Ted Hisokawa
Mar 05, 2026 20:24
New Anthropic research measuring real AI usage finds no unemployment spike in exposed jobs, though hiring of workers 22-25 in AI-vulnerable roles shows concerning decline.
Anthropic has released what may be the most comprehensive analysis yet of AI’s actual impact on employment—and the headline finding challenges both doomsayers and optimists. Using a new metric called “observed exposure” that tracks real-world AI usage rather than theoretical capability, researchers found no systematic increase in unemployment for workers in AI-vulnerable professions since ChatGPT’s November 2022 launch.
But buried in the data is a warning sign: hiring of workers aged 22-25 into highly exposed occupations has dropped roughly 14% compared to pre-ChatGPT levels.
The Gap Between Theory and Reality
The study’s core innovation is measuring what AI actually does versus what it theoretically could do. Computer programmers top the exposure list at 75% task coverage, followed by customer service representatives and data entry keyers at 67%. Yet even in computer and math occupations—where 94% of tasks could theoretically be accelerated by AI—Claude currently covers just 33%.
“AI is far from reaching its theoretical capability,” the researchers note. Legal constraints, software requirements, and human verification steps all slow adoption. A pharmacist’s prescription authorization task, for instance, scores as fully automatable in theory but shows zero AI usage in practice.
Who’s Most Exposed?
The demographics of vulnerability may surprise some. Workers in the top quartile of AI exposure are 16 percentage points more likely to be female, earn 47% more on average, and hold graduate degrees at nearly four times the rate of unexposed workers. The unexposed group includes cooks, motorcycle mechanics, bartenders, and lifeguards—jobs requiring physical presence or manual dexterity.
Bureau of Labor Statistics projections through 2034 align with Anthropic’s findings: for every 10 percentage point increase in observed AI exposure, projected job growth drops 0.6 percentage points.
The Youth Employment Question
The study’s most actionable finding concerns young workers. While overall unemployment rates for exposed occupations remain stable, job-finding rates for 22-25 year-olds entering these roles have declined. Workers over 30 show no such pattern.
This echoes February 2025 research from the Dallas Fed showing employment in the most AI-exposed sectors declined 1% since late 2022, and separate findings that job postings for highly automatable roles dropped 13% post-ChatGPT while AI-enhanced positions grew 20%.
Anthropic researchers acknowledge limitations: young workers not finding jobs in exposed fields may be staying in school longer, taking different roles, or leaving the workforce entirely rather than appearing as unemployed.
What Would Sound the Alarm?
The framework can detect differential unemployment increases around 1 percentage point. A “Great Recession for white-collar workers”—doubling unemployment in exposed occupations from 3% to 6%—would be visible. Mass layoffs affecting the top 10% of exposed workers would push aggregate unemployment from 4% to 13%.
Neither scenario has materialized. The researchers plan periodic updates as AI capabilities advance and adoption spreads, noting their approach “won’t capture every channel through which AI could reshape the labor market” but should identify disruption before it becomes obvious in headline statistics.
Image source: Shutterstock
