Labour market impact of AI: IT, Finance and Legal are the most exposed sectors

A recent study by Anthropic introduces a new way to measure AI’s impact on the workforce. Instead of just looking at what large language models (LLMs) can theoretically do, the researchers created a metric called “Observed Exposure.” This combines theoretical capabilities with actual, real-world usage data from their AI assistant, Claude.

How is "Observed Exposure" measured?

To move beyond predicting what AI might do, the researchers built this metric by combining O*NET occupational task lists, existing theoretical AI capability estimates, and real-world usage data from the Anthropic Economic Index. An occupation scores higher if its core tasks are actually being executed by Claude in professional, work-related contexts. The calculation weights how the AI is used—giving full weight to fully automated workflows and half weight to augmentative uses (where AI simply assists the worker)—and then averages the score based on the fraction of time a worker typically spends on those specific tasks.

Here is a brief breakdown of their early findings on how AI is actually reshaping the labor market.

Which sectors and workers are most affected?

As shown in the radar chart above, the gap between what AI can do (blue) and what it is actually used for (red) is still large. However, adoption is heavily concentrated in knowledge-work sectors.

• Top Sectors: Computer & Math (IT), Business & Finance, Management, Legal, and Office & Admin show the highest levels of both theoretical capability (>90% of these tasks can be substituted by AI) and real-world AI usage.

• Top Occupations: Computer programmers, customer service representatives, data entry keyers, and financial analysts are currently the most exposed specific roles.

• Worker Demographics: The workers in the top quartile of AI exposure do not fit the traditional vulnerable worker profile. They are significantly more likely to be older, female, higher-paid, and highly educated (almost four times as likely to hold a graduate degree compared to unexposed workers).

  
  

Is AI causing unemployment right now?

The short answer is: not yet, but..... By tracking data from the Current Population Survey since late 2022, the researchers found no systematic increase in unemployment for highly exposed workers.

However, AI does appear to be subtly shifting the pipeline for new talent. While existing workers aren't losing their jobs, there is suggestive evidence that hiring for young workers (ages 22–25) has slowed down. New job starts for young adults in highly exposed occupations dropped by roughly 14% post-ChatGPT compared to 2022. Furthermore, the Bureau of Labor Statistics projects weaker job growth through 2034 for the occupations seeing the highest AI usage today.

The Bottom Line and Outlook

AI is currently still operating at only a fraction of its theoretical capacity. So far, its impact on the labor market looks more like a quiet restructuring, primarily reducing the hiring of entry-level workers. However, a look into the future paints a much more pessimistic picture: Once real-world AI usage (the red area in the chart) continues to grow and reaches its maximum theoretical potential, we must assume that the number of workers needed in highly exposed sectors will drop massively. When AI systems become capable of fully automating core tasks in IT, finance, and administration, the current stability in unemployment figures is likely just the calm before the storm, preceding widespread job cuts in these traditionally secure knowledge professions.