A law changed who counts as unemployed in France.

The reading, and the six quarters behind it

INSEE put French unemployment at 8.3% of the labour force in the second quarter of 2026: 2.7 million people, 62,000 more than the previous quarter and 261,000 more than a year earlier. It is the highest reading since the third quarter of 2020, a quarter deformed by the pandemic, and before that since the third quarter of 2019. It remains 2.2 points below the mid-2015 peak, which is the part most likely to be dropped from a headline.

The increase is broad rather than concentrated. Over the quarter the rate rose for every age band: youth unemployment rebounded 0.4 points to 21.6%, the 25-49 band rose 0.2 to 7.5% - its highest since early 2021 - and the over-50s rose 0.3 to 5.5%. Over the year the youth figure moved furthest, +2.5 points. Women's unemployment rose 0.4 points on the quarter to 8.2% while men's was stable at 8.5%; across the year both moved the same 0.8 points.

Long-term unemployment is rising with it. 671,000 people reported having been out of work and searching for at least a year, 36,000 more than the previous quarter and 116,000 more than a year before - 2.1% of the labour force, the highest since mid-2022. That series is the one to watch, because it cannot be produced by a change in who registers.

A law that changed who is searching

The loi pour le plein emploi, passed in December 2023, provided that from the start of 2025 RSA recipients, young people supported by the Missions locales, and disabled people supported by Cap emploi would be registered with France Travail automatically. INSEE says plainly that this produced a mechanical increase in registrations in the first quarter of 2025, and it has published the affected groups' contributions every quarter since.

Why an administrative act moves a labour-market statistic is worth stating slowly, because it is the hinge of the piece. The international definition counts someone as unemployed only if they are without work, available to start, and actively searching. Registering with the public employment service is an act of search. So a person who wanted work, was available, and was not formally searching - counted as inactive yesterday - can become unemployed in the statistics today without their employment, their income or their intentions changing at all.

The size of that effect is INSEE's own number, not an inference. Across the six quarters since the law took effect the unemployment rate rose 0.98 points. Of that, RSA recipients contribute 0.31 points and young people registered with France Travail contribute 0.11. INSEE summarises the pair as nearly half of the rise. Their rates are extraordinary in isolation - unemployment among RSA recipients reached 48.6%, up 7.9 points since the end of 2024 - but it is their weight in the labour force, 3.2% and 3.9%, that converts those rates into the modest point contributions above.

Why this is not "the rise is fake"

It would be easy, and wrong, to read the decomposition as an unmasking. Two things in the same release refuse it.

The first is arithmetic. The population the law never touched - 92.9% of the labour force, the line INSEE calls "other people" - contributes 0.57 of the 0.98 points, the majority of the rise. Whatever the registration effect is doing, most of the increase is happening among people whose relationship to the statistics did not change.

The second is the direction of a different number. If the RSA line were purely a reclassification - the same people, relabelled - their employment rate would be untouched, because nobody would have lost work. It fell: down 2.5 points since the end of 2024, to 24.5%. A relabelling cannot do that. Something real is happening to that population as well as something statistical, and the release lets you see both without choosing between them.

The honest summary is narrower and less quotable than either slogan. A measurement change and a labour-market deterioration are both present, and INSEE has published the split rather than leaving readers to guess it. A desk that prints only the 8.3% has told the truth; a desk that prints only the decomposition has implied a fiction.

The halo did not move - and it sits on a different base

The halo around unemployment is INSEE's count of people who are inactive by the international definition yet want a job, either because they are not searching or not available to start. In the second quarter of 2026 it stood at 1,865,000 people, 4.4% of the 15-64 population - stable on the quarter, and down 0.1 point on the year.

Set that beside a rate that rose 0.7 points over the same year and an obvious question appears: if people were being moved from wanting-work-without-searching into searching, the halo should be falling as unemployment climbs. It barely moved. That comparison is ours, not INSEE's, and it is a question rather than a finding: the two figures sit on different denominators and different populations, and the law's registrations may draw from parts of inactivity the halo never counted. We print the observation and decline the conclusion.

And a warning this desk has now earned three days running. The unemployment rate is a share of the labour force. The halo is a share of the 15-64 population. Yesterday's European slack figure was a share of the extended labour force. Three editions, three denominators, and none of them can be added to or ranked against another. When a page shows you 8.3% and 4.4% side by side, the one operation that suggests itself is the one that produces a number corresponding to no population at all.