When to consult the CSE before deploying AI (October 2026)

Update of 7 October 2026. First version of this reference page. It lists the published French decisions on consulting the works council (CSE) before an AI deployment, the timetable of the EU regulation for HR tools and the cases still pending. The log at the bottom of the page dates every change.

An employer that introduces an artificial intelligence tool altering working conditions must inform and consult its works council, the comité social et économique (CSE), before deciding, and summary proceedings judges now sanction a failure to do so by suspending the tool. The table below lists nine summary or appeal decisions handed down between February 2025 and September 2026, in six cases. In four of the five cases concerning consultation, the judge ordered it or suspended the tool until it was completed, including a ruling by the Paris Court of Appeal targeting ChatGPT. At France Télévisions, the judge required it for only one of the two tools at issue. A single case ruled it out entirely, for Copilot 365. The obligations of the European AI regulation specific to high-risk HR tools will only apply from 2 December 2027.

What changes this month

For this first edition, two recent developments. On 30 September 2026, CFE-CGC Groupe RATP filed an alarme sociale (a formal early-warning notice) for "failure to respect employee representative bodies regarding information and consultation on projects introducing new technologies in the company". On the morning of 6 October, no follow-up had been published on the union's website and no record of agreement or disagreement had been made public. According to the count by ActuIA on 5 October, the five working days given to management to draw up a record of agreement or disagreement expired on 6 October. On the European side, Regulation (EU) 2026/1744, in force since 27 July, rewrote Article 4 of the AI Act on AI literacy. The obligation to ensure a sufficient level becomes an obligation to "promote" its development (details below).

L. 2312-8: introducing "new technologies" triggers consultation

The foundation fits in one line of the French Labour Code. Article L. 2312-8 provides that the CSE "is informed and consulted" on matters concerning the organisation and running of the company. The text refers in particular to "the introduction of new technologies, any significant change affecting health and safety conditions or working conditions". The government's Code du travail numérique shows a last update of 25 August 2021, well before generative assistants arrived in companies. Judges now apply it to ChatGPT just as they do to HR software.

Three provisions set out how it works. Article L. 2312-14 states that "the employer's decisions are preceded by the consultation" of the council. Article L. 2312-15 requires a "sufficient period for review" and "precise, written information". Failing an agreement, Article R. 2312-6 sets this period at one month, extended to two months if the CSE calls in an expert, after which the council is deemed to have delivered a negative opinion.

For HR uses, Article L. 2312-38 adds prior information on recruitment assistance techniques and on "automated personnel management processing". It requires consultation before any decision on the means "allowing the monitoring of employees' activity". Each employee is also protected individually: under Article L. 1222-4, no information concerning them may be collected by a device that has not been brought to their attention beforehand.

Table of French decisions (February 2025 - September 2026)

The "Source" column links to the order when its text has been read. Otherwise, it names the commentary that reports it ("according to"). The two Paris Court of Appeal rulings decide the appeals against the two Créteil orders of 15 July 2025: they form a single case.

Paris Court of Appeal, 21 May 2026: a consumer tool comes into scope

The two rulings of 21 May decide the appeals against the orders handed down in Créteil on 15 July 2025. Two subsidiaries of the same trade press group had given their employees access to an internal writing assistant and authorised ChatGPT through an IT charter, without consulting their CSE, according to Village-Justice. One of them is GISI, a subsidiary of Infopro Digital, and the assistant is named DIGI, according to ActuIA. According to the elements of the decision reported by ActuIA in June, these tools were used to transcribe audio or video content, summarise documents, correct or rephrase texts and suggest headlines. The court saw this as a new technology likely to affect working conditions. It upheld the suspension pending consultation and the interim payment granted to the councils, and added 3,000 euros per case in legal costs, according to Village-Justice.

The significance of these rulings lies less in the internal tool than in ChatGPT. A consumer service thus joins business software within the scope of Article L. 2312-8. For an IT department rolling out licences for a general-purpose assistant, the optional nature of its use did not remove the need for consultation in this case: the court found it irrelevant, according to ActuIA's report of 5 October. It is also the only case in the table decided on appeal, and these summary decisions leave each judge to assess the deployment brought before them.

Pilot, test, licence: where judges draw the line

The pilot scenario was decided in the Nanterre order of 14 February 2025. The employer argued that the five tools were "being trialled". The judge noted that an operations director had opened access to Finovox to all his subordinates by email on 20 November 2024, and that a director had mentioned training her teams on Synthesia. This phase, the judge wrote, "cannot therefore be regarded as a mere experiment", but "on the contrary amounts to an initial implementation". The early deployment constituted "a manifestly unlawful disturbance". The order adds that once consultation has begun, "whether mandatory or not", the employer cannot implement the project before the council has given its opinion.

Créteil clarified the timing. The CMS Francis Lefebvre note of 4 September 2025 quotes this reasoning: "Where the measure is part of a complex procedure involving staggered decisions, the council must be consulted on each of them". It notes that the working groups on AI were not suspended: only the use of the tools was.

In Paris, on 2 September 2025, the order against France Télévisions rejected the optional-use argument. The MedIAGen platform, which provides secure access to generative AI tools on the market, had been used since the second quarter of 2024 by nearly 800 employees. As no consultation was established when it was first made available in 2024, the judge concluded "that this is indeed the first introduction of these new technologies". She found it "manifest" that a technology relying on artificial intelligence constitutes a new technology.

The order of 29 January 2026 concerns HR software. CS Group France was replacing Whoz, a skills and CV generation software that already had AI features, with Cornerstone Skills and Napta. The judge accepted that the new features "do not substantially exceed" the old ones. He nevertheless ordered consultation for two reasons. Use of the software, which "was not presented as optional", extended to all employees, whereas only two business units used Whoz. The data would also be used for annual reviews, appraisals and assignment to projects.

The opposite line also exists. The Paris Judicial Court ruled out consultation for Copilot 365 on 10 February 2026, according to the CMS note of 13 April 2026. In the France Télévisions order, version 2 of the Raiponse HR chatbot, whose first version had been submitted to the central CSE in 2022, did not require a new consultation. According to the judge, it "merely adds topics and documents", with no established change to the technology and no new feature likely to affect employees. The authors of the CMS note conclude that the analysis of uses and their effects on the organisation "takes precedence" over that of technical characteristics. The editorial team reads it the same way: the judge looks at who uses the tool, whether it is mandatory and what the data it produces is used for, far more than at the "pilot" or "AI" label.

Obstruction offence and summary proceedings: the employer's exposure

The immediate risk is civil. Works councils bring the matter before the summary proceedings judge on the grounds of a manifestly unlawful disturbance, and obtain a suspension, a penalty and an interim payment. In Nanterre in 2025, the council sought a penalty of 50,000 euros per day and per breach; it obtained 1,000 euros per breach recorded over 90 days. On top of this came an interim payment of 5,000 euros and 2,000 euros in legal costs. In January 2026, CS Group France was ordered to open the consultation and suspend the deployment, each injunction under a penalty of 500 euros per day of delay for up to 90 days.

The criminal risk is set by Article L. 2317-1. Obstructing the establishment of the CSE or the free appointment of its members is punishable by one year's imprisonment and a 7,500 euro fine. Obstructing its "proper functioning", which covers a failure to consult, is punishable by a 7,500 euro fine. These amounts apply to individuals: for a company, the maximum fine is multiplied by five, to 37,500 euros, under Article 131-38 of the Criminal Code. None of the decisions listed here falls under the criminal courts.

CSE expert assessment on new technologies: funding and deadlines

Article L. 2315-94 allows the CSE to call in an accredited expert "in the event of the introduction of new technologies or of a significant project altering health and safety conditions or working conditions". Funding follows from Article L. 2315-80: of the cases listed in Article L. 2315-94, it places only the assessment for serious risk and, under conditions, the one on gender equality at work entirely on the employer. New technologies therefore fall under "one-off consultations": 20% from the council's operating budget, 80% paid by the employer. The employer pays the full amount when that budget is insufficient and no surplus has been transferred to social and cultural activities in the previous three years. The council may also pay alone for any other type of expert assessment (Article L. 2315-81).

Deadlines are set by regulation. The expert requests information within three days of appointment and the employer responds within five days (R. 2315-45). The estimated cost, scope and duration are notified within ten days (R. 2315-46). The report must be submitted no later than fifteen days before the end of the consultation period (R. 2315-47). An employer contesting the need for the assessment, the choice of expert or its estimated cost must refer the matter to the judge within ten days (R. 2315-49). The judge rules within ten days, with no appeal possible, and the referral suspends the consultation deadlines (L. 2315-86).

The procedural route matters. In the CS Group case, the central CSE had voted for an expert assessment on 17 September 2025 and asked the summary proceedings judge to order compliance under penalty. The judge ruled that there were no grounds for summary proceedings, as this request fell under the expedited procedure on the merits provided for by Article L. 2312-15.

Beyond consultation: risk prevention

A consultation carried through to completion does not end the litigation. On 15 September 2026, the Créteil summary proceedings judge suspended a reorganisation at GISI that cut eight subeditor positions and handed about 70% of their tasks to AI. The judge found that the project's consequences for working conditions had been insufficiently assessed, in light of the safety obligation under Article L. 4121-1. The expert assessment commissioned by the CSE disputed the time estimates for the remaining tasks. Resumption is conditional on real-world testing, under a penalty of 8,000 euros per day for three months, as ActuIA detailed in September. Management reserved the right to appeal.

AI-assisted HR decisions: Article 22 of the GDPR already applies

For tools that sort, score or exclude employees, the text enforceable today is Article 22 of the GDPR. The data subject has "the right not to be subject to a decision based solely on automated processing, including profiling", which produces legal effects concerning them or similarly significantly affects them. The exceptions (contract, law, explicit consent) require at least the right to obtain human intervention, to express one's point of view and to contest the decision.

The precedent comes from a work platform. On 24 August 2026, the CNIL reported on the 824,990,000 euro fine imposed by the Dutch data protection authority on Uber B.V. and Uber Technologies Inc. Deactivations of drivers' accounts, temporary in cases of suspected fraud, temporary or permanent in cases of low ratings, were classified as automated individual decisions "due to the total absence of human intervention". The proceedings stemmed from a collective complaint filed in 2020 with the CNIL by the Ligue des droits de l'Homme on behalf of more than 170 drivers. The authority found that these blocks affect drivers "significantly", since they can no longer take rides or earn income. The text does not distinguish by employment status: it applies equally to an employee excluded from a project or a position by a tool, without human review.

The CNIL has included in its 2026 work programme the completion of its work on AI uses in the employment sector. This work will address "the risks of algorithmic bias and the safeguards to be put in place to protect the rights of employees and patients".

AI Act: high-risk HR tools wait until 2 December 2027

Regulation (EU) 2024/1689 classifies as high-risk, in point 4 of its Annex III, systems intended for recruitment, filtering applications and evaluating candidates. The same point covers systems used to make decisions on promotion or termination, to allocate tasks based on behaviour or personality traits, or to monitor and evaluate performance. Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July, postpones the application of sections 1 to 3 of Chapter III to 2 December 2027 for these systems, as ActuIA detailed in May.

Article 26(7) is part of this postponed block. It requires employers deploying a high-risk system to inform "workers' representatives and the affected workers" before putting it into service. This information is provided "where applicable, in accordance with the rules and procedures laid down in Union and national law". In France, this procedure already exists: it is the CSE procedure, which applies without waiting for 2027.

Two provisions already apply. Article 5(1)(f) has prohibited, since 2 February 2025, AI systems used to "infer emotions of a natural person in the areas of workplace", except for medical or safety reasons. Article 4 on AI literacy was rewritten by Regulation 2026/1744. The original text required measures "to ensure, to their best extent, a sufficient level of AI literacy". The new version requires measures "to promote the development of AI literacy" and specifies that it "does not oblige" a specific level to be ensured for each individual. The obligation, applicable since 2 February 2025 as ActuIA recalled in June, remains in a lighter form.

RATP and Ubisoft Paris: two open cases

At RATP, the letter of 30 September signed by Valérie Vovk, central union delegate of the CFE-CGC, invokes Article 70 of amendment no. 3 to the agreement on the quality of social dialogue of 7 July 2022. It names no tool. ActuIA analysed this notice and the context of the workplace elections from 19 to 23 October.

At Ubisoft Paris, the Solidaires Informatique union section announced on 16 July 2026 that management was suspending the deployment of AI tools "while an expert assessment is carried out", at the request of the CSE of the Ubisoft Paris studio. The assessment covers their impact on employees' health and safety. The union criticised the company for "insistently" urging teams to use generative AI without informing the CSE beforehand. It is calling for a moratorium, a charter on tool transparency and the involvement of employees.

Checklist before deploying an AI tool

1. Assess the tool by its effects: number of employees concerned, mandatory or optional use, use of the data in appraisals, reviews or assignments.
2. Consult the CSE before any decision, including for a pilot open to an entire team, then at each stage of a phased deployment.
3. Provide a precise written file: features, scope, timetable, data processed.
4. Check Article L. 2312-38 for any recruitment, automated personnel management or activity monitoring tool.
5. Allow for a two-month period and the funding of an expert assessment on new technologies.
6. Inform employees individually of any device that collects information about them.
7. Guarantee genuine human intervention in any HR decision with significant effects (Article 22 of the GDPR).
8. Rule out any emotion inference at work, prohibited since February 2025, and take AI literacy measures (Article 4).
9. Identify tools falling under point 4 of Annex III and prepare the information required by Article 26 for 2 December 2027.
10. Keep a record of the dates of information, the documents provided and the opinion delivered: summary proceedings turn on these documents.

Update log

• 7 October 2026: page created. Nine decisions listed in six cases, including the France Télévisions order of 2 September 2025 and the suspension of a reorganisation at GISI on 15 September 2026. Also covered: RATP alarme sociale of 30 September, rewrite of Article 4 of the AI Act by Regulation 2026/1744, CNIL communication of 24 August on the Uber sanction.