One month, two with an expert.
Absent an agreement, the CSE is deemed to have given a negative opinion after one month; two months when an expert is involved; three months when expert assessments take place at central and establishment level.
French CSE and AI / Court decisions
CSE · AI at work · 2025–2026 case law
French court decisions, from February 2025 to September 2026, on AI tools rolled out without consulting the works council (CSE) or without assessing their health effects. Each row names its source.
The essentials
Three findings, as of .
Since February 2025, summary judges have repeatedly ordered consultation of the CSE and suspended AI tools rolled out without it. The Paris Court of Appeal confirmed this on 21 May 2026.
What triggers consultation is not the word “AI” but the project’s effect on work: a short, voluntary trial with no identified effect was not enough (Paris Judicial Court, 10 February 2026).
On 15 September 2026, an AI-based reorganisation was also suspended on the basis of the employer’s safety obligation.
The CSE (comité social et économique) is the elected employee representative body in French companies. In companies with at least 50 employees, it must be informed and consulted before new technologies are introduced. Article L. 2312-8 of the Labour Code (in French) ↗
01 / Timeline
Court, date, case number, tools at issue, outcome and source consulted.
| Date and court | Tools or project | Decision | Source |
|---|---|---|---|
| Nanterre Judicial CourtSummary order (référé)Case no. 24/01457 | Several applications with AI features, rolled out as a “pilot phase” while the CSE consultation was still under way. | Consultation required Manifestly unlawful disturbance: implementation suspended until the consultation is completed. | Order (PDF, CFDT website, in French) ↗CFDT commentary, 29 April 2025 (in French) ↗ |
| Créteil Judicial CourtSummary orders (référé)Case nos. 25/00850 and 25/00851 | An in-house editorial assistant and access to ChatGPT governed by an IT policy, in two companies of a trade press group. | Consultation required AI is characterised as a new technology likely to affect working conditions. Its use is suspended until the CSE consultation is closed. | CMS Francis Lefebvre, 4 September 2025 (in French) ↗ |
| Paris Judicial CourtSummary order (référé)Case no. 25/53278 · France Télévisions | A platform giving all employees access to generative AI; a new version of an HR chatbot. | Required for one tool of two Consultation of the central CSE ordered and rollout of the platform suspended. No new consultation for the chatbot: same technology, already submitted to the committee. | CMS Francis Lefebvre, 17 September 2025 (in French) ↗ |
| Nanterre Judicial CourtSummary order (référé)Case no. 25/02856 · CS Group France | Cornerstone Skills and Napta, skills-management software using AI, replacing Whoz. | Consultation required Opening of the information and consultation of the central CSE ordered and rollout suspended, with a penalty of €500 per day of delay. The replaced tool already used AI, which does not rule out consultation: use is extended to all employees and the data feed appraisal and staffing. | Order (PDF, CFDT website, in French) ↗ |
| Paris Judicial CourtSummary order (référé)No. 25/57412 | Copilot 365: a four-month trial with volunteers, with no rollout decision. | Consultation not required No consultation required at this stage: the judge found neither a major project nor an identified effect on jobs or organisation. | CMS Francis Lefebvre, 12 March 2026 (in French) ↗ |
| Nanterre Judicial Court | Generative AI tools. | Consultation required Duty to consult the CSE upheld. | Decision on Judilibre (in French) ↗ |
| Paris Court of AppealDivision 1, chamber 2 · two rulingsCase nos. 25/13232 and 25/13234 | ChatGPT and the in-house editorial assistant: appeal against the Créteil orders of 15 July 2025. | Consultation required · appeal Orders upheld: the suspension stands until the information and consultation procedure is completed. The fact that employees had already used these tools on their own initiative does not remove the duty to consult. | Ruling no. 25/13234 on Judilibre (in French) ↗Flichy Grangé Avocats, 4 August 2026 ↗ |
| Créteil Judicial CourtSummary order (référé)Gisi (Infopro Digital group) | A reorganisation cutting eight sub-editor posts, with about 70% of their tasks handed to AI. | Project suspended · occupational health Reorganisation suspended: its effects on workload and health had not been sufficiently assessed in light of the employer’s safety obligation (Article L. 4121-1). Resumption made conditional on tests in real conditions, with a provisional penalty of €8,000 per day of delay. | ActuIA, 17 September 2026 ↗Force Ouvrière, 9 October 2026 (in French) ↗ |
02 / Reading
What these decisions support, and the decision behind each point.
In Nanterre on 29 January 2026, the replaced tool already used AI: consultation was grounded on extension to all employees and on the use of data for appraisal and staffing. In Paris on 10 February 2026, the lack of an identified effect on jobs or organisation led the judge to rule it out.
In Nanterre on 14 February 2025, applications used by employees for several months were treated as implementation, not as a mere experiment. Conversely, a four-month voluntary trial with no rollout decision was not enough to require consultation (Paris, 10 February 2026).
On appeal, on 21 May 2026, the Paris court held that making an assistant available on the intranet was likely to affect working conditions, and that employees’ earlier use of AI tools on their own initiative did not release the employer from consulting.
In Paris on 2 September 2025, the new version of an HR chatbot, with no change of technology and no new feature likely to affect employees, did not require a new consultation. The generative AI platform opened to all employees did.
In Créteil on 15 September 2026, the judge did not rule on consultation: a reorganisation was suspended because its effects on workload and health had not been sufficiently assessed in light of the employer’s safety obligation. According to ActuIA’s report, the risk assessment document mentioned AI-related risks without rating them.
03 / Reference points
The provisions these decisions rely on, set out on the French CSE and AI page.
Absent an agreement, the CSE is deemed to have given a negative opinion after one month; two months when an expert is involved; three months when expert assessments take place at central and establishment level.
The committee may call one in when new technologies are introduced or for a major project. The cost is shared: 80% for the employer, 20% for the committee.
The summary judge can order the consultation and suspend the project. Obstructing the committee’s regular functioning is punishable by a fine of €7,500.
The employer protects workers’ physical and mental health and assesses risks; the CSE is consulted on the risk assessment document and its updates.
04 / Method
What was read, what is taken from a commentary and what these decisions do not allow us to conclude.
Each row points to the source consulted: the text of the decision where available, otherwise the commentary cited. The Nanterre order of 29 January 2026 was read in full. Case numbers and dates were cross-checked against the Éditions Tissot guide (in French) ↗ and ActuIA’s reference page ↗.
Apart from the Paris Court of Appeal ruling, these are first-instance summary orders: provisional decisions, made on the facts of each case, that do not settle the case law.
This page is an information and prevention resource. It is not legal advice and does not replace analysis by a legal professional.