Publications · Research · Occupational health
Two publications to understand AI at work.
An experimental study published in Références en Santé au Travail (INRS) and a narrative review of psychosocial risks associated with large language models. For each: the question, approach, findings and limitations.
01 / INRS · Experimental study · June 2026
Artificial intelligence in occupational health: assessing its potential to improve the precision and clarity of medical recommendations.
Study overview
The study compared the o1 model with a multidisciplinary human reference on 385 medical recommendations, assessed against five wording criteria.
Across these criteria, 74.60% of assessments agreed with the human reference; 10.54% of disagreements were judged justified on review. The findings support an aid to professional review, with performance varying by criterion.
The experiment does not demonstrate improved worker health or validate autonomous medical decision-making. Results concern one model, one prompt and a specific corpus.
02 / Narrative review · 2025
Large language models and new psychosocial challenges at work: an occupational health challenge.
Study overview
This narrative review examines possible links between large language models, changes in work and psychosocial risks, using the six dimensions of the Gollac framework.
It discusses job insecurity, deskilling and loss of meaning, cognitive overload, and changes in working relationships. It proposes organisational prevention, training and worker participation.
It is a synthesis of literature and prevention proposals, not a cohort study or an experiment establishing a causal effect on health.