Among workers who left jobs highly exposed to artificial intelligence between 2019 and 2026, 41% landed in the same occupational cluster where they began, according to a new brief published by the Bipartisan Policy Center in 2025. Of those who stayed within their cluster, nearly half saw no meaningful reduction in their AI exposure, meaning their job changes didn't alter either their vulnerability to automation or the range of occupations available to them in the future. The report examines whether changing jobs actually expands workers' options or simply shuffles them among similarly exposed roles.
Workers who remained within their original cluster saw the AI exposure of jobs they could reach next drop by just 0.2 percentage points on average. By contrast, the 59% who crossed into a different occupational cluster landed in positions from which the next available jobs carried 6.5 percentage points less exposure on average. These cluster crossings flowed through a narrow channel: 57 "bridge occupations"—representing just 12% of the 472 occupations measured—carried 47% of all moves between clusters and 52% of crossings made by workers leaving highly exposed jobs. The largest group of workers who switched clusters, about a quarter of all those who left exposed occupations, moved into other cognitive roles such as business, finance, education, and computer jobs. Another 11% transitioned into physical and manual work, 9% into management, and 8% into service roles including food service and healthcare support.
The report identifies that most bridge occupations aren't highly vulnerable to AI themselves—only 10.5% qualify as highly exposed, compared with 19% of other occupations. However, two of the most broadly connected occupations in the labor market sit at or near the high exposure threshold: customer service representatives, which is highly exposed, and the broad "other managers" category, which sits just below the threshold at 48%. Both connect 15 of the 16 occupational clusters tracked. The authors note that for workers who already face limited alternatives, disruption to these transitional roles could substantially narrow their options. Workers in the administrative, legal, and office support cluster—which holds much of what the report calls the "trapped worker population"—made 10.1 million transitions into bridge occupations between 2019 and 2026, moving from jobs averaging 70% AI exposure to bridge roles averaging 42%.
Bridge occupations function as career infrastructure, yet the report notes they're rarely priorities for workforce investment. For most workers they serve as steps on a path rather than destinations, but they're involved in roughly half of all cluster crossings. The occupations tend to be physical and generalist, meaning that crossing the labor market typically means passing through work built on transferable, hands-on tasks rather than specialized cognitive ones. The report argues that as AI capabilities advance and labor markets shift, forecasting how well the market can absorb displaced workers will require monitoring which occupations carry the most traffic and watching the exposed ones closely.
In 2025, the direction of worker movement reversed for the first time in the data. Net flow into AI-exposed occupations turned negative—1.6 million more workers left these jobs than entered them—after years of steady decline from a peak of 2.2 million net inflow in 2020. The swing from 2024, when net inflow stood at just 0.2 million, was 1.8 million, more than twice the largest annual decline in the preceding five years. The authors caution that the 2025 figure covers only 11 months due to a government shutdown that interrupted data collection, and that several labor market shocks including tech-sector layoffs and federal workforce reductions could account for part of the shift. Data through February 2026 shows continued net outflow at a slower pace, but the authors say two months can't establish a trend and it's not yet clear whether 2025 marks a temporary shock or the start of a more durable change in where workers land.