Researchers at Stanford Digital Economy Lab say that workers between the ages of 22 and 25, who are most exposed to AI, have seen a 13% decline in employment compared to other occupations.
This comes as the employment of more experienced workers in the same job, and workers of all ages in occupations that are not exposed to AI are either stable or grown.
The findings detailed in the working paper entitled “Canarians in Coal Mine? Six Facts on the Recent Employment Effects of Artificial Intelligence” echoes pre-reports on the challenging job markets of recent graduates in the field of computers.
For example, data released in February 2025 by the Federal Reserve Bank of New York found that the unemployment rate among university graduates was 5.8% in recent years, compared to an average of 4.0%. And among those who have studied computer science, the unemployment rate in the US is 6.1%, and has skyrocketed to 7.5% due to computer engineering.
A report from Signafire in May found that new graduate employment had dropped to 50% of pre-pandemic levels. The report describes some shifts due to AI, a situation as a job reset where companies want more experienced workers and want fewer entry-level employment.
However, this seems to have been the case for a while. Last September, James O'Brien, a professor of Berkeley, California, Science and Science, published a post on LinkedIn, confirming the anecdotes reported in a Wall Street Journal report.
“Previously, Berkeley CS graduates would receive multiple attractive job offers in terms of job type, location, salary and employer, even if not top students,” he wrote. “But great students like students with a major 4.0 GPA have become worried because they have zero offers. I think this trend is irreversible and part of a wider trend affecting almost all employment sectors.”
The new Stanford Digital Economy Lab Paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen is based on analyses of monthly administrative data from US payroll company ADP from January 2021 to July 2025, representing millions of workers and tens of thousands of companies.
The study focuses on six findings and relies on data from AI companies that are not completely indifferent like humanity. [PDF] Openai to define occupational exposure to AI. Microsoft, another company that owns dogs in the race, has also released an AI role hit list. The most exposed occupations are said to include customer service representatives, accountants, auditors, software developers, secretaries and administrative assistants.
Researchers say the declining outlook for employment exposed to AI contributes to worse overall employment growth between the ages of 22 and 25, even if the employment of other workers grows at a higher rate. The authors also assume that not all use of AI involves a decline in employment. Effects vary depending on whether AI use automates roles or simply supports hiring tasks.
Boffins also notes that AI appears to have more impact on employment than wages at this time. Therefore, the softening employment market for computer science students in recent years has not cut salaries across the board.
Finally, they say their findings are not explained by other factors, such as the role of entry-level computer offshoring and outsourcing and support.
“We find that our results are not driven solely by computer occupations or by occupations susceptible to remote work and outsourcing,” the authors state.
The authors acknowledge that other researchers had no real impact on April studies of Danish workers that found generative AI had no real impact on employment or wages, and February studies that found AI was balanced with other productivity gains (M Hampole). However, the Stanford authors say these papers are flawed.
“These previous papers use data that lack either granularity or immediacy enough to ensure that AI exposure and changes in employment with age,” the authors argue. “In contrast, this paper uses large real-time data to take a step towards solving the ongoing debate about the employment effect of AI on young workers.” ®
