CESifo Study Finds No Widespread AI Job Loss Among Recent US Graduates

The feared surge in joblessness among new US college graduates has not appeared in a new analysis of data through summer 2026. A CESifo working paper finds no evidence of significant, widespread displacement or reduced hiring, even as other research points to weakness in some AI-exposed jobs.

Why the researchers looked at new hires

Robert Fairlie and Jane Wu focused on recent graduates because shifts in labor demand may first appear in employers’ hiring decisions. AI could take over some standardized tasks in entry-level office jobs, leading firms to recruit fewer beginners rather than lay off experienced staff. That makes young workers a useful group to examine, but does not show that firms have broadly made that choice.

The paper defines its group narrowly: people ages 22 to 25 with bachelor’s degrees who are not pursuing higher degrees. Using the US Census Current Population Survey, the researchers examined unemployment from 2022 through summer 2026. Summer matters because joblessness predictably rises as graduates enter the labor market; comparing one summer with another helps separate that seasonal pattern from an unusual change.

What the summer numbers show

Unemployment among those graduates was 7.3% in summer 2026, within the 6.3% to 7.8% range recorded in earlier summers from 2022 to 2024. The pattern held when researchers included graduates who wanted a job but were not actively looking—people excluded from the official unemployed group.

7.3%Recent graduates

The 2026 rate fell within the 6.3% to 7.8% range observed in the earlier summers cited by the researchers.

Fairlie and Wu also tested whether the broad rate concealed a relative disadvantage. They compared recent graduates with people the same age without college degrees and with college graduates ages 30 to 49, and divided jobs by potential AI exposure using an earlier assessment of tasks AI systems could perform. Across almost all comparisons, differences in trends from 2022 to 2026 were not statistically significant. The tests therefore did not identify reliable differences in those trends, but they do not establish that no individual job was affected.

Why a different study saw trouble

The result appears to clash with earlier Stanford research that found entry-level employment lagging in occupations judged more exposed to AI. But the studies ask different questions. Stanford used ADP payroll data to examine employment in particular kinds of jobs. The CESifo paper used a wider household survey to ask whether recent graduates, as a group, were unusually likely to be unemployed.

Those findings need not cancel each other out. Employment can weaken in a set of occupations without producing an unusual unemployment rate across all recent graduates. The surveys also cover workers differently. Neither result, on its own, settles how much of any change in hiring is caused by AI. The narrower study identifies a warning sign in exposed jobs; the broader one does not find widespread harm in graduates’ unemployment through summer 2026.

The limit of a first test

The authors describe summer 2026 as an initial test, not a verdict on future graduating classes. They warn that graduates entering in 2027 or later could face different conditions if workplace AI use continues to intensify. More years of data will be needed to determine whether an effect emerges as use deepens. For now, the paper challenges claims of widespread harm already visible in the graduate job market, not the possibility of a later or more concentrated impact.