IN BRIEF
Fairlie and Wu used the US Current Population Survey for June, July and August 2026, comparing results with earlier summers and other worker groups.
Recent graduates showed no statistically significant relative unemployment increase. A broader measure including people who want work did not change the conclusion.
The observational analysis does not directly measure all job vacancies, wages or firm-level AI use.
WHY IT MATTERS
New graduates face headlines about disappearing entry-level jobs. Testing that concern requires distinguishing plausible pressure on hiring from an observed unemployment surge.
An absent statistical spike does not make early-career difficulties disappear or establish that AI is irrelevant. It sets a boundary around what current data support.
WHAT IS NOW POSSIBLE
Before: fear of an early AI-linked unemployment surge outpaced evidence controlling for summer seasonality. Now: a working-paper analysis of current CPS observations finds no significant relative surge for this period.
WHAT MADE IT POSSIBLE
The US Current Population Survey supplies monthly employment status and demographic information. The authors compare recent college graduates with older graduates and younger workers without degrees.
They use seasonal comparisons, difference-in-differences and event-study models, and examine a broader group who say they want work.
THE STORY
For someone looking for a first job, the question is practical: is artificial intelligence already making it harder to find work after college? US federal data for summer 2026 do not show a statistically significant, unusual relative unemployment spike among recent graduates. That is the finding of economists Robert W. Fairlie and Jane Wu in a national survey comparison. It does not mean that entering the workforce is easy or that AI has been ruled out as a source of change.
The calendar matters. Every summer, large numbers of people finish college and begin looking for work. Graduate unemployment can rise seasonally without a new technology shock. The study therefore compares June, July and August 2026 with previous summers, and also looks at older college graduates and young people without degrees. That helps distinguish ordinary transitions into the labor market from an exceptional new pattern.
Using the Current Population Survey, the authors found no statistically significant relative increase for recent graduates in their reported models. The conclusion held under a broader definition that counts people who say they want a job even if they are not conventionally classified as unemployed. That wider measure raised the measured level by almost two percentage points but did not produce a distinctive summer 2026 surge. A higher unemployment level and an unusual year-on-year increase are not the same claim.
The researchers also examined occupations' exposure to AI and remote-work opportunities. Their analysis found some signals involving remote work but no general causal evidence that AI raised the unemployment rate of recent graduates. The research is observational. It does not randomly assign comparable firms to adopt or avoid AI, and cannot isolate every hiring decision made by employers.
Circulated through IZA and NBER in September 2026, the paper adds an empirical check to a debate full of forecasts and anecdotes. Employers may offer fewer junior positions, change entry-level tasks or alter wages before a change becomes visible in headline unemployment figures. Longer data series, vacancy records, pay information and direct measures of AI adoption are needed. The useful finding today is limited but meaningful: the feared unemployment spike does not appear in this comparison, while the longer-term consequences for early careers remain unsettled.
WHAT WE KNOW
The CPS observations cover June-August 2026 and several comparison groups. [1][2]
The main models do not find a statistically significant relative graduate-unemployment increase.
Including people who want a job raises the level by nearly two percentage points, not the statistical significance of the 2026 change.
WHAT WE DON'T KNOW
The result may not hold in later quarters or narrower occupations.
A non-significant estimate is not proof that the true effect is exactly zero.
The data do not causally attribute each unemployment change to AI adoption.
WHAT HAPPENS NEXT
Extend comparisons with later data on hiring, employment and pay.
Measure which employers actually deployed AI tools in different tasks.
Study specific occupations and degree fields where aggregate averages may hide change.
SOURCE TRACE
No statistically significant relative graduate-unemployment spike was found.
The study uses June-August 2026 CPS microdata.
The expanded measure raises the level by nearly two points without a significant 2026 spike.
SOURCES
- The Early Impacts of AI on Employment among Recent College GraduatesOpen source
- The Early Impacts of AI on Employment among Recent College GraduatesOpen source
EVIDENCE & UPDATE HISTORY
- Initial publication record
EARLY EVIDENCE
- Story updated
EARLY EVIDENCE
Updated: Deck, In Brief, Why It Matters, The Story, 10-Second Takeaway
CURRENT EVIDENCE
EARLY EVIDENCE
REAL-WORLD SURVEY EVIDENCE / PRELIMINARY ECONOMETRIC COMPARISON
WHAT WOULD CHANGE THIS?
- Later labour-force data could show persistent changes after summer 2026.
- Actual company AI-use measures would improve causal attribution.
- Occupation-specific analysis could uncover changes masked in aggregate unemployment rates.