← Back to the dossier

2026, the evidence · The economists

Economists see few lost jobs so far, except for young people starting out

Does AI cost jobs? So far the studies find little. Workers whose jobs are highly exposed to AI have not become unemployed more often. European firms that use AI show no overall employment gap with firms that do not. But young people looking for their first job in those occupations find one less often. The Belgian High Council for Employment sees that risk in Belgium too.

By Mara Masaeva · Updated 29 September 2026

ArgumentDevelopingThe story is still developing. There is one source so far, or the numbers are still changing.

What happened

The overall picture. Maxim Massenkoff and Peter McCrory, economists at Anthropic, found no systematic rise in unemployment for workers in the most exposed US occupations since late 2022. By their own estimate, a rise of about 1 percentage point would have been visible. An ECB survey of around 5,000 firms, discussed in March 2026 by Laura Lebastard and David Sondermann of the ECB, found no overall employment gap between firms that use AI and firms that do not. According to Futurum's summary, intensive AI users were about 4 percent more likely to add staff.

The entry level. For young people starting out, the data do show a change. In the same Anthropic study, the rate at which 22- to 25-year-olds start a new job in a highly exposed occupation fell by 14 percent compared with 2022. That is about half a percentage point on a monthly rate of around 2 percent. The authors call it “just barely statistically significant”, and they see no such fall for workers over 25. Earlier work at Stanford by Erik Brynjolfsson and colleagues, as cited by Anthropic, found a fall in employment of 6 to 16 percent for the same age group in exposed occupations. Summaries of this research also give larger numbers. They mention a drop of more than three percentage points since a peak in November 2023. They also say that exposed entry-level jobs that are about seven times more likely than other entry-level jobs to ask for senior skills like judgement and leadership.

The most exposed occupations in the Anthropic data include computer programmers, customer service staff, data entry and financial analysts. Those are jobs where many people used to start.

In Belgium. The High Council for Employment sees no fall in the share of Belgian workers in the jobs most at risk of being replaced by AI. But it points to the same foreign studies on young workers and warns that it may become harder for young people to take the first step on the career ladder. It names administrative staff, such as secretaries, as the most vulnerable group. Their work is highly exposed, and AI replaces more of it than it supports. The Flemish AI barometer says something similar. Some school leavers, including highly educated ones, may find a place on the labour market less quickly.

The announcements. Challenger, Gray & Christmas counted 116,175 job cuts attributed to AI in the first eight months of 2026, about 22 percent of all announced cuts. For all of 2025 the same firm counted about 55,000, according to JobsPikr. I would be careful with these numbers. They count what companies say. For the share price, “AI is doing the work” sounds better than “demand fell”.

The other direction. In 2024 the AI sector directly created about 119,900 jobs in the United States, against about 12,700 counted as lost to AI automation. I do not expect that ratio to last. For now, more jobs were added than lost.

What may follow

Jobs are the part of the AI question that touches most people. In my view, the loudest claims on both sides have little support here. Mass unemployment does not show in the data yet. A fall in job starts for young people does.

An economy can keep the same number of jobs and still become much harder to enter. That hurts the people who can least afford to wait. Those are young people without experience. In Belgium it also affects women, according to the High Council, because they more often work in administrative jobs where AI replaces more than it supports.

It matters who does the measuring. The most-cited reassuring study comes from two economists employed by Anthropic, an AI lab. That does not make their numbers wrong. But anyone who quotes them should mention it.

What I do not know

I found no Flemish or Belgian study that follows young workers in exposed occupations the way Anthropic and Stanford do in the United States. For that point the High Council relies on foreign studies.

Some numbers here come from summaries, not from the studies. I could not find these in the linked sources: the drop of more than three points since November 2023, the seven-times figure, Challenger's 116,175 cuts, and the 119,900 against 12,700 US jobs. The three-point drop also fits badly with Anthropic's own monthly rate of about 2 percent. Check them before you use them in front of a room.

My notes

Whoever is up front will get this question, probably first, and probably from someone with a child about to graduate. I would give the Anthropic number, 14 percent fewer job starts for 22- to 25-year-olds, and say how thin it is. I would not make it sound worse or better than that.

My short version is that the total number of jobs holds for now, but young people find it harder to get in.

Read next

Sources

  1. Anthropic: labor market impacts of AI, a new measure and early evidenceresearch · main source

    Massenkoff and McCrory, 5 March 2026. Research by an AI lab into whether AI costs jobs. I recommend it, with that in mind. I have not read the appendix.

  2. Hoge Raad voor de Werkgelegenheid: Artificiële intelligentie op de Belgische arbeidsmarkt, februari 2026primary · not read end to end yet

    Source of the Belgian findings. Summary and recommendations read, the full analysis not yet.

  3. Departement WEWIS: Vlaamse AI-barometer 2025primary

    Press release. Source of the line on school leavers.

  4. IntuitionLabs: AI and entry-level jobs, what hiring data really showsresearch · not read end to end yet

    Searched, not read in full. The entry-level figures once attributed here do not appear in the current version of the page.

  5. Futurum: AI isn't coming for your job yet, and maybe never willargument · not read end to end yet

    Source of the summary of the ECB survey by Lebastard and Sondermann. Searched, not read in full.

  6. JobsPikr: AI layoffs 2026, the ROI reality checkresearch · not read end to end yet

    Source of the Challenger figure for 2025. Searched, not read in full.