It’s a subtle adjustment that doesn’t rely on layoffs.
Since the end of 2022, French companies have been hiring fewer young people on permanent contracts in roles where generative AI can more easily perform the work.
That is the finding of a joint study by the Bank of France and the Laboratory of Innovation and the Future of Work at HEC, coordinated by economist Antonin Bergeaud.
The researchers first classified occupations by their potential exposure to AI. To do this, they fed a language model the skill requirements from France Travail’s job repertoire, which describes more than 1,500 professions.
The model was designed to assess to what extent AI can perform them or assist, taking into account what hinders substitution: tacit expertise, physical presence, human interactions. This index thus measures technical capability, not actual use. It is nevertheless strongly correlated with usage data observed from Google.
The authors then confronted this index with administrative records of labor-market movements.
Among 21- to 29-year-olds, permanent-contract hires in the top 20% most exposed occupations (the quintile) stood in March 2026 at around 28% below their level in November 2022.
In the 20% least exposed, the decline is only about 8%.
Across all contracts, hires of young people fall by 22.1% in the most exposed occupations and rise by 1% in the least exposed.
The effect fades with age. It is reduced among 30- to 49-year-olds and almost non-existent from age 50 onward. Juniors thus bear most of the impact: among young developers and computer engineers, permanent-contract hires are about 42% below their end-2022 levels.
Fewer hires, fewer departures
The phenomenon does not translate into a shift toward fixed-term contracts, which also declined.
Youth are also leaving their jobs less frequently. Among 21- to 29-year-olds, contract terminations in permanent contracts are about 23% lower in March 2026 than in November 2022 in the most AI-exposed occupations, versus about 6% lower in the least exposed occupations. Employees thus stay longer in their roles and teams are renewed more slowly. In the most exposed occupations, however, hires fall more than departures (about 28% versus 23%): relatively fewer young entrants than exits.
The authors offer an explanation. A company that entrusts AI with part of the tasks for beginners does not need to lay off workers; it suffices to hire less and not replace some departures.
This margin of maneuver is all the more relevant in France, where ending a CDI is more tightly regulated than in several other countries. According to Antonin Bergeaud, AI excels at tasks easy to codify and leaving little autonomy, which corresponds to entry-level positions. Employers, he argues, avoid hiring into roles that could disappear, without discarding employees whose expertise might still be useful.
The finding is consistent with what HR directors report.
According to a survey by the National Association of HR Directors conducted in September among 220 professionals, a quarter of the responding companies have reduced or are considering reducing their recruitment of young people. A third report changes in the tasks assigned to interns, apprentices and young hires, and 9% note a significant reduction in execution tasks.
Declared adoption, a more ambiguous signal
The Bank of France has also cross-checked its data with its business-climate survey, enabling monitoring of around 6,000 firms, nearly 600 of which reported intensive AI use in early 2026.
Their hiring of 21- to 29-year-olds on permanent contracts has evolved less favorably since late 2022 than that of other firms, and the gap is much less pronounced among older workers. Their total employment does not diverge in a statistically significant way. The adjustment therefore concerns the pace and composition of recruitment, not a visible contraction in headcount.
What the study does not prove
The authors themselves urge caution.
Other forces have hit the same jobs, such as monetary tightening and the post-Covid correction of the tech sector that have weighed on hiring since 2022.
Telework, which would also have a negative effect on junior recruitment, largely concerns the same categories of jobs. These factors are not necessarily competing with AI, as they may have acted together on the same companies.
The survey also measures the intensity of use only at the start of 2026, whereas the hiring gaps appeared several quarters earlier.
Intense AI-using firms were already hiring before ChatGPT in more exposed occupations, even within the same sector. Adoption is therefore not distributed randomly. The comparison establishes an association but does not yet isolate AI’s own effect from the selection of firms that adopt it.
