Anthropic study warns AI job losses rival the Great Recession

Anthropic researchers have published data showing AI can theoretically handle 94% of tasks in computer and math roles but currently completes only 33% in practice -- a gap that makes the scale of eventual displacement harder to predict and easier to underestimate. The study, "Labor market impacts of AI: A new measure and early evidence," by Maxim Massenkoff and Peter McCrory, introduces an "observed exposure" metric comparing what AI is capable of against what employers are actually deploying, and finds the coming wave has barely started. Meanwhile, early signals are already visible: the February 2026 jobs report recorded 92,000 positions shed, unemployment rose to 4.4%, and job-finding rates for workers aged 22-25 in AI-exposed fields dropped 14%.

The research carries unusual weight because it comes from the company building the systems driving this displacement. The workers most at risk are not low-wage laborers -- they earn 47% more than the median, are 16 percentage points more likely to be female, and are nearly four times as likely to hold graduate degrees. In a scenario the researchers describe as analogous to the Great Recession, unemployment in the top quartile of AI-exposed occupations could double from 3% to 6%. That ceiling has not been reached because legal, regulatory, and technical constraints still limit deployment -- but those constraints are narrowing, and office and administrative roles face a theoretical AI capability ceiling of 90%.