Early NBER Evidence Finds No Large Spike in 2026 Unemployment for Recent College Graduates

# What the paper studied Robert W. Fairlie and Jane Wu (UCLA) present a September 2026 NBER working paper titled "The Early Impacts of AI on Employment among Recent College Graduates." The authors use Current Population Survey (CPS) microdata to estimate whether the arrival of more widely used AI in workplaces produced detectable increases in unemployment among recent college graduates in June, July, and August 2026.

# Why recent college graduates The paper focuses on recent college graduates because changes in labor demand often first appear through reductions in hiring. Entry-level office tasks are relatively standardized, so the argument goes that firms could reduce new hiring for simpler roles rather than immediately laying off more experienced workers.

# Main findings Using multiple comparison groups and modeling approaches, the authors do not find evidence that unemployment rates for recent college graduates rose in summer 2026 relative to prior summers. They compare summer 2026 to the same months in earlier years and run difference-in-differences and event-study interaction specifications with two primary comparison groups: older college graduates and young workers without a college degree.

The paper also expands the usual unemployment definition to include people who report "wanting a job." Adding these "sidelined unemployed" raises the recent-graduate unemployment rate by nearly two percentage points, but even with this broader measure there is no statistically significant increase in summer 2026.

# Tests for AI exposure and remote work Beyond broad comparisons, the authors interact 2026 unemployment with occupational measures of AI exposure and remote-work availability. Results show some evidence of a positive relationship between higher relative unemployment and remote-work availability. The paper does not report a clear, consistent link between measured AI exposure and a relative rise in unemployment for recent graduates in summer 2026.

# How the authors approach timing The paper explicitly takes an agnostic approach to treatment timing, noting that 2026 might be the first year when AI use is widespread enough in workplaces to generate measurable impacts on recent college graduates. Their approach therefore searches for departures in summer 2026 without presuming a single, fixed treatment date.

# Interpretation offered in the post Blog post summarizing the working paper (Conor Clarke) interprets the rough bottom line as "not that bad, so far." That phrasing signals that the paper finds no large, detectable displacement of recent graduates in the specific early-window it studies, while leaving open the possibility that dynamics could change later.

# Practical takeaway for readers The working paper provides an early, empirical check on concerns that AI would immediately and sharply reduce hiring of recent college graduates. For summer 2026, using CPS data and multiple comparison strategies, the paper finds no statistically significant spike in unemployment for that cohort. There are occupational patterns worth watching, particularly where remote work is common, but the aggregate early signal is limited.

# What to watch next Future research should track later months and different outcome measures (earnings, underemployment, sectoral hiring), and continue refining occupational measures of AI exposure. The paper establishes an early baseline against which later changes can be compared.