An April 2026 Census paper tracked matched employer-employee records and found that hiring of workers ages 22 to 24 fell sharply in the industries most exposed to AI, while hiring in less-exposed industries held steady. Employment for early-career workers in the most AI-exposed fifth of industries dropped 12% over the ten quarters after ChatGPT’s release. Lee Tucker, one of the coauthors of the paper, said “the decline in hires is the primary cause” of that rate of unemployment, not people losing jobs they already had.

That mattered most for one type of graduate. The most AI-exposed industries, Tucker found, cluster heavily around software and information-technology work, which are the very fields computer science and other highly AI-exposed majors are built to feed into.

A second paper from last week, also coauthored by Tucker, follows the graduates of the most AI-exposed decile of college majors. Their odds of holding a job one quarter after graduation fell by five percentage points, and full-quarter initial earnings dropped 13% following ChatGPT’s release. A 13% earnings decline is roughly the size economists would expect from graduating into a severe recession—except there wasn’t one, since the rest of the labor market held up fine.

It’s lower-paying jobs, not no jobs at all

Young grads still need to work and still have jobs, even if they’ve received highly exposed degrees. So the decline in earnings is less about a lack of employment and more about pursuing lower-wage occupations to make ends meet.

About half of the earnings loss came from graduates earning less within the industries that did hire them. The other half came from a shift into different industries altogether, mainly lower-wage sectors like restaurants and retail. Together, the papers suggest that a computer science graduate applies for the same kind of entry-level software job an earlier class would have landed easily, finds the posting isn’t there, and eventually takes a job ringing up groceries or bussing tables instead.

The more recent paper found the earnings and employment gaps shrink over time but remain substantial for the most AI-exposed majors even years out. The previous one shows hiring volumes had largely recovered by early 2025, but off a smaller base of jobs, meaning the door reopened only partway.

What is striking is how long it took for the students to notice: undergraduate computer science enrollment fell 8.4% in spring 2026 from a year earlier, following a 3.6% drop the prior fall. A Goldman Sachs analysis in June found computer science and computer programming enrollment each fell more than 10% in the 2025-26 academic year, the first year Goldman’s economists saw students visibly reacting to AI in their major choices. A Gallup and Lumina Foundation survey cited in that report found that about 42% of bachelor’s degree students had reconsidered their major because of AI.

But in the earlier Census paper, hiring data show the market turning within months of ChatGPT’s release. Enrollment data shows students didn’t start abandoning computer science in visible numbers until three years later.

Still, that trend shows up beyond the two Census papers. The Federal Reserve Bank of New York’s ongoing tracker put the underemployment rate for recent college graduates at 42% in the second quarter of 2026, with unemployment for that cohort running at 5.6%, above the national rate. A Strada Institute and Burning Glass Institute analysis found 52% of graduates were working in jobs that don’t require a degree—retail, food service, hospitality, and office administration among them—within a year of leaving school, and 45% were still there a decade later.