Workers displaced by artificial intelligence (AI) are trapped in a cycle they cannot easily escape. New research from Goldman Sachs shows that AI’s impact on the labor market has spread from the tech sector to finance, insurance, real estate, and professional services—and that workers pushed out of their jobs by AI not only have a harder time finding new work than other laid-off workers, but often land back within range of AI’s next wave of displacement after they are re-employed.
Goldman Sachs economist Pierfrancesco Mei noted in the report that in non-manufacturing sectors, industries with higher AI adoption are now showing weaker employment growth and higher unemployment rates. The data show that workers displaced from jobs highly exposed to AI take about one extra month on average to find new work, and are roughly 10 percentage points more likely than other laid-off workers to still be unemployed a year after losing their jobs.
Goldman Sachs Chief Economist Jan Hatzius estimates that every 10-percentage-point increase in AI adoption corresponds to about a 0.9-percentage-point decline in employment growth and a roughly 0.9-percentage-point rise in the unemployment rate in the affected industry.
The study draws on new data from the US Census Bureau’s Displaced Worker Supplement, covering more than 15,000 involuntarily unemployed workers between 2016 and 2025, and defines the most recent cohort, from 2023 to 2025, as the “first wave of AI-displaced workers.” The research team built a composite score combining occupational AI exposure, industry AI adoption, and the manner in which jobs were eliminated, then identified the top 30% of workers most likely to have lost their jobs because of AI and compared them with other laid-off workers.
The results show that before 2024 the differences between the two groups were not pronounced, but in 2025 a clear gap emerged: AI-displaced workers took markedly longer to find new jobs, and their probability of long-term unemployment rose significantly. Goldman Sachs noted that the magnitude of this short-term shock is “comparable to the impact of past waves of technological change on displaced workers,” meaning the argument that “AI only creates jobs and does not destroy them” is becoming hard to sustain.
The finding contrasts with earlier market views. Apollo analyst Torsten Slok said in August this year that AI’s impact on employment was “negligible,” with his research finding that AI mainly compresses wages rather than the number of jobs—wage growth for positions with the highest AI exposure has run 6.7% slower than for low-exposure positions since 2023, but headcount has shown no statistically significant change.
Where Displaced Workers Land Is the Real Problem
The most unsettling finding in the Goldman Sachs research is not the lengthening of unemployment spells but the direction of re-employment. In past technological revolutions, workers displaced by technology tended to move “downward,” shifting into more routine, less complex roles as their existing skills gradually depreciated with each wave of technological change.
The current wave of AI displacement follows a similar—or even more severe—pattern. When AI-displaced workers find new jobs, they tend to seek positions where they can use their existing skills, but this is precisely what makes them more vulnerable to the next round of automation risk.
| Period | Probability of landing in a job with equal or higher AI exposure after re-employment |
|---|---|
| 2018–2023 (median level) | About 55% |
| 2024 | About 65% |
| 2025 | Nearly 75% |
Note: Data are from Goldman Sachs research, covering workers displaced from jobs highly exposed to AI.
In other words, three out of four workers displaced by AI are walking back into AI’s next hunting ground.
US Bureau of Labor Statistics data confirm the trend: between 2023 and 2025, 7.4 million workers were involuntarily unemployed, up from 6.3 million in the previous survey cycle; and among long-term unemployed workers who found new full-time jobs, only about 49% achieved pay at or above their previous position, down sharply from about 62% two years earlier.
Goldman Sachs historical data show that the effects of technological unemployment are long-lasting. In the years after job loss, the real wage growth of technology-displaced workers consistently trails that of other laid-off workers, and their probability of falling into unemployment again remains elevated over the long term. In four decades of individual-level data, workers who were older, lacked a college degree, lived in rural areas, or had held their previous position for a long time showed a cumulative real wage gap of about 8 to 10 percentage points a decade after job loss—roughly twice that of younger, college-educated, urban, short-tenure workers.
Still, the new data contain one relatively encouraging signal: younger workers are clearly more adaptable to AI’s impact. In 2025, AI-displaced workers under 35 needed about three extra weeks to return to full-time work compared with other unemployed workers; those 35 and older needed about five extra weeks and went through more transitional periods of temporary and part-time work.
Goldman Sachs also noted that college students “are increasingly choosing majors with lower AI exposure.” However, data from the Stanford Digital Economy Lab show that young people aged 22 to 25 in occupations highly exposed to AI have employment levels about 19% below what would be normal—their problem is not slow adaptation but the fact that no one is hiring.
Layoff Wave Could Be Triggered When the Economy Turns Down
The US unemployment rate remains relatively low at 4.2%, and the AI displacement effect has yet to fully show up in macroeconomic data.
Goldman Sachs economist Joseph Briggs explains: historically, the employment shocks from automation of routine jobs have been overwhelmingly concentrated during recessions. The data show that the share of American workers in routine jobs held roughly steady during each economic expansion, then fell in stepwise fashion during the downturns of the early 1980s, 2001, 2008, and 2020—and never rebounded.
Based on three historical projection methods, Briggs concluded that every 1% increase in labor productivity driven by technological progress raises the unemployment rate by about 0.3 to 0.4 percentage points at its peak, driven mainly by job destruction. Building on Goldman Sachs’ forecast that AI will ultimately boost productivity by 15%, his estimates show that about 9% of workers will ultimately be displaced during the AI transition, with an additional 3% of hiring contraction pressure—roughly 15 million American workers in total, up from an estimate of 6% to 7% earlier this year.
He explicitly noted that companies tend to retain redundant positions during booms, and once the economy turns, they use the cover of a cyclical downturn to push automation forward in a concentrated way. More alarmingly, the AI capital expenditure boom has itself become an important pillar of current economic growth, and Hatzius warns that this momentum is peaking—meaning the fuse that triggers the next recession may be AI itself.
Bank of America economists take a relatively milder view, arguing that AI’s impact on the labor market may remain relatively limited through 2026, with broader structural restructuring more likely to occur in 2027. Even so, this “optimistic scenario” only delays the large-scale shock by about two years.
Notably, Goldman Sachs points out that currently only about 23% of US companies have integrated AI into some routine workflows, and that this figure “overstates the degree of automation actually occurring today.” Yet even at this relatively limited adoption rate, AI has already produced statistically significant job losses outside the tech sector—and displaced workers who find new jobs are highly likely to be exposed once again to the risk of AI displacement.
If AI adoption doubles or even triples from current levels, each of these effects will be further amplified: more displaced workers will chase the limited number of jobs that are “hard for AI to replace,” unemployment spells will lengthen further, and workers who move into new positions will face the risk of a second displacement. Briggs’ conclusion is that the bulk of this shock will likely be released in a concentrated way when the next recession arrives—precisely when the economy is least able to absorb it.
He warned that the Federal Reserve and other central banks “should pay close attention to any emerging downside risks to growth or the labor market,” and that once recent inflation constraints ease, “a rapid pivot toward easing could help ensure that automation-related job losses do not accelerate at the economy’s most vulnerable moment.”
Cooling in the job market is already reflected in corporate actions. According to TheStreet, Visa announced at the end of July that it would cut about 7% of its global workforce, equivalent to roughly 2,600 positions—one of the latest cases this year of a company citing AI as a reason for layoffs.