Intervene when the project can still evolve.
When consultation is required, it precedes the decision. The CSE can request information and safeguards while the choices remain open.
French CSE · AI at work · Prior assessment
The French CSE and AI at work: assess a project before issuing its opinion
A practical framework to help the French Social and Economic Committee (CSE) obtain the information it needs, assess how AI may affect work and identify missing safeguards.
10 essential questions7 review stagesNo employee scoring
Why it matters
Why should the CSE engage with AI at work?
Direct answer
Introducing AI changes more than the tools people use. It can redistribute decision-making, objectives, review work, autonomy, skills, oversight and accountability. The CSE provides a forum in which these changes can be examined collectively before they become embedded in the organisation.
Examine actual capabilities.
An assistant can also access emails, files, data or software. The risk depends on the entire system, not just the intended use.
Make additional work visible.
Checking, correcting, handling exceptions and recovering from errors takes time. This workload must be measured and recognised.
Organise collective monitoring.
The system may evolve after launch. Incident handling, stopping thresholds and changes that trigger reassessment must be specified.
Occupational health
Can AI at work cause psychosocial risks?
Direct answer
Yes, depending on how it is used and organised. AI can contribute to work intensification, reduce scope for discretion, increase monitoring, undermine meaning at work or shift accountability. These risks must be assessed through the realities of the work itself.
- Workload, pace and interruptions
- Autonomy and scope for discretion
- Skills, quality and meaning of work
- Monitoring and evaluation of activity
- Teams and working relationships
- Occupational change and job security
01 / Minimum evidence base
Ten questions to ask before the CSE issues its opinion.
For each question, indicate whether the employer’s response is documented, incomplete or missing. The summary updates automatically.
01Purpose
What problem are we trying to solve with AI?
A clearly defined need, the alternatives considered and the expected benefit.
02Capabilities
What can the system actually do?
Its available functions — not only the stated use case.
03Decision
What decisions can it make or influence?
Distinguish between assistance, recommendation, scoring and decision-making.
04Autonomy
What actions can it perform without human intervention?
Identify the actions performed, their scope and frequency.
05Access
What data and tools does it have access to?
Employee data, emails, files, internet access and workplace software.
06Scope
Which employees and occupations will be affected?
Groups exposed, tasks transformed and situations requiring particular attention.
07Actual work
What effects are expected on work?
Workload, pace, autonomy, skills, teamwork, monitoring and employment.
08Recourse
How can a human challenge, correct or stop the system?
An identified person with the necessary time, information and authority.
09Incidents
How will unexpected effects be detected?
A reporting channel, logs, a named owner and a feedback process.
10Suspension
Under what circumstances will the system be suspended?
Written shutdown thresholds and a tested rollback procedure.
02 / Authority given to the system
Assess the system’s actual autonomy.
The higher the level, the stronger the justification, oversight, monitoring and reversibility must be.
Human control test
Who can stop the AI? Does this person actually have the time, information, skills and authority to do so?
Work and health indicator
Is the project likely to change the organisation, workload, autonomy, pace or working conditions?
03 / The full assessment framework
Seven review stages, in decision-making order.
A concise pathway that starts with the system, examines the work as it is actually done and ends with a reasoned conclusion.
Understand the system
Model, vendor, version, hosting, memory, internet, tools and agents.
See the transformed work
Workload, autonomy, skills, working relationships, assessment and employment.
Assess autonomy
From A0 information to A4 organisational delegation.
Demonstrate effective human oversight
Time, information, skills, authority and real possibility of stopping.
Prepare for incidents
Reporting, severity, analysis, suspension and rollback.
Organise reassessment
New model, new data, new tools or a new population.
Formulate a conclusion
What is documented, what is missing and which safeguards need strengthening.
Your summary
Additional information required
Some answers are still missing to understand the system and its effects.
04 / Essential answers
Frequently asked questions about the CSE and AI.
Clear answers about consultation, occupational health and the information the CSE should request.
Should the CSE be consulted before any use of AI?
An employee’s occasional, individual use of an AI tool does not in itself trigger CSE consultation. In undertakings with at least 50 employees, however, an employer-led introduction of AI may fall within the CSE’s prior information and consultation remit, notably where new technologies are introduced or the organisation and working conditions are affected. Recent decisions have required consultation before generative AI tools were introduced, including in a case where their use was presented as optional.
Can the CSE seek advice from the occupational physician during an AI project?
Yes. The occupational physician advises employee representatives on health and working conditions. When AI may change the organisation, workload, autonomy or pace of work, or expose employees to psychosocial risks, the CSE may seek the physician’s advice. The occupational physician also attends relevant CSE meetings in an advisory capacity.
Why should the CSE engage with AI projects at work?
Because an AI system can redistribute decisions, review work, autonomy, skills, monitoring and accountability. The CSE enables these changes to be examined collectively before they become embedded in everyday working arrangements.
What information should the CSE request about an AI system?
At a minimum: its purpose, actual capabilities, the decisions it influences, its level of autonomy, accessible data and tools, the employees affected, expected effects on work, human oversight, incident monitoring and suspension criteria.
Can an optional AI tool still fall within the CSE’s remit?
Yes, depending on the circumstances. Describing a tool as optional does not rule out effects on working methods, workload, organisation or skills. The project must be assessed by reference to its capabilities, scope and actual effects.
05 / Legal framework
A preparation tool to use alongside the applicable legal framework.
Not every use of AI automatically triggers consultation. The workforce, the nature of the project, its capabilities and its effects must be assessed on a case-by-case basis.
01
Labour Code · L.2312-8
↗
02Information and consultation in undertakings with at least 50 employees, notably on new technologies and significant changes affecting health, safety or working conditions.
Labour Code · L.2312-14
↗
03Where consultation is required, it must take place before the employer’s decision.
Labour Code · L.2312-38
↗
04Prior information on recruitment-assistance methods and automated staff-management processing; prior information and consultation before implementing means of monitoring employees’ activity.
Labour Code · R.4623-1
↗
05The occupational physician advises staff representatives on the prevention of risks and the improvement of working conditions.
Labour Code · L.2314-3
↗
06The occupational physician participates in an advisory capacity in meetings devoted to health, safety and working conditions.
INRS · Artificial intelligence and health and safety
↗
AI can transform workload, autonomy, skills, monitoring and meaning at work, creating psychosocial risks that need to be prevented.
Primary sources · Verified on · Legal review recommended before relying on this tool to inform a decision