Lifelong Scars: Goldman Sachs finds that workers replaced by AI often struggle to find new jobs and frequently become targets in AI’s next wave

Workers displaced by artificial intelligence are falling into a cycle that is increasingly difficult to escape. According to the latest research from Goldman Sachs, signs of AI impacting the labor market have spread from the technology sector into broader fields such as finance, insurance, real estate, and professional services. Displaced workers not only face greater difficulty finding new employment compared to other layoffs, but after being re-employed, they often fall into the next round of AI-driven replacements.

Goldman Sachs economist Pierfrancesco Mei wrote in the latest report that non-manufacturing sectors with high rates of AI adoption are now showing weaker employment growth and higher unemployment rates. Data shows that workers whose positions are highly exposed to AI and subsequently eliminated take, on average, about one month longer to find new employment, and their probability of remaining unemployed a year after displacement is about 10 percentage points higher than other laid-off workers.

Meanwhile, Goldman Sachs Chief Economist Jan Hatzius pointed out that for every 10-percentage-point increase in AI adoption, the corresponding industry experiences around a 0.9 percentage point decrease in employment growth and a 0.9 percentage point increase in unemployment.

Even more concerning is that the full force of this impact may not have arrived yet. Goldman Sachs economist Joseph Briggs warns that, historically, automation-driven layoffs often concentrate during economic recessions. While companies are currently deferring layoffs during periods of strong profits, a downturn could trigger a concentrated release of years-worth of AI-driven layoffs, amplifying the impact by coinciding with cyclical unemployment.

Previously, concerns about AI replacing jobs were widely questioned. Apollo analyst Torsten Slok stated in August that AI's impact on employment was "negligible," his research finding that AI mainly suppressed wages rather than job counts—positions most exposed to AI saw wage growth 6.7% slower than low-exposure jobs after 2023, but employment numbers did not show statistically significant change.

However, Goldman Sachs' latest data indicates a clear shift. Mei, utilizing new data from the U.S. Census Bureau’s Displaced Worker Supplement covering over 15,000 involuntarily unemployed workers from 2016 to 2025, defines the latest cohorts from 2023 to 2025 as "the first wave of AI-replaced workers." By creating a composite score based on occupational AI exposure, industry AI adoption rate, and the nature of job eliminations, Goldman Sachs isolated the top 30% most likely to be displaced by AI, comparing them to other laid-off workers.

The results show that, before 2024, differences were insignificant—yet by 2025, a clear divergence emerges: AI-displaced workers experience significantly longer reemployment periods, and their probability of long-term unemployment increases notably. Goldman Sachs points out that the short-term impact magnitude is "comparable to the effects of previous technological revolutions on unemployed workers," indicating that the assertion "AI only creates jobs but never destroys them" is no longer tenable.

The most unsettling finding in Goldman Sachs' research is not the prolonged unemployment duration, but the trajectory of reemployment. In past technological shifts, workers replaced by technology typically experienced "downward mobility," moving into more routine and less complex roles, with their original skills depreciating as technology evolves.

The current wave of AI-driven replacement reveals similar or even starker patterns. When seeking new jobs, AI-replaced workers tend to look for positions where their existing skills remain relevant, but this makes them even more vulnerable to the next wave of automation risk. According to Goldman Sachs data, the probability that AI-replaced workers end up in equally or more AI-exposed jobs after reemployment rose from a median level of about 55% between 2018 and 2023, to about 65% in 2024, and is expected to climb further to nearly 75% in 2025.

In other words, three-quarters of workers replaced by AI are stepping right into AI’s next hunting ground.

Data from the U.S. Bureau of Labor Statistics confirms this trend: between 2023 and 2025, 7.4 million workers were involuntarily unemployed, compared to 6.3 million in the previous survey cycle; and among long-term unemployed workers who found new full-time jobs, only about 49% earned wages at or above the level of their previous positions, down sharply from about 62% two years earlier.

Goldman Sachs' historical data shows that the impact of technological unemployment is long-term. Over several years following displacement, actual wage growth for technologically replaced workers remains persistently below that of other laid-off workers, and their probability of repeated unemployment stays high. Over the past forty years, older, non-college-educated, rural, and long-tenured workers experience cumulative real wage gaps of about 8 to 10 percentage points ten years after displacement—roughly twice as high as younger, college-educated, urban, and short-tenured counterparts.

However, new Goldman Sachs data reveals a relatively positive signal: younger workers adapt to AI shocks more quickly. In 2025, AI-displaced workers under 35 will take about three weeks longer to return to full-time employment than other unemployed, whereas those 35 and over take about five weeks longer, often undergoing more temporary or part-time transitions.

Goldman Sachs also notes that university students are "increasingly choosing majors with lower AI exposure." Yet, data from Stanford Digital Economy Lab shows that among 22- to 25-year-olds in AI-exposed occupations, employment levels are already about 19% below normal—a predicament caused not by slow adaptation, but by a lack of recruitment altogether.

Currently, the U.S. unemployment rate remains at a relatively low 4.2%, and the effect of AI substitutions is not yet fully reflected in macro data.

Goldman Sachs economist Briggs explains: Historically, the vast majority of job losses from routine position automation occurred during recessions. Data shows the proportion of U.S. workers in routine roles remains steady during each economic expansion and then falls stepwise during downturns—in the early 1980s, 2001, 2008, and 2020—and has never rebounded.

Using three historical estimation methods, Briggs finds that for every 1% improvement in labor productivity due to technological progress, the unemployment rate rises by about 0.3 to 0.4 percentage points at the peak, driven mainly by job losses. Based on Goldman Sachs’ forecast that AI will ultimately raise productivity by 15%, it is estimated that about 9% of workers will be displaced by AI during the transition, with an additional 3% facing employment contraction—totaling roughly 15 million U.S. workers, higher than the 6% to 7% estimated at the start of the year.

Briggs points out that companies tend to retain redundant positions in good times, but once the economy turns, they leverage downturns to accelerate automation. More alarmingly, the current AI capital spending boom has itself become a key growth driver, and Hatzius warns that this momentum is peaking—which means that the very trigger for the next recession could be AI itself.

Bank of America economists take a relatively moderate stance, suggesting "AI’s influence on the labor market may remain limited before 2026, with broader structural changes likely by 2027." Even so, this "optimistic scenario" would only delay large-scale shocks by about two years.

Of note, Goldman Sachs points out that currently only about 23% of U.S. companies integrate AI into some routine workflows—and this number "overstates the actual extent of automation currently underway."

Yet, even with this relatively limited adoption rate, AI has already caused statistically significant job losses outside of the technology sector, and unemployed workers re-entering the workforce are highly likely to remain at risk of replacement by AI.

If AI adoption were to double or even triple from current levels, each of these effects would be further magnified: more unemployed workers would vie for the limited number of "AI-hard-to-replace" roles, unemployment durations would lengthen, and workers entering new positions would face a fresh wave of displacement. Briggs concludes that the bulk of this impact may very well burst forth during the next recession—precisely when the economy is least able to absorb it.

Briggs warns that the Federal Reserve and other central banks "should closely monitor any emerging downside growth or labor market risks," and once current inflation constraints are gone, "a swift pivot to easier policy may help ensure automation-driven unemployment does not spike at the economy’s most vulnerable moment."