Private French Firms Hire Fewer Young Workers in AI-Exposed Jobs

In late November 2022, the monthly trend for permanent hires of people aged 21 to 29 in France’s most AI-exposed occupations stood at roughly 22,000. By March 2026 it was about 6,200 lower. Hiring also fell in less exposed work, though much more slowly. The contrast is the central finding of a new HEC Paris study, in collaboration with the Banque de France. 

HEC professor Antonin Bergeaud and his coauthors Antoine Bertheau, Simon Bunel, Oscar Degos and Marco Guido Palladino tracked every hire and contract ending recorded in the study’s private-sector data from January 2019 through March 2026. Their figures cover mainland France, excluding agriculture, temporary agency workers and household employers (the scale is unprecedented, as the INSEE counted 20.96 million private-sector employees, roughly 78% of its 27.02 million salaried workers in France in mid-2026). 

The researchers sorted occupations into five groups reflecting how readily AI can perform their constituent tasks, (drawing on 16,740 skills listed by France Travail). The index accounts for the need for physical presence, human contact and experience that is hard to write down. Among the most exposed, note the researchers, are translators, journalists and bank operations technicians. A high score describes what the technology could do, not how widely employers use it.

A Slower Route Into Work

Among people aged 21 to 29, permanent hiring fell 28.3% in the most exposed group between November 2022, when ChatGPT was released, and March 2026. It fell 7.5% in the least exposed group (nurses, kindergarten teachers, qualified workers in the building sector), a gap of 20.8 percentage points. Those are changes in seasonally adjusted, smoothed monthly hiring trends, rather than the number of people employed.

Departures from permanent jobs fell too: about 23% in the most exposed occupations, compared with 6% in the least exposed. But hiring dropped further than departures. The balance between the two narrowed by around 2,700 monthly movements among young people in the most exposed group. Fewer people enter, and fewer leave. That pattern may take time to appear in overall employment figures, according to the academics.

According to Associate Professor in Economics and Decision Sciences Antonin Bergeaud there is a possible mechanism, although he admits the data cannot establish that companies made those choices because of AI: “A company that assigns some of the tasks usually given to beginners to AI need not lay people off. It can simply hire fewer people and leave some departures unfilled.” 

The researchers focused on software as a sector at the frontline of these evolutions. Permanent hires of young developers and computer engineers fell around 42%, against 11% in occupations across the other four exposure groups. Yet the study showed that removing developers from the most exposed group still leaves a 17-percentage point gap relative to the least exposed group. The divergence also survives when fixed-term contracts are included: total hires fell 22.1% in the most exposed occupations and rose 1.0% in the least exposed. The result extends beyond software and permanent jobs, though fixed-term contracts had begun diverging before ChatGPT appeared.

The age pattern is uneven. The gap in permanent hiring between the highest and lowest exposure groups is 20.8 points at ages 21 to 29, 18.6 at 30 to 39, 15.3 at 40 to 49 and 3.8 after 50. This is strongest at the start of working life, though the comparison does not isolate an AI effect from other pressures on younger recruits.

European Studies Measure Different Outcomes

So, how does France compare to other countries in Europe? The comparisons give a mixed picture. A Norwegian study covering private-sector workers through February 2026 finds no statistically clear employment gap between young people in the most and least AI-exposed occupational groups. It counts people employed, while the French study counts hiring and departures. 

Danish researchers, meanwhile linked surveys on chatbot adoption to employment records and found changes in tasks, without measurable effects on earnings or recorded hours above 2% in the first two years. Neither result measures the same French hiring flows. France’s public sector, meanwhile, poses another question. A French government report estimates that one in five public employees holds a role theoretically exposed to AI, with administrative work among the areas concerned. This sector, was outside the researchers’ hiring data, so a similar slowdown there remains possible but unmeasured.

Other Pressures on Hiring

The authors also matched hiring records with a Banque de France survey of roughly 6,000 firms. Firms reporting intensive AI use in early 2026 showed a 17% - 25% relative shortfall in permanent hiring of young people in 2025 compared with other firms in the same sector, depending on the quarter. Their total employment did not show a statistically clear difference. But the survey dates AI use only in 2026: it cannot show whether these businesses had adopted it when their hiring began to diverge in 2023.

Other explanations for these evolutions remain open. The study shows  that software hiring had surged after the pandemic before a correction and higher interest rates. Occupations exposed to AI also lend themselves to remote work, which complicates the comparison. Apprenticeships have expanded in fields such as business and computing: among people aged 21 to 29 working in the most exposed occupations, apprentices’ share rose from 6% at the end of 2019 to 10% at the end of 2023. Some young entrants could be taking that route instead of a permanent job. The authors are now examining whether fewer hires have begun to translate into fewer young people employed in these occupations.

For employers, the question remains how young people acquire the experience that exposed occupations still require. “The long-term issue is the career ladder,” explained Bergeaud. “Junior positions also help train, through practice, tomorrow’s experienced professionals.” These reflections are likely to find their way into the explorations by LIFT, a new research laboratory directed by Bergeaud, launched by HEC Paris this year in partnership with Accenture, Google, and the Région Île-de-France. Its core mission is deciphering and measuring the impact of artificial intelligence (AI) on the economy, productivity, business models, work organization, and workforce skills.