Is AI Actually Taking Jobs? Anthropic’s New Study Reveals the Truth

Why It Matters

The findings signal that firms and policymakers should prepare for targeted workforce shifts—retraining, hiring strategy changes and productivity adjustments—rather than broad job destruction, with implications for education, labor markets and corporate talent planning.

Summary

Anthropic’s new study introduces “observed exposure,” a metric comparing AI’s theoretical capabilities to how it’s actually used on the job, and finds that only a fraction of automatable tasks are currently being performed by AI. Jobs with higher observed exposure—notably computer programmers, computer service reps and data-entry keyers—are projected to grow more slowly over the next decade, though there has been no large spike in unemployment to date. Workers in highly exposed roles tend to be older, more often female, better educated and higher paid, and entry-level hiring (ages 22–25) into these occupations has dipped slightly. Overall, the study suggests AI’s workplace impact is uneven: concentrated in certain roles and affecting hiring patterns more than widespread immediate layoffs.

Original Description

A new study from Anthropic analyzed how AI is impacting jobs using a new metric called observed exposure. The result? AI could automate many tasks, but real workplace usage is still much lower. Job losses haven’t spiked yet, but hiring trends may already be shifting.