Anthropic's New Measure Finds AI Hasn't Yet Displaced Workers — But Could

Anthropic is trying to build a better early warning system for AI-driven job displacement. The company's new research, published this week, introduces a metric called "observed exposure" — combining theoretical AI capability with real-world usage data from Claude — and the early results are cautiously reassuring.

According to the company's research, actual AI adoption remains "a fraction of what's feasible." In other words: AI can do a lot more than it's currently doing. The gap between theoretical capability and real-world usage is significant across nearly every occupation studied.

According to the findings, occupations with higher observed exposure are projected by the U.S. Bureau of Labor Statistics to grow less through 2034. Workers in the most exposed professions tend to be older, female, more educated, and higher-paid — counter to the intuition that AI will primarily affect entry-level roles.

But here's the key finding: according to Anthropic, there's no systematic increase in unemployment for highly exposed workers since late 2022. The researchers did find "suggestive evidence" that hiring of younger workers has slowed in exposed occupations — a signal worth watching.

The researchers themselves are careful. They frame this as groundwork laid before meaningful effects emerge, not a definitive verdict. Their stated goal is to establish an approach for measuring AI's labor market impact and revisit it periodically — to identify disruption earlier than past attempts, which famously overshot.