As AI systems enter more workplaces, their labor market impact could range from mild productivity gains to serious social disruption. Recent research from Anthropic tries to move beyond speculation, by comparing AI's theoretical capabilities with Claude use in actual work-related tasks. Their findings suggest that current adoption is still far below the technical potential, but that some occupations are already rather exposed.

In our next benkyoukai, we will use this research as a starting point for discussing the labor market impacts of AI. The session may touch on Anthropic’s Economic Index, early evidence on unemployment and hiring, and what today’s usage patterns can tell us about future disruption. Join us to discuss one of the most concrete ways AI may reshape society!

This report introduces “observed exposure”, a new measure for estimating the labor market impact of AI by combining theoretical task-level capability with real-world usage data from Claude. Rather than asking only which tasks language models could perform, the measure emphasizes tasks that are already appearing in professional use, especially automated rather than augmentative applications. This allows the authors to distinguish between AI’s technical potential and its current deployment in actual work settings.

Using U.S. occupational and labor market data, the report finds that current AI use remains far below its theoretical potential, though exposure is concentrated in specific professions such as programming, customer service, and financial analysis. The authors find no systematic increase in unemployment among highly exposed workers since the release of ChatGPT, but report suggestive evidence that hiring of young workers has slowed in exposed occupations. The report presents these results as an early measurement framework rather than a definitive estimate of AI’s long-term labor market effects.

— Maxim Massenkoff and Peter McCrory, Labor market impacts of AI (2026)
— Anthropic, AI Economic Index (2026)