Published in March 2026 by Anthropic economists Maxim Massenkoff and Peter McCrory, this working paper introduces a new framework for measuring AI's actual impact on labour markets. Rather than relying on theoretical capability assessments alone, the authors construct an 'Observed Exposure' metric that combines data on what tasks AI can theoretically perform with real-world usage data from Anthropic's Economic Index.

​

The key finding is that AI is far from reaching its theoretical potential: actual coverage of job tasks remains a fraction of what models could in principle handle. That said, occupations with higher observed AI exposure are projected by the US Bureau of Labor Statistics to grow less through 2034, suggesting the measure tracks real economic signals.

​

The paper also profiles which workers are most exposed: those in high-exposure roles tend to be older, female, more educated and higher-paid. Despite this, the authors find no systematic increase in unemployment for highly exposed workers since late 2022. There is, however, suggestive evidence that hiring of younger workers has slowed in exposed occupations.

​

For destination organisations and DMOs thinking about workforce planning, the paper provides one of the most rigorous early frameworks for understanding where AI displacement risk is actually concentrated, and how to monitor it before effects become visible in aggregate employment data.

• Key findings and overview of method
• The case for a new exposure measure: limitations of prior approaches
• Measuring exposure: combining O*NET, Eloundou et al. and Anthropic usage data
• Observed Exposure: the new measure explained
• Theoretical capability vs actual observed exposure by occupation
• How exposure tracks with projected BLS job growth (2024-2034)
• Worker characteristics in high and zero exposure occupations
• Unemployment analysis: evidence to date
• Hiring slowdown: suggestive evidence for younger workers
• Appendix: mathematical details