AI Was Supposed to Hit New Grads Hard. Unemployment Data Says Otherwise

A new working paper from economists at Munich's CESifo finds no evidence that AI is displacing recent college graduates from the labor market — directly contradicting a Stanford study published just weeks earlier that found the opposite.

The CESifo researchers state their conclusion point blank: "there is no evidence of any significant, widespread displacement or reduction in hiring of recent college graduates in absolute or relative levels."

That sentence lands in the middle of an increasingly heated debate over whether generative AI is hollowing out the bottom rung of the white-collar career ladder. For the past year, policymakers, labor economists, and the graduates themselves have watched as chatbots took on writing, coding, and analysis tasks once handled by junior staff. The question of whether that shift shows up in hiring data has real stakes — for students choosing majors, for universities setting curricula, and for regulators assessing whether AI-driven unemployment is a present danger or a speculative one.

Two studies, opposite findings

Last month, a Stanford study reported that entry-level employment in occupations it classified as "AI-impacted" lagged well behind employment in other fields. That finding traveled fast. It fit a popular narrative: AI handles the routine work first, so firms stop hiring the people who used to do it.

The CESifo paper, titled "The Early Impacts of AI on Employment Among Recent College Graduates," now pushes back. Its authors, Robert Fairlie and Jane Wu, examined the same broad question — what is AI doing to the job prospects of people fresh out of college — and reached the opposite answer from the unemployment data.

The discrepancy matters. When two credible research teams analyze overlapping questions and diverge this sharply, it signals that the empirical picture is unsettled. Any confident claim that AI is already destroying entry-level work — or that it is provably harmless — outruns what the data currently supports.

Why recent graduates are the right test case

Fairlie and Wu explain their choice of subject in the paper: they focused on recent graduates "because changes in labor demand may first appear through reductions in hiring."

The logic is straightforward. Companies facing an AI-capable workforce do not typically fire experienced employees en masse. It is cheaper and quieter to simply stop hiring. New positions go unfilled; entry-level roles get absorbed by software; the headcount shrinks through attrition rather than layoffs. If AI were reshaping labor demand, the graduates of 2025 and 2026 would be the first to feel it.

The researchers' theoretical concern is specific. As AI gets good enough to perform what they call the "relatively standardized tasks" that fill many entry-level office jobs, firms could reduce new hiring for simpler roles rather than laying off more experienced long-term employees. Junior analysts, assistants, and coordinators do work that is procedural enough to automate but cheap enough that firms once tolerated the cost of training someone through it.

That mechanism is exactly why the null result is notable. The channel most likely to transmit AI's labor-market effects early — reduced hiring of new graduates — does not yet show a measurable signal.

Why 2026's graduates might be more exposed

Fairlie and Wu do not dismiss the risk. Their paper acknowledges reasons to believe 2026's graduating job seekers might face more exposure to AI displacement than those who graduated a year or two earlier.

They point to three indicators, all moving in the same direction over roughly the last twelve months. A recent Census survey recorded a sharp increase in the number of firms "replacing a large number of employee tasks with AI." AI spending per employee has risen broadly. ChatGPT Enterprise token use has climbed as well.

Read together, those trends describe companies moving AI from pilot projects into production at accelerating speed. More firms are substituting AI for tasks, and the scale of usage — measured in tokens, spending, and surveyed task replacement — is climbing quarter over quarter. The exposure of entry-level work to these systems is growing, even if the employment data has not yet registered damage.

What the null result does and does not mean

The paper's finding is a snapshot of the early period of AI adoption, and the authors frame it as exactly that. The title itself says "early impacts." Absence of displacement in current unemployment data does not rule out displacement later, particularly if the adoption indicators the researchers cite continue their sharp upward trajectory.

The working paper also stands against a growing stack of anecdotes and studies suggesting the opposite. The Stanford finding — that entry-level employment in AI-impacted occupations trails the rest of the economy — remains on the table. Reconciling the two will likely require better occupational classifications, longer time series, or more granular hiring data than either study had.

For now, the strongest empirical claim available is the CESifo team's own: no significant, widespread displacement of recent college graduates, in absolute or relative levels. The graduates of 2026 are job hunting into an economy where firms are aggressively deploying AI — and where the unemployment numbers, so far, have not broken in either direction.

If the Census-reported task replacement keeps climbing and the hiring data still shows nothing a year from now, the "AI kills entry-level jobs" thesis will need serious revision. If the data turns instead, this paper becomes the benchmark against which the damage gets measured.