AI risks ‘Great Recession’ for white-collar workers, Anthropic warns

NEW YORK, UNITED STATES — A new Anthropic research finds that while artificial intelligence (AI) is currently used for only a fraction of white-collar tasks, its theoretical capability to perform far more professional work raises concerns about future labor disruption.

The research introduces a new metric, “observed exposure,” which compares actual AI usage with its potential. The findings suggest that highly educated, high-earning professionals could face a disruption comparable to past economic downturns if AI adoption approaches its technical limits.

“We introduce a new measure of AI displacement risk, observed exposure, that combines theoretical LLM capability and real-world usage data, weighting automated (rather than augmentative) and work-related uses more heavily,” the researchers wrote. 

The gap between AI potential and workplace adoption

Anthropic’s research, authored by Maxim Massenkoff and Peter McCrory in a report titled “Labor Market Impacts of AI: A new measure and early evidence,” establishes a stark divide between what AI can theoretically do and what it is actually doing in the workplace. 

By analyzing real-world usage data from Anthropic’s Claude model, the study found that while large language models (LLMs) are technically capable of handling 94% of tasks for computer and math workers, Claude currently performs about 33% of those tasks in observed professional usage.

Reporting on the study, Fortune notes that similarly, in office and administrative roles, AI’s theoretical capability stands at 90%, yet actual integration remains a fraction of that potential. 

This “observed exposure” metric, derived directly from professional interactions with Claude, provides the first data-driven map of AI’s real-time penetration into the labor market.

The researchers attribute this lag to legal constraints, model limitations, the need for additional software infrastructure, and continued human oversight. The study also notes a significant gap between AI’s technical potential and its real-world use.

“Many tasks, of course, remain beyond AI’s reach—from physical agricultural work like pruning trees and operating farm machinery to legal tasks like representing clients in court,” the report states.

However, for knowledge-based professions, the findings indicate that current limits on AI adoption may offer only temporary insulation, warning that automation risks will continue to grow as adoption expands.

High-income jobs most at risk of AI automation

Contrary to popular narratives that emphasize risk to manual laborers, Anthropic’s analysis identifies high-income, highly educated professionals as among the workers most exposed to AI capabilities.

The study reveals that the group most exposed to AI is 16 percentage points more likely to be female, earns 47% more than the least exposed group, and is nearly four times as likely to hold a graduate degree. This category encompasses lawyers, financial analysts, and software developers—roles heavily reliant on cognitive tasks. 

Computer programmers, customer service representatives, and data entry keyers top the list of most exposed occupations, while jobs requiring physical presence, such as cooks and mechanics, show zero exposure. 

The report outlines a scenario it calls a ‘Great Recession for white-collar workers,’ where unemployment in the top quartile of exposed occupations could theoretically double.

Although systematic mass layoffs have yet to occur, the study provides early indications of the hiring trend, specifically among younger workers entering the workforce. 

In the post-ChatGPT era, the job-finding rate in AI-exposed occupations has fallen by about 14% relative to 2022. This concurs with individual results, emphasizing a 16% decrease in employment among employees aged 22 to 25 in AI-prone jobs, indicating that the first effect would be a slowing of new job recruitment rather than a mass-firing spurt. 

The loss of 92,000 jobs reported in February by the United States Bureau of Labor Statistics and Block layoffs citing AI remain small relative to the broader labor market.

However, the researchers suggest the early impact may appear in a narrower hiring pipeline for young professionals, stating they “may be remaining at their existing jobs, taking different jobs, or returning to school.”

As AI capability rapidly outpaces its workplace adoption, the future of work may be defined less by sudden mass layoffs than by a gradual tightening of entry-level opportunities and a structural reshaping of professional careers.