Use first
Describe the operational purpose, users, affected workers, outputs and consequences.
Thomas Cole, The Architect’s Dream, 1840
V. The architectBuild the framework.
Build the framework.
Before deployment.
AI systems, work organisation and occupational health.
The framework depends on the actual use, the data processed and the effects on work. The AI Act, GDPR, labour law, social dialogue and prevention duties must be considered together before deployment.
In 30 seconds
Essential points
- The same model can be low-risk for public-text reformulation and high-impact when used for recruitment, worker management or surveillance.
- EU AI rules, GDPR, French worker consultation and occupational risk prevention are cumulative frameworks.
- Recent court orders are informative but should be interpreted cautiously when issued in interim proceedings or at first instance.
EstablishedEmergingAnalysis
On this page
This English overview is informational and does not constitute legal advice. The French page contains the detailed legal references and case citations.
Method
A layered legal framework
Legal classification should not begin with the brand or model name. It should begin with what the system does in the workplace. Relevant questions include whether it processes personal or health data, influences employment decisions, allocates tasks, monitors performance, changes working conditions or produces a decision with significant effects.
Multiple frameworks
Compliance with one framework does not remove obligations under the others.
Prevention lens
A legally permitted system may still create workload, autonomy or psychosocial risks requiring prevention.
European regulation
EU AI Act
The EU AI Act uses a risk-based approach. Workplace-relevant issues include prohibited practices, transparency duties, obligations for general-purpose AI and the high-risk regime for certain systems used in employment, worker management and access to self-employment. The exact obligations and application dates should always be checked against the current consolidated text.
Employment systems
Recruitment, selection, promotion, termination, task allocation and worker evaluation may fall within high-risk categories depending on the use.
Human oversight
Oversight must be meaningful: the reviewer needs competence, authority, information and time to intervene.
AI literacy
Organisations need role-appropriate knowledge for people operating or overseeing AI systems.
Data protection
GDPR and personal data
When an AI system processes employee, candidate or patient data, the GDPR applies. Purpose limitation, data minimisation, lawful basis, transparency, retention, security and individual rights remain central. Health, biometric and inferred psychological data require particular caution.
Data protection impact assessment
High-risk processing may require a DPIA before deployment.
Automated decisions
Solely automated decisions with legal or similarly significant effects trigger specific protections and require careful analysis.
External AI services
Sending confidential or personal data to a provider creates questions about roles, retention, reuse, transfers and contractual safeguards.
French labour law
Worker information and consultation
In French companies with at least 50 employees, the Social and Economic Committee (CSE) must be informed and consulted on matters including new technologies and major changes affecting health, safety or working conditions. Monitoring tools, automated personnel management and recruitment methods require particular attention.
Timing
Consultation should occur early enough for the opinion to influence the project, not after general deployment.
Information quality
Worker representatives need the purpose, functioning, data, affected tasks, risks, safeguards, pilot results and alternatives.
Expert assistance
Depending on the legal conditions, the CSE may seek authorised expert support for a major project or new technology.
Occupational safety and health
Risk assessment and prevention
French employers must protect workers’ physical and mental health, assess occupational risks and record them in the DUERP. An AI project that changes pace, autonomy, monitoring, skills or organisation should therefore be analysed as a work transformation.
Primary prevention
Redesign the project to remove or reduce hazards before relying on individual coping or training alone.
Occupational-health contribution
The occupational-health service can help identify exposed groups, work effects, warning signs and appropriate follow-up.
Traceability
Record assumptions, consultation, testing, incidents, corrective actions and the reasons for scaling or stopping.
Courts
Emerging case law should be read cautiously
French litigation has increasingly examined whether AI pilots and tools should have been subject to worker consultation. Some interim orders have suspended deployments until consultation was completed, while earlier decisions emphasised the need to demonstrate a significant effect on working conditions. These decisions are context-specific and may be appealed.
Pilot versus deployment
A project described as experimental may already be an implementation if workers are using it in organised conditions.
Impact matters
The label “AI” does not automatically determine the outcome; courts also examine actual effects and project maturity.
Do not wait for litigation
Early consultation and documented prevention are more useful than attempting to reconstruct the process after a dispute.
Key references
- European Commission, AI regulatory framework.
- CNIL resources on artificial intelligence and personal data.
- Detailed French Labour Code articles and court references are listed in the French reference edition.
Verified legal sources
- Regulation (EU) 2024/1689 — Artificial Intelligence Act.
- Council of the EU, 29 June 2026 — final adoption of the simplification regulation.
- French Labour Code, Article L.2312-8 — works council consultation on new technologies and important changes to working conditions.
- French Labour Code, Article L.2312-38 — automated personnel management and employee monitoring.
- French Supreme Court, Social Chamber, 12 April 2018, no. 16-27.866.
- Paris Court of Appeal, 21 May 2026, no. 25/13232 — confirmation of suspension pending works council consultation.
Court decisions are fact-specific and this page is not legal advice.