Unemployment among recent US college graduates averaged 7.3% from June through August 2026, close to the previous summer's 7.2% and below the 7.8% recorded in 2024. Those figures come from a September working paper by Robert Fairlie and Jane Wu that tests whether a widely discussed AI effect on new entrants is visible in household survey data. The authors do not find a statistically clear increase in unemployment associated with summer 2026. Their analysis cannot establish that AI has had no effect on entry-level work.

The big change

  • What changed: A new analysis measures recent graduates who are out of work, including some who want a job but are excluded from official unemployment because they have not recently searched. That broadens the evidence beyond payroll employment and job postings.
  • Why it matters: Young graduates' joblessness rises each summer, so a June increase alone says little about AI. The 2026 standard rate stayed inside the range of the previous four summers, while the broader measure reached its highest point across all five summers studied. They capture different parts of the difficulty new graduates face in finding work.
  • What to watch: Future measurements will need to distinguish AI exposure from other changes in the same occupations and capture graduates who lack a previous occupation. This study's comparison groups and survey questions leave both issues unresolved.

The study compares like months and three worker groups

Fairlie and Wu used individual records from the US Current Population Survey, which underlies the government's employment figures, from January 2022 through August 2026. They defined “recent graduates” as people aged 22 to 25 whose highest degree is a bachelor's and who are no longer enrolled in school. That is an age and education group, not a verified list of people who graduated in 2026. They compared its June-to-August unemployment with the same months in 2022 through 2025, then checked whether its change differed from that of college graduates aged 30 to 49 or young adults without a college degree.

The same-month comparison matters because unemployment in the selected graduate group normally rises during summer. The paper reports 7.1% in summer 2022, 6.3% in 2023, 7.8% in 2024 and 7.2% in 2025, against 7.3% in 2026. In its adjusted models, which account for seasonal patterns, earlier trends and demographic and geographic characteristics, the 2026 differences were not statistically significant. The relative changes against either comparison group were also not statistically significant.

A broader measure shows more people wanting work

The Bureau of Labor Statistics definition generally counts a person as unemployed when they have no job, are available and actively sought work in the previous four weeks. Fairlie and Wu also counted respondents outside the labor force who said they wanted a job. This is the authors' expanded measure, not the official unemployment rate.

On that measure, the summer 2026 rate was 10.4%, up from 9.4% in summer 2025 and just above the earlier high of 10.1% in 2024. The raw increase deserves to be seen alongside the adjusted result: the models did not show a statistically significant rise relative to prior summers or the comparison groups. A non-significant estimate leaves room for smaller changes the data cannot reliably distinguish, as well as changes in particular kinds of work hidden by the aggregate rate.

The paper cannot isolate AI's effect on hiring

The study observes employment status, not employers' reasons for hiring or rejecting a graduate. Its comparison groups may also be exposed to AI. The authors tested whether unemployment changed more in occupations with greater theoretical or observed AI exposure. Those estimates were positive but not statistically significant. Their separate measure of whether an occupation can be performed remotely had a positive association with unemployment, but remote work and AI exposure overlap. The paper cannot disentangle their effects.

Occupation analysis has a further gap: some people without work have no current or previous occupation to match to an exposure score. The authors say this removes about 16% of unemployed recent graduates and 36% of the broader unemployed-plus-wants-work group from those models. The overall unemployment comparisons include people without an occupational history.

Fairlie and Wu call their analysis descriptive and caution against a causal reading. It checks one proposed sign of AI displacement: an unusual jump in recent graduates' unemployment. It does not directly measure starting pay or employers' hiring decisions. The data through August 2026 do not settle how AI is changing those outcomes.