AI is cutting youth hiring in France, central bank data show

Yann LeCun does not believe in the end of work, but the first French data already show a tightening at the entrance to the job market. According to 01net (14 November 2026), which draws on an interview the researcher gave to Le Point alongside the economist Andrew McAfee, Yann LeCun says that "only a small share of tasks can be automated today" and expects new occupations to appear rather than a wave of unemployment. Statistics published in October by the Banque de France, France's central bank, nevertheless tell a more nuanced story for young people. In this article, "generative AI" means software that can produce text, code or images on demand, such as ChatGPT, launched in November 2022.

What does Yann LeCun say about the future of work?

Still according to 01net, the position of the French deep-learning pioneer contrasts with the gloomiest forecasts from AI labs. Anthropic chief executive Dario Amodei considers it possible that half of junior positions in tech, law, consulting and finance could disappear within one to five years, with unemployment rising as high as 20%. Yann LeCun called that scenario "ridiculously stupid" as early as spring, 01net reports, and the interview confirms he has not changed his mind.

Yann LeCun, who is based in New York where he teaches, chose Paris for the headquarters of AMI Labs, the start-up founded after he left Meta, which is betting on so-called world models rather than on large language models.

What does the Banque de France measure among young people?

In its blog post no. 463, dated 13 October 2026, the Banque de France tracks hires and contract endings on permanent contracts (CDI) separately between January 2019 and March 2026. To identify the occupations where AI can act, the authors build an exposure index from the Répertoire opérationnel des métiers et des emplois of France Travail, which describes more than 1,500 occupations. A language model assesses to what extent AI can perform or assist each skill. The bank specifies that the index measures a potential technical capability and not actual use, while noting a high correlation with Google Atlas usage data.

The result comes down to a few figures, all taken from the Banque de France post:

• In March 2026, among 21-29-year-olds, permanent hires in the fifth of occupations most exposed to AI are about 28% below their November 2022 level, against 8% in the least exposed fifth, a gap of roughly twenty percentage points.
• The gap is almost nonexistent from age 50.
• Contract endings for 21-29-year-olds also fall, by about 23% in the most exposed occupations against 6% in the least exposed, which reflects lower labour turnover.
• Total employment at firms reporting intensive AI use does not diverge clearly from that of other firms.

To cross these data with declared use, the Banque de France follows about 6,000 firms, of which nearly 600 report intensive AI use in early 2026. At those firms, permanent hires of 21-29-year-olds have evolved less favourably since late 2022 than at others. The change therefore shows up in entries and exits more than in a visible drop in the number of employees.

01net completes this picture with other figures, which we repeat as reported: about 75,000 fewer permanent contracts a year for 21-29-year-olds, a fall of about 42% among young developers and engineers, and, according to Insee, a 7.4% drop over one year in employment of 15-29-year-olds in IT in the fourth quarter of 2025, excluding apprenticeships. The authors take care not to establish a direct causal link with AI.

Is AI really the cause of this decline?

The Banque de France itself urges caution. According to its post, monetary tightening and the post-Covid correction in the technology sector weighed on hiring from 2022, in occupations that are themselves highly exposed to AI, and remote work, which affects the same jobs, further complicates the reading. The survey also measures AI use only in early 2026, whereas the hiring gaps appear several quarters earlier: the comparison establishes an association, not yet the specific effect of AI.

The foreign studies cited by the bank are also mixed. In the United States, Brynjolfsson and co-authors (2026) observe a relative deterioration in young people's employment in exposed occupations, mainly through fewer hires. In Norway, Hernæs and Kostøl (2026) detect no relative decline, and in Denmark, Humlum and Vestergaard (2026) mostly find a reorganisation of tasks, with no average effect on hours worked or income.

What do young British job seekers say?

On 14 November 2026, The Guardian launched a series following eight young people in their job search, with four profiles in the first instalment. Aamir, 25, who holds a BA and an MA from two London universities, says he has applied for more than 1,000 jobs since graduating and lives with his parents on Universal Credit. Alice, 24, a graduate of Oxford and then Cambridge, has submitted 50 tailored applications and reached five interviews.

Peter, 21, a computer science graduate of the University of Leicester, has applied for more than 50 jobs without ever reaching the interview stage. He links this blockage to AI: "Why would you employ 10 graduates when you can have one graduate doing the work of 10 with the help of AI?" These are individual accounts reported by the Guardian: they illustrate how a generation feels but do not, on their own, prove that AI is the cause.

What this changes for a French company

The Banque de France warns that automating the tasks given to beginners can reduce learning by doing and, in time, tomorrow's human capital. It specifies that its results do not establish that this mechanism is already at work in France, but that hiring of young people in exposed occupations has begun to contract. For a French company, this suggests steering headcount not only by the number of employees but also by the pace of new entries. If beginner tasks are handed to tools, it remains to be decided how juniors will learn the trade, through apprenticeships, mentoring or supervised assignments. This reading is a line of analysis, not a finding of the studies cited.