A Stanford University study found that employment declines among entry-level workers in their early 20s are widening in occupations with high AI exposure.

On Aug. 24 local time, IT outlet Ars Technica reported that Stanford economists said in their August 2026 report that employment among workers aged 22 to 25 fell more in high AI-exposure jobs than in low-exposure ones.

The study updates last year's report, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," with the latest data. The researchers said the weakness in entry-level employment they identified last year is not temporary, and is persisting while also broadening in scope. Employment levels for workers aged 22 to 25 in the most AI-exposed occupations were 19 percent lower than for the same age group in jobs with less AI disruption. The gap was 13 percent last year.

The researchers used anonymised, high-frequency payroll data from human resources management firm ADP. They combined labour market impact indicators from prior studies with Anthropic's economic index to assess AI exposure by occupation. Anthropic's index reflects how each occupation uses AI in day-to-day tasks, based on Claude use cases.

Looking at the overall labour market, there was little difference in relative employment levels between jobs more affected by AI and those less affected. The pattern differed for the 22 to 25 age group. The researchers estimated that employment for this age group fell by about 11 percent in the top 40 percent of AI-affected occupations since 2022. Over the same period, total employment rose 10 percent in the bottom 60 percent of occupations with lower AI impact.

The employment deterioration appeared mainly in slower hiring rather than increased layoffs. The researchers said the pattern was driven mainly by lower hiring rates for entry-level workers in AI-affected fields, rather than by rising layoffs or voluntary quits. They also found that jobs themselves were shrinking rather than wages falling.

Differences by occupation were also clear. Anthropic distinguishes between automation, where AI replaces human tasks, and complementary use, where it raises human productivity. Accountants and auditors, and receptionists and information clerks were classified as occupations with a strong automation impact. Chief executives and nurses were presented as occupations with more complementary use. The researchers wrote, "Automation-focused AI use is consistent with labour replacement, and complementary use is linked to employment stagnation or growth."

The researchers said entry-level workers may be more vulnerable in jobs that rely heavily on formalised knowledge. They said knowledge that can be taught through education, textbooks and documented procedures is easier for AI to replace, while tacit knowledge built through practical experience, repeated exposure and mentoring is more likely to pair with skilled workers. To test this, they used education requirements in the O*NET occupational database as a proxy indicator and found that the higher the share of formalised knowledge in an occupation, the slower the increase in entry-level employment. By contrast, in occupations with a higher share of tacit knowledge, employment growth was faster among mid-career and more experienced workers.

The study also suggested education could act as a buffer. Occupations with a higher share of college graduates had relatively smaller employment differences between high and low AI-exposure jobs. In occupations with a lower share of college graduates, the pattern was clearer, with low-exposure jobs increasing and high-exposure jobs decreasing.

Erik Brynjolfsson (에릭 브린욜프슨), who led the study, warned that current trends point to a direction in which existing workers keep their jobs but opportunities shrink for new cohorts entering the labour market. "The entry-level job effects we measured are real, persistent and growing," he said. "A labour market in which overall employment levels hold up while the onramp for people starting careers quietly closes is more concerning."

The findings differ from claims that AI is immediately replacing the entire labour market. Still, it was clear that entry-level hiring is declining first in areas where companies are applying AI early. This leaves how AI adoption evolves as a key variable for youth employment, including whether it solidifies around automation or shifts toward complementing skilled workers.