France Travail is developing a generative artificial intelligence (AI) tool for the control of registered individuals, a project announced on June 30 at its Ethical Advisory Committee.

The short-term goal is to use AI to analyze the files of controlled individuals, including their situation and interview reports, in order to produce a summary and "points of attention." This automation aims to reduce human processing time to reach a target of 1,500,000 controls in 2027, an increase of 400,000 compared to 2026, without additional staff. The final sanction decision would remain human. By 2030, the project envisions a "systematization of controls" where AI could select individuals, interact with them, and decide to forward files for sanctions.

The tool could also be used for "pathway monitoring" to verify the obligation of 15 weekly "hours of activity" stemming from the Plein Emploi law. The project presentation document raises several risks, including the AI's "black box" operation making decisions unexplainable, automation bias among human agents, reproduction of discriminatory biases, and the possibility of centrally and opaquely modifying control policies through simple changes in instructions given to the tool.