Unemployment among college graduates is increasing

“The discussion about AI taking jobs, or at least good jobs, is generally framed as tomorrow’s problem. However, there are some hints that AI may already be taking ‘knowledge worker’ jobs,” said Michael Feroli, chief U.S. economist at J.P. Morgan.

For starters, the unemployment rate among college graduates has increased, reaching 5.8% in March — the highest in more than four years. “Disturbingly, the unemployment rate has been trending above the aggregate rate, which is highly unusual by historical standards,” noted Murat Tasci, senior U.S. economist at J.P. Morgan. While the trend could be partially due to a secular rise in the relative supply of college graduates and unusually low labor market churn, the effects of AI could also be at play.

To investigate the impact of AI on graduate unemployment, the economics team at J.P. Morgan Global Research examined the cross-sectional distribution of unemployment rates by college major, noting that the technology could affect some fields more than others. Overall, anthropology graduates — who are not particularly exposed to AI — have had most difficulty finding jobs. However, other majors exposed to AI, including computer engineering, graphic design, industrial engineering and architecture, have also experienced significant increases in graduate unemployment, highlighting AI’s impact on job growth.

“To the extent that exposure to AI displaces some workers with specific majors, new entrants from those majors could be at a greater disadvantage, as experience could offset some of the skills that are now obsolete with AI-performed tasks,” Tasci said.

Job growth is slowing across certain tech industries

In addition to rising unemployment among college graduates, job growth across several white-collar sectors has been tepid, pointing to AI’s growing role in the workforce.

To this end, the economics team compared changes in employment trends with AI exposure across industries, as well as within detailed subsectors of professional employment, using data from the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS) and Anthropic AI. On the whole, there is a mildly negative correlation between employment trends and AI usage, suggesting that AI may be depressing job growth.

This trend is especially evident in certain tech industries, including cloud, web search and computer systems design. After years of steady employment gains, these three industries stopped growing at the end of 2022, just after the release of ChatGPT. “While firms may also have been retrenching after their post-COVID hiring, the rise of large language models (LLMs) produced a product that automated some of their labor via code assistance, while at the same time reshaping their business strategy,” Feroli noted.

On the other hand, this trend isn’t as apparent within the professional and business services segment. “The most likely explanation for our non-result across non-tech industries is that AI is simply too recent to have materially changed business practices, even if it were to ultimately prove impactful,” Feroli said. Indeed, BTOS data shows that as of mid-2025, less than 10% of firms in the overall economy indicate they are using AI regularly, with the figure rising to just over 20% in the professional, scientific and technical industries.

“All in all, we find little association between various measures of AI intensity and job growth outside of selected tech industries, and conclude that so far AI has not been a major driver of the composition of employment gains, except perhaps in tech,” Feroli added. 

AI could affect the labor market in the next recession

In addition to its impact on job growth, AI could prolong the recovery of the labor market in the event of a downturn. “We think that during the next recession, the speed and breadth of the adoption of AI tools and applications in the workplace might induce a large-scale displacement for occupations that consist of primarily non-routine cognitive tasks,” Tasci said.

Back in the 1980s, middle-skill occupations that rely on routine tasks — such as sales, manufacturing construction and maintenance — started disappearing due to automation. Notably, their decline had a pronounced cyclical behavior. Throughout the downturns over the following decades, it took increasingly longer for these occupations to bounce back from recession-induced job losses, each time resulting in a jobless recovery — an economic situation in which the overall economy improves but payrolls grow only slowly.  

Today, AI exposes non-routine cognitive occupations — such as scientists, engineers, designers and lawyers — to similar risks. “As the cost of adaptation comes down with the wide dissemination of AI tools and applications, these traditionally high-wage occupations might come under pressure,” Tasci noted.

This is compounded by the rising unemployment risk among these occupations, which has now surpassed that of workers from non-routine manual jobs. “Workers who were last employed in non-routine cognitive jobs have always accounted for the smallest share of the unemployed, until recently. This changing pattern might be indicative of rising unemployment risk for these workers going forward,” Tasci said.

Taken together, these factors suggest that a jobless recovery could recur in the future. “Overall, a much larger unemployment risk and anemic recovery prospects for workers in non-routine cognitive occupations might cause the next labor market downturn to look pretty dismal,” Tasci added.