Free resource · Word + Excel kit

AI kit and social dialogue of the sponsor

The method for deploying an AI project without suspension or mistrust: define the project by answering eight questions, establish a timeline for consulting the Works Council, compile the required documentation for the expert, and negotiate a procedural agreement. Updated to include the seven court decisions of 2025-2026 and the amended AI Act of 2026.

📄 Word, 37 pages📊 Excel Workbook, 6 tools⚖️ Case law 2025-2026🆓 Free

Download the kitView the decisions

The scene has become commonplace by 2025. A generative AI assistant project is ready, the licenses are purchased, a pilot program has been running for three months across an entire department, and widespread deployment is planned. Then the Works Council (CSE) takes the matter to court, and the judge suspends the tool until the end of a consultation process that never even began, with a penalty of 500 or 1,000 euros per day. This is almost never because the elected representatives are hostile to AI. It's because no one on the project side knew that the pilot program was already a deployment, that the charter authorizing ChatGPT was already an introduction of new technology, and that the documentation submitted to the committee had to be written and precise.

This toolkit helps prevent such situations and ensures that social dialogue, when conducted effectively, becomes what it is: an accelerator of adoption. It treats the consultation with the Works Council (CSE) as a project milestone, with its own deadlines, deliverables, and stakeholders, not as a mere formality to be rectified afterward. It focuses on the decisions the commissioning body actually makes: defining, planning, documenting, discussing, and monitoring, because these decisions determine the procedure's validity, well before any formal arguments are presented.

1. The Works Council has become the obligatory step for AI projects

Until 2024, consulting the Works Council (CSE) on a new digital tool was routine. Within eighteen months, judges in summary proceedings had classified artificial intelligence as a "new technology" within the meaning of Article L. 2312-8 of the French Labor Code and suspended, under penalty of a fine, any deployments undertaken without prior consultation. An appeals court upheld this position in May 2026. The toolkit includes a fact sheet for each of the six decisions below, and a boxed section for the French Supreme Court's (Cour de cassation) ruling of March 18, 2026; here are the key points.

Decision	What she said	Sanction
Nanterre Court of Justice, February 14, 2025	A pilot phase where employees actually use the tools is already a deployment: consultation must precede the pilot.	Suspension until the end of the consultation
TJ Créteil, July 15, 2025 (Moniteur Group)	An AI-based editorial assistant and a charter authorizing ChatGPT constitute a new technology affecting the working conditions of journalists.	Suspension, penalty of €1,000 per day, provision of €5,000
TJ Paris, September 2, 2025 (France Télévisions)	Updating an HR chatbot does not require consultation; the generative AI platform, initially launched with 800 pilots and then rolled out more broadly, does, although its impact "is not precisely known at this stage".	Suspension of the platform
Nanterre Court of Justice, January 29, 2026	Replacing one talent management tool with two AI-integrated tools, generalized to all employees, is a new technology: the criterion is the effect on work, not the technical novelty.	A penalty of €500 per day
TJ Paris, February 10, 2026	A four-month, voluntary Copilot 365 test, with no demonstrated impact on health or organization, is not an "important project": the expertise voted on by the CSE is cancelled.	Decision in favor of the employer
CA Paris, May 21, 2026 (Moniteur Group)	Confirmation on appeal: the pre-existing use of the tools or their optional nature does not exempt from consultation; authorizing and framing ChatGPT by a charter is to introduce it.	Penalty and provision confirmed

The toolkit derives four rules of conduct. The Works Council (CSE) must be consulted before any deployment: a pilot program that equips an entire department, is implemented over time, or impacts working conditions already constitutes a deployment (Nanterre Court of Appeal, February 14, 2025), and only a truly voluntary test, limited in time and scope, can precede it (Paris Court of Appeal, February 10, 2026). Replacing a tool can constitute the introduction of new technology when its scope, the number of employees using it, or its uses expand. Authorizing and regulating a tool that employees have adopted on their own is to introduce it. And the evaluation criterion is the effect on working conditions: the number of employees involved, the jobs affected, the organization, and the duration.

2. The legal framework in ten minutes

The French Labor Code establishes three levels of employee representative involvement, and confusing them is the primary source of disputes: information provided in a meeting is not a consultation, and negotiations with unions do not exempt employees from consulting the Works Council (CSE). The toolkit details the relevant legislation, the stakeholders to be involved (Works Council, Central Works Council and establishment Works Council, Health, Safety and Working Conditions Committee (CSSCT), unions, employees, Data Protection Officer (DPO)), deadlines and expert assessments, and what the AI Act adds following Omnibus Regulation (EU) 2026/1744, adopted on July 8 and published on July 24, 2026.

Diet	When	Basis	What this implies
Prior information	Methods of recruitment assistance; automated personnel management processes and their modifications.	Article L. 2312-38, paragraphs 1 and 2	Inform the CSE before use, in a meeting and in writing; no opinion to be obtained.
Information and consultation	New technologies; significant changes in working conditions; effects on staffing levels, employment and training; means of controlling activity.	L. 2312-8, II, 1°, 3° and 4°; L. 2312-38, paragraph 3	Written and precise file, meetings, written questions and answers, one month deadline (two with expert, three in multi-establishment), opinion, reasoned response.
Negotiation	Methodological agreement on the content, modalities and deadlines of ad hoc consultations; broader AI agreement.	L. 2312-55; L. 2242-20	Signed with the union representatives or, failing that, with the majority of the CSE; it is binding on the consultations it organizes.

Regarding the AI Act, Regulation (EU) 2026/1744 postponed until December 2, 2027, the obligations of deployers of high-risk systems listed in Annex III, including all HR systems (recruitment, evaluation, assignment, monitoring) and the requirement to inform employee representatives before implementation (Article 26, paragraph 7). It did not postpone what has already been in effect since February 2, 2025: the obligation to promote the development of staff proficiency in AI (Article 4, which the omnibus bill reformulated as an obligation of means) and the prohibition of emotion recognition in the workplace (Article 5). The toolkit translates each provision into a consultation argument or an agreement clause.

3. Qualify your project: eight questions, four levels

Qualification is the step that clients most often skip, yet it determines everything: the applicable regulations, the timeline, the expert budget, and the tone of the relationship with elected representatives. Eight questions, asked in order (will employees use the tool, does it process employee data for evaluation or monitoring, is it new or expanded, is it a simple update, what is the workforce size, what are the effects on the organization and skills, is it formalizing an existing practice, is there an agreement in place?), lead to one of four levels. The "Self-Assessment" tab in the Excel workbook calculates the level and the default legal deadline.

Level	Examples	Procedure	Indicative duration
1. Information	Update without functional changes; technical tool without employee users; recruitment assistance method.	Written note, item on the agenda, mention in the minutes.	Two to four weeks
2. Consultation	Generative AI assistant deployed to a population without processing individual data; generalization of a pilot; charter governing tools for the general public.	File, one or two meetings, opinion, reasoned response.	Month €320.00
3. Consultation with probable expertise	HR tools, job transformation, large workforce, individual performance indicators.	Enhanced file, expertise financed 80% by the employer, training plan, AIPD, DUERP.	Two months (three in multi-establishment settings)
4. High Risk AI Act	Recruitment, selection, promotion, dismissal, task allocation, monitoring and evaluation of individuals.	Level 3, plus register, human control, informing workers before commissioning, logs.	Bonds due on December 2, 2027, to be anticipated

4. The textbook case: a well-executed Copilot pilot

The decision of the Paris Judicial Court of February 10, 2026, is the most useful for organizations deploying Microsoft 365 Copilot. A four-month trial with volunteers, integrated into existing licenses, without any demonstrable impact on health or the organization, is not a significant project warranting expert review; a productivity gain does not automatically generate a psychosocial risk. The lesson is not that a pilot program eliminates all responsibility: it is that a truly voluntary trial, limited in duration and number of participants, without individual evaluation data, announced to elected representatives before its launch and followed by a shared assessment, can precede the consultation, which then focuses on the generalization. The toolkit makes this point in Article 7 of the methodology agreement and includes a checklist of four rules.

5. The typical timeline and the file that the expert expects

The kit's reverse timeline starts from the targeted deployment date and works backward: scoping and qualification at D-120, informal contact with the CSE secretary at D-110, finalized file at D-95, summons and submission of the file at D-90 (the two-month period begins from this submission), first meeting at D-85 with possible appointment of the expert, expert report at D-45 (fifteen days before the deadline, article R. 2315-47), advisory meeting at D-30, reasoned response at D-25, individual information and training for employees from D-20 to D-5, deployment at D0, monitoring committees at D+30 and D+90. The "Reverse Timeline" tab of the binder recalculates each milestone according to the timeframe.

The information and consultation file is the key document: the law requires "precise and written information," and the judge verifies that this information enabled the committee to formulate an informed opinion. The kit offers a framework in twelve sections (context, tool, scope, schedule, data, organization and workload, skills and training, employment, health and working conditions, activity monitoring, support and rules of use, governance and follow-up), with, for each section, the questions the expert will ask, followed by the wording of the agenda and a template for the reasoned response to the committee's observations.

Diet	Default legal deadline	Submission of the file	Advisory meeting	Launch preparation
Level 2, without expertise	One month (R. 2312-6)	J-45	J-15	J-90
Level 3, with expertise	Two months	J-90	J-30	J-120
Central works council and establishments with expertise	Three months	J-120	J-30	J-150
Current Methodology Agreement	Conventional time limits (L. 2312-55)	According to the agreement	According to the agreement	According to the agreement

6. The methodology agreement and the FAQ for elected officials

Article L. 2312-55 of the French Labor Code allows for the establishment, by agreement, of the content, procedures, and deadlines for ad hoc consultations. This is the ideal vehicle for moving beyond a case-by-case approach, and this is what AXA France, BPCE, MetLife Europe, and Prisma Media did in 2025: company agreements mentioning AI doubled between 2024 and 2025, from approximately 115 to 215 texts, according to the CEET (Center for Economic and Social Studies) of the Cnam (National Conservatory of Arts and Crafts). The template kit comprises twelve articles: shared principles, joint AI committee, four-level qualification grid, standard file content, contractual deadlines per level, framework for pilot projects, training for elected representatives and employees, individual guarantees, annual monitoring and review, and review before December 2, 2027. It is based on the agreements discussed but does not reproduce their text verbatim, and it should be reviewed by an employment lawyer before signing.

The FAQ has two parts. Twenty questions from employee representatives (when should we be consulted, does a pilot count, what documents should be required, what is the timeframe, who pays the expert, what if management deploys without us, can the tool monitor us, what changes with the AI Act, can an employee refuse, can we negotiate an agreement) that the client can submit as is to their Works Council: this gesture of transparency defuses many tensions. And twelve questions from the client: Is Copilot a new technology, can I launch a pilot before the consultation, my software vendor has added AI to an existing tool, is the charter sufficient, can I deploy during the consultation, the Works Council doesn't give an opinion, how can I legally shorten the deadlines, what Works Council is required in a company with multiple establishments, what risks does the manager face?

7. What the kit contains

The Word document has 37 pages in eleven parts: the user guide; why the CSE has become mandatory (the seven decisions, what a forgotten consultation costs, three misconceptions, social dialogue as an accelerator); the legal framework in ten minutes (three regimes, key texts applied to an AI project, stakeholders, deadlines and process, AI Act after the omnibus); the qualification (decision tree, four levels, Copilot case study, four rules); the typical schedule (reverse planning, three variants, what runs in parallel); the information-consultation file template (twelve sections, agenda, reasoned response); the method agreement template (twelve articles and three annexes); the FAQ (twenty questions from elected representatives, twelve from the commissioning body); six case law summaries; checklists and five pitfalls; the public sector; annotated resources.

The Excel workbook includes the tools in six tabs: the self-diagnosis in fifteen questions, which calculates the level of procedure, the default legal deadline and the parallel obligations to be undertaken; the reverse planning, which recalculates each milestone from the targeted deployment date and the deadline regime; the register of AI projects and systems with their social dialogue status; the impact analysis grid in seven dimensions and twenty-four criteria; the log of questions and answers with deadlines; the dashboard of indicators of the monitoring committee.

The kit, part by part: what each part contains and what it's for. The templates to fill out are in the Word document and the Excel workbook.

Part 1. Why the Works Council has become the obligatory step for AI projects

What a missed consultation costs

The direct cost: a suspension of deployment obtained through summary proceedings for manifestly unlawful disruption (Article 835 of the French Code of Civil Procedure); a penalty of €500 to €1,000 per day; an advance payment to the Works Council (€5,000 in the Groupe Moniteur case); and the offense of obstruction, punishable by a €7,500 fine (Article L. 2317-1). The indirect cost is heavier: three to six months of delay, expiration of licenses, and a Works Council wary of the next project.

Part 2. The legal framework in ten minutes

Key texts applied to an AI project

Each text is translated into practical consequence: an incomplete file allows the CSE to refer the matter to the court and extend the deadline (L. 2312-14 and L. 2312-15); the expert report (L. 2315-94, 2°) can be challenged within ten days (L. 2315-86); articles L. 1222-3 and L. 1222-4 require individual information for employees, L. 1121-1 the proportionality of any activity monitoring, R. 4121-2 the updating of the DUERP, and article 35 of the GDPR a DPIA for the evaluation of employees.

Who needs to be informed or consulted

From 50 employees onwards, the Works Council (CSE) has full consultative powers; from 11 to 49 employees, it does not have these powers regarding new technologies, but individual information, the GDPR, and the AI Act apply. With multiple establishments (Article L. 2316-1), a project without local adaptation measures falls solely under the jurisdiction of the central Works Council; otherwise, it falls under the central Works Council and then the establishment Works Councils. The Health, Safety and Working Conditions Committee (CSSCT) prepares the health component without issuing an opinion.

Timeframe and process of a consultation

The time limit begins from the date the information is communicated or made available in the BDESE (Article R. 2312-5). The expert submits their requests within three days of their appointment, the employer responds within five days (Article R. 2315-46), and the report is submitted fifteen days before the expiry of the time limit (Article R. 2315-47). Four conditions for a proper consultation: it precedes the decision and any deployment, including a pilot program that already has its effects; it is based on written and precise information; the review period is sufficient; and the employer provides a reasoned response before making a decision.

Part 3. Qualifying your project

The decision tree in eight questions

No employee user: Level 1, documented qualification. Employee data processed for recruitment, evaluation, assignment, training, or activity monitoring: Level 3 minimum, Level 4 if the function falls under point 4 of Annex III of the AI Act, Data Protection Impact Assessment (DPIA) mandatory. New or expanded tool, or existing use formalized by a charter: Level 2 minimum. Simple update: Level 1, with an explanatory note. Existing agreement: its framework applies, and in case of doubt, the most protective level applies.

Part 4. The standard calendar

Reverse planning from D-120 to D+90

Each milestone has its stakeholders and deliverables: qualification form, minutes of the meeting with the works council secretary, complete file, dated written responses, expert report, opinion, reasoned response, memo to employees, monitoring committee dashboard. Cardinal rule: never shift J0 (day 0) to the left; if the schedule requires earlier deployment, the deployment must wait, not the consultation.

What needs to run in parallel

Seven areas of focus: the DPIA with the DPO; the update of the DUERP (mental workload, intensification, isolation, loss of meaning, dependence on the tool); registration in the register and the AI Act classification; the rules of use, written before the consultation and attached to the file; the training plan, managers first; individual information for employees; the clauses of the supplier contract on data, hosting, reversibility and the technical documentation that the expert will request.

Part 5. Sample Information and Consultation File

The twelve sections of the dossier

Allow fifteen to twenty-five pages for a level 3 project. Each section lists the questions the expert will ask, for example: is the project linked to a staff reduction objective? what activity monitoring functions does the tool contain, even when deactivated? Avoid attaching the supplier's sales presentations: elected officials read them as such.

Impacts on health, safety and working conditions

This section lists psychosocial risks (workload, loss of autonomy, dependency, deskilling, isolation, perceived surveillance) and undetected errors, then preventive measures, the right to make mistakes during skills development, updating the Single Risk Assessment Document (DUERP), and the role of the Health, Safety and Working Conditions Committee (CSSCT) and occupational health services. The expert will ask how psychosocial risks will be measured before and after implementation, and who is responsible for any errors produced by the tool.

Monitoring activity and using indicators

What indicators does the tool produce, individual or collective? Who has access to it, for what purposes, and what is prohibited: evaluation, sanction, individual comparison? If the tool does not produce any individual indicators, this must be explicitly stated in the documentation. The expert will ask if a manager can see who uses the tool and how many times, and if usage data can be included in the annual performance review.

Agenda item

The wording defines the legal scope of the consultation: too vague, it risks being challenged; too narrow, it will necessitate a second consultation. It is jointly determined by the chair and secretary of the Works Council. The kit provides two versions, one for the information meeting (tool, function, scope, articles covered, schedule, potential expert), and the other for the advisory meeting.

Template for a reasoned response to the observations of the Works Council

Required by Article L. 2312-15 before any decision is made, this written and dated response is also the best protection in the event of a dispute: it proves that the consultation had a real purpose. The template has five parts: a summary of the consultation, point-by-point responses, management commitments, rejected observations and their reasons, and the decision and timeline.

Part 6. Model agreement on the methodology of "social dialogue and artificial intelligence"

An agreement does not replace a consultation

The agreement is concluded with the representative trade unions (Article L. 2232-12) or, in the absence of a union representative, with a majority of the elected members of the Works Council (CSE). It organizes future consultations but does not constitute consultation on the projects themselves: each project from level 2 to 4 remains subject to the opinion of the Works Council, and an agreement claiming to waive this requirement would be invalid.

Article 2. Shared Principles

AI systems assist work and do not make decisions on their own that have an individual effect; employees know when they interact with them; usage data is not used for monitoring or evaluation, except when consulted on a specific control device; no practices prohibited by section 5 of the AI Act are deployed; biases in HR systems are controlled; a right to make mistakes covers the appropriation period.

Article 6. Procedures and time limits for consultations

The model proposes one month and two meetings at most for level 2, two months (three with central and establishment CSE) and three meetings for level 3, three months for level 4. The committee transmits its questions within five days following each meeting, the company responds in writing within five days, and everything is recorded in a log of exchanges.

Article 7. Experiments

An experiment may precede consultation on a level 2 or 3 project, provided five cumulative conditions are met: voluntary participation with the possibility of withdrawal at any time, a capped duration (four months in the model), a capped number of participants, no individual evaluations, and no organizational changes. The AI ​​committee is informed before launch; level 4 projects are excluded. The evaluation (Appendix 3: framework, uses, effects on work, incidents, training, recommendations) precedes any expansion, which is itself subject to consultation.

Article 9. Individual Guarantees

Before implementation, each employee concerned receives written information: purpose, data concerning them, indicators and recipients, and appeals. They can request a clear explanation of a decision made using a system and have it reviewed by a person. Any good-faith report of a malfunction, bias, or non-compliant use is protected.

Part 7. FAQ

Can an employee refuse to use the tool?

In principle, an employer can mandate the use of a regularly deployed tool, subject to training and adaptation to the job (Article L. 6321-1). Refusal is not considered misconduct if the tool was deployed without consultation or individual notification, or if it disproportionately infringes upon the employee's rights. The right to make mistakes and the option to not use certain functions are guarantees that should be negotiated.

Part 8. Case Law Summaries

Also of note: Court of Cassation, March 18, 2026

The Social Chamber of the Court of Cassation ruled on March 18, 2026 (No. 23-22.270) that, when a technological project is accompanied by a collective dismissal with a job protection plan, the expert assessment falls exclusively under Article L. 1233-34 and not under Article L. 2315-94.

Part 9. Checklists and pitfalls to avoid

Before, during and after the consultation

Before: Project qualified, initial approval (J0) by the management committee, complete file, no access outside of a structured trial, budgeted expertise. During: Project presented by its sponsor, not just by HR, dated written responses within five days, no launch communication before the review. After: Reasoned response, individual information, monitoring committees at one and three months, then annual review, requalification for any extension.

The five traps that cost him a suspension

Deploying surreptitiously with the intention of regularizing the process later; dismissing AI as "simple software" or an "update"; invoking strategic urgency, which no decision has acknowledged; disguising a deployment as a pilot program; or formalizing existing practices through a charter. For each approach, the toolkit outlines the necessary steps, such as integrating the consultation period into the timeline from the outset.

Part 10. Public Sector

The bodies and their powers

Since January 1, 2023, the social committees at the administrative, territorial, and establishment levels have replaced the technical committees and the CHSCTs (Health, Safety and Working Conditions Committees). Negotiations for a framework agreement on AI across the three branches of the civil service have been open since June 18, 2026. As of December 2, 2027, public bodies deploying a high-risk system will be required to analyze its impact on fundamental rights (Article 27 of the AI Act).

8. Frequently Asked Questions

Is Copilot or ChatGPT Enterprise a "new technology" that requires consultation with the Works Council?

As soon as employees started using it in their work, judges treated it as such: Groupe Moniteur for ChatGPT, France Télévisions for a generative AI platform. The Copilot decision of February 10, 2026, did not contradict this; it ruled that a voluntary four-month test without demonstrated impact was not a significant project justifying an expert assessment. Assume that the widespread adoption of a generative AI assistant is subject to review.

Can I launch a pilot project before consulting the Works Council?

Only if it is a truly voluntary test, limited in time and scope, without individual evaluation data, without organizational changes, announced to the Works Council before its launch, and followed by a shared review (Paris Court of Appeal, February 10, 2026). The Works Council must be consulted before any deployment, and a pilot program that equips an entire department, is implemented over time, or affects working conditions already constitutes a deployment (Nanterre Court of Appeal, February 14, 2025).

Employees are already using ChatGPT: is a charter sufficient to regularize the situation?

No. The Paris Court of Appeal ruled on May 21, 2026, that authorizing and regulating ChatGPT through a charter constitutes its introduction into the company, and that pre-existing use or its optional nature does not exempt users from consulting with the committee. Present the charter, the secure tools offering, and the training as a single project submitted to the committee.

What happens if the CSE does not issue an opinion within the deadline?

Upon expiry of the statutory or contractual deadline, the committee is deemed to have been consulted and to have issued a negative opinion; the employer may then decide after responding in writing to the observations made. Verify that the deadline has indeed begun: it only starts running from the date of submission of complete information.

The works council is requesting an expert assessment: who pays, and can I contest it?

The employer finances 80% of the cost and the Works Council (CSE) 20% from its operating budget, unless the budget is insufficient. The employer may challenge the necessity, the choice of expert, the cost, or the duration before the judicial court within ten days; this recourse is successful when the project has no significant impact on working conditions, and unsuccessful when it transforms jobs.

How to shorten delays without taking risks?

A complete file from the first meeting, written responses within five days, an interim meeting, and a methodological agreement based on Article L. 2312-55, which sets shorter deadlines in exchange for a standardized file, a framework for experimentation, and a monitoring committee. Conversely, an incomplete file or a pilot launched prematurely will delay everything.

9. Going further

This kit completes the AI Act kit from the sponsor, which addresses AI literacy, transparency and high risk, and the sponsor's GDPR kitThis covers impact assessment and informing individuals: the three are read together, and the information and consultation file in this kit is designed to accommodate the conclusions of the GDPR impact assessment. AI charter template provides the rules of use that the file and the agreement presuppose. Organizations that want to prepare for consultation on an ongoing project, or bring together management, managers, and employee representatives around a real-world case, entrust this work to a half-day workshop: this is the purpose of the mission. Governing your agents and AI training in companies, the foundation of which includes informing employees and elected officials.

Read also : AI Project Sponsor Security and Data Kit, Sponsor's AI Act Kit: The 2026-2027 Roadmap, GDPR Kit for AI Project Sponsors, AI Charter in Business: The 2026 Model (Downloadable) et AI Manager's Kit.

The entire series of free kits. To drive adoption: AI charter template, AI training plan template, AI Ambassadors Kit, Manager's toolkit for dealing with AI, Adoption Roadmap Kit, AI maturity grid, AI Measurement and ROI Kit, AI Governance Kit, Gift Sets Cowork, Copilot Studio kit, Claude Skills kit et Vibe coding kitFor the AI ​​project sponsor: AI Act kit, GDPR kit, Security and data kit, Shadow AI kit, frugal AI kit, AI Sovereignty Kit et responsible AI kitBy profession: AI HR kit, AI marketing kit, AI finance kit, AI kit for lawyers, AI kit for salespeople, AI toolkit for project managers, AI innovator kit et communicator's AI kit.

10. Sources

The positions cited on this page and in the toolkit are based on the following dated sources. The decisions are cited from the commentaries of referenced lawyers and legal experts; the text of the Paris Court of Appeal's ruling is available online.

• Y. Bouaïcha, Consultation with the Works Council and artificial intelligence tools: what the judges decided in 2025-2026, Village of Justice; CA Paris, May 21, 2026, n° 25/13232, Papers Justice.
• CMS Francis Lefebvre Avocats, AI suspended: Judge demands consultation with the Works Council before any deployment ; Artificial intelligence: forcing consultation with the Works Council ; AI deployment: when effects matter more than technology ; Copilot 365 is not a major project.
• Lamy Liaisons, Decision of the Nanterre Judicial Court of February 14, 2025 Capstan Avocats, AI: Towards a nuanced approach to the obligation to consult the Works Council.
• European Union, Regulation (EU) 2026/1744 amending Regulation (EU) 2024/1689 on artificial intelligence, adopted on 8 July and published on 24 July 2026; Labour Code, Articles L. 2312-8, L. 2312-14 to L. 2312-16, L. 2312-38, L. 2312-55, L. 2315-80, L. 2315-94 and R. 2312-6, Digital Labor Code.
• IRES, Social dialogue, collective agreements and artificial intelligence, July 2025; N. Greenan, S. Napolitano and J. Pillosio, CEET (Cnam), working paper no. 220, September 2026 (analysis: Mister AI Observatory) ; Lexia Conseil, Artificial intelligence: what the first company agreements stipulate ; HR mind, An agreement on social dialogue and AI within AXA.
• LaborIA (Ministry of Labour and Inria), Guide to deploying AI at workMarch 2025; DIAL-IA, for social dialogue in service of the good uses of AI CFE-CGC The AI ​​Guide 2026 ; CESE, Artificial intelligence, work and employmentJanuary 2025; Lefebvre Dalloz, 8 questions and answers about the role of the Works Council in the face of AI, October 2026.
• Directorate General for Administration and the Civil Service, the negotiation of a framework agreement on AI in the civil serviceJune 2026; DLA Piper, The EU Quality Jobs Act, 2026.