Research update · 16 September 2026
The U.S. early-career AI warning just failed its first clean cross-country stress test.
A population-wide Norwegian study sees the same direction of risk for young workers in highly exposed occupations. It does not find a robust, statistically significant effect. That distinction cuts directly into the easiest AI-jobs narrative.
What the Norwegian study measured
Dennis Facius and Roberto Iacono use Norwegian employer-employee administrative records from January 2015 through March 2025. Their main sample covers private-sector workers with a full-time main job and maps occupational generative-AI exposure onto them. The focal group is workers aged 22–25, the same age band at the centre of the U.S. payroll warning.
Three designs, one uncomfortable result
The paper tests the pattern within firms, across occupations using synthetic difference-in-differences, and across firms using a predicted-exposure shift-share design. The young-worker estimates point down, but none is individually statistically significant. A backdating exercise produces placebo estimates that can exceed the estimate after ChatGPT became widely available.
What it means for the 19% U.S. gap
It does not make the U.S. payroll, Census hiring or graduate findings disappear. Those studies examine different people, labor markets, outcomes and periods. It does mean that a relative decline after ChatGPT cannot be sold as a universal AI effect when another rich-country administrative system sees a non-robust result and longer pre-trends expose a rival story.
Why this is not an all-clear
Norway's institutions differ from the United States, and this study measures occupational exposure rather than which employer adopted a particular system. Its post-period ends in March 2025, and the authors note later young-worker coefficients become more negative. The result narrows the claim; it does not settle it.
Firm adoption still does not produce one answer
An EIB paper using broadly defined AI adoption reports no adverse firm-level employment effect in a 2019–2024 European-firm model. A U.S. spending study reports growth at selected heavy adopters. Both are useful brakes on an economy-wide collapse claim. Neither observes the same entry routes, jobs or counterfactuals as the U.S. exposure studies.
The bottom line
The evidence now supports a narrower warning: entry routes may be tightening in some settings, but the cause and reach are unresolved. Demand a result's country, outcome, exposure or adoption measure, comparison and pre-trend before using it to forecast a career.