, the employer had authorised ChatGPT under a policy and had deployed an internal writing assistant. The court held that it had introduced a new technology, regardless of whether employees had used it before, outside any framework. Consulting the CSE was therefore required (Paris Court of Appeal judgment, 21 May 2026, in French).

How do you consult the CSE, step by step?

Open the consultation before the decision to deploy, with a precise file, then wait for the opinion before going live. Here is the procedure, with who is responsible:

  1. List the tools and uses concerned (IT director). Validated tools, ongoing pilots, tools already in use without validation: this is the topic of shadow AI.
  2. Run each tool through the five questions above (IT, legal, DPO, the data protection officer). A single "yes" is enough to open the consultation. Keep a dated written note of your answers: it shows you examined the question.
  3. Build the information file (IT, with the DPO). Details below.
  4. Formally notify the CSE and ask for its opinion, before any go-live, pilot included if your lawyer recommends it (management or HR, supported by IT).
  5. Let the period run and answer questions (IT, HR). See the timeline table below.
  6. Collect the opinion, keep it in writing, then decide (management).
  7. Deploy, monitor, and reopen the question at each change (IT). A change of scope or functions may require a new consultation.

What does the information file contain?

A precise file is better than a general presentation: the CSE must be able to assess the tool's real effect. At a minimum, provide:

  • the tool, the vendor and the purposes;
  • the tasks and populations concerned, as a number of people;
  • the data processed and where it is located;
  • functions that record, measure or evaluate activity, if any;
  • the duration and scope of the pilot;
  • the expected effects on work (tasks, organisation, training);
  • training and human-oversight measures;
  • the DPO's opinion, if personal data is processed.

How much time should you plan for the consultation?

As a rule, plan one month, two if the CSE calls in an expert, three if several CSEs are involved: these periods run before the decision and add to your deployment schedule.

SituationPeriod for the CSE's opinion
Simple consultationOne month
The CSE calls in an expertTwo months
Several levels of CSE (for example a central CSE)Three months

If you have several sites, plan on three months in your schedule. If the CSE does not give an opinion within the period, it is deemed to have given a negative opinion. An appeal by the employer to a court may suspend these periods.

The CSE may also call on an approved expert when new technologies modifying working conditions are introduced (article L. 2315-94). Budget for the cost of this expertise in the project: how it is shared between the employer and the CSE is set by the Labour Code, and your lawyer will specify it.

We recommend building the consultation into the schedule from the scoping phase, rather than discovering it at the end of the project. It is the "governance and compliance" step of our method.

Who does what in the company?

The IT director owns the technical file, but the consultation falls to management and human resources: none of these functions can run it alone.

FunctionRole in the consultation
IT directorDescribes the tool, the data and the functions that measure activity; owns the deployment schedule
HR directorThe CSE's usual contact; prepares the presentation with the IT director
Executive managementDecides after the opinion, owns the schedule
Legal or lawyerQualifies the situation, reviews the file and the policy
DPOAssesses personal data and the need for an impact assessment
CSE (central and site-level)Issues its opinion, may ask for clarifications or an expert

The HR director is an ally: involve them before writing the file, not after.

What does a company risk by deploying without consulting?

It risks having a judge suspend the deployment, with a financial penalty (an astreinte) if it does not comply. Since 2025, three urgent-procedure rulings (référés) have suspended AI tools introduced before the CSE's opinion, one of them upheld on appeal in Paris in May 2026. They are cited from their text or from the appeal judgment.

Nanterre Judicial Court, 14 February 2025 (case no. 24/01457)

A pilot phase, used at least in part by all the employees concerned, is a first implementation and not a mere experiment. Deployment suspended until the CSE's opinion, a penalty of €1,000 per breach for 90 days, and €5,000 in provisional damages.

Takeaway: once the consultation is open, the pilot cannot precede the opinion.

Read the decision (PDF, in French)

Paris Judicial Court, 2 September 2025 (case no. 25/53278)

A platform giving access to generative AI requires consulting the central CSE, and its deployment is suspended, even if its use is optional. By contrast, no consultation for version 2 of an existing HR chatbot, which only adds topics and documents.

Takeaway: optional use does not protect you, and a minor update may go through.

Read the decision (PDF, in French)

Créteil Judicial Court, 15 July 2025, upheld by the Paris Court of Appeal on 21 May 2026 (judgment no. 25/13234)

The employer had authorised ChatGPT under a policy and deployed an internal writing assistant. This choice amounts to introducing a new technology, regardless of prior use by employees. The suspension, the penalty of €1,000 per day for three months and the €5,000 in provisional damages are upheld.

Takeaway: making a tool available to employees and framing its use is introducing it, especially if an internal assistant is deployed as well.

Read the Court of Appeal judgment (in French)

Two takeaways. First, making an AI tool available, even a consumer one, and framing its use was enough on appeal to be judged the introduction of a new technology, with the internal assistant deployed as well. Second, these rulings were issued under the urgent procedure or are an appeal of one of them: to our knowledge, the Court of Cassation had not yet ruled on generative AI and the CSE as of 10 October 2026, and judges decide case by case.

Does personal data change the procedure?

It adds a separate obligation, which does not replace the CSE's. A tool that processes employee or customer data also falls under the GDPR, the European data protection regulation, and the CNIL publishes resources on AI (CNIL AI page, in French).

Involve the DPO in the file given to the CSE. If the tool is used to measure or evaluate employees' activity, ask them whether a data protection impact assessment (DPIA) is needed.

Which cases fall outside this method?

Three situations call for a different reading:

  • Fewer than 50 employees. The company is not subject to the consultation under article L. 2312-8: prior information of employees and the GDPR apply.
  • A tool with no effect on tasks. A minor update to existing software does not necessarily justify a new consultation, but the line remains blurry: the Paris court dismissed the consultation for a new version of an existing HR chatbot that only added topics and documents (2 September 2025).
  • A group or several sites. How consultations are split between the central CSE and the site-level CSEs requires specific legal analysis.

And after the consultation?

After the consultation, you still need to deploy, train and measure adoption. This is the continuation of the approach described in our method, illustrated by our client cases. Your teams stay in control. To prepare the file for the CSE and the deployment plan together, contact us.

Sources: Labour Code (Code du travail numérique), articles L. 2312-8, L. 2312-38, L. 1321-4 · CNIL, artificial intelligence · Decisions: Paris Court of Appeal, 21 May 2026, Nanterre court, 14 February 2025, Paris court, 2 September 2025 (accessed 10 October 2026). Sources are in French.

Pick the one that fits.

Your first agent in production.

We start from your real situation, not a generic template. Free diagnostic, 45 min, concrete results on your context.

Book my diagnostic45 min, on your context. No generic pitch.

Free, 45 min, no commitment, on your real data