The specter of job displacement, once primarily associated with physical labor, has shifted dramatically, now casting a long shadow over highly educated, well-compensated professionals. Anthropic, an artificial intelligence company known for its Claude model and its focus on AI safety, recently published a comprehensive study detailing the current and potential impact of AI on the workforce. Their findings suggest a significant disparity between what AI is theoretically capable of doing and its actual application, a gap that, if closed, could trigger a “Great Recession for white-collar workers.”
This new research, titled “Labor market impacts of AI: A new measure and early evidence,” introduces a metric called “observed exposure.” This innovative approach directly compares AI’s theoretical capabilities with real-world usage data gleaned from Claude’s interactions in professional environments. What emerges is a striking picture: AI is currently performing only a fraction of the tasks it could theoretically handle. This underutilization, however, is not expected to last, and when the gap narrows, the demographic most at risk diverges sharply from popular perception.
Contrary to the common image of manufacturing or service industry workers being most vulnerable, the study highlights a different cohort. The group most exposed to AI’s capabilities is 16 percentage points more likely to be female, earns 47% more on average, and possesses nearly four times the likelihood of holding a graduate degree compared to the least exposed segment. These are the lawyers, financial analysts, and software developers, not the warehouse staff. Occupations such as computer programmers, customer service representatives, and data entry keyers are identified as particularly susceptible.
The researchers, Maxim Massenkoff and Peter McCrory, point to several factors contributing to the current lag in AI adoption. Legal constraints, technical hurdles like model limitations, the necessity for additional software tools, and the ongoing requirement for human oversight all play a part. Yet, these are largely seen as temporary impediments. The study projects that as these barriers diminish, the “red area” of actual AI usage will expand to fill the “blue area” of what is technically possible.
Consider the field of computer and math workers, where large language models are theoretically capable of handling 94% of tasks. Yet, in observed professional use, Claude currently covers just 33% of these. A similar pattern holds for office and administrative roles, with a 90% theoretical capability contrasting sharply with a much smaller fraction of actual deployment. This substantial difference underscores the latent potential for disruption. However, it’s also worth noting that approximately 30% of workers, including cooks, mechanics, bartenders, and dishwashers, appear to have zero AI exposure, given their roles demand a physical presence that current large language models cannot replicate.
The implications of this potential shift are profound. The researchers draw a parallel to the 2007–2009 financial crisis, which saw the U.S. unemployment rate double from 5% to 10%. They suggest that a comparable doubling in the top quartile of AI-exposed occupations, from 3% to 6%, would be clearly detectable within their framework. This scenario, while not yet fully materialized, is gaining traction as a serious possibility across various analyses, extending beyond more speculative doomsday narratives. Federal Reserve Governor Michael S. Barr, for instance, outlined this as one of three potential outcomes for AI adoption in a recent speech.
While some companies, like Jack Dorsey’s Block, have already cited AI as a rationale for significant workforce reductions, the broader picture is more nuanced. The Anthropic research suggests that for younger workers, the primary impact might not be immediate layoffs but rather a slowdown in hiring within AI-exposed fields. They found a 14% drop in the job finding rate in these occupations in the post-ChatGPT era compared to 2022, although they caution that these findings are only marginally statistically significant. There has been no systematic increase in unemployment across the board, according to the study. Nevertheless, this subtle shift could signal a new reality for employment in the age of artificial intelligence, potentially leading young professionals to explore different career paths or pursue further education.
