Anthropic's "Observed Exposure" Study Is the First Real Early-Warning System for AI Labor Disruption

Anthropic published a dataset based on millions of real workplace interactions with Claude, revealing that AI-exposed roles earn 47% more than low-exposure roles—reversing historical automation patterns. A significant gap exists between theoretical automation capability (94% for computer/math occupations) and actual observed usage (33%), representing an 'acceleration zone' where adoption will catch up. The study also found a 16% decline in junior hiring in AI-exposed fields, signaling the collapse of traditional apprenticeship pathways, and outlines three phases of disruption from early exposure through structural reorganization by 2038.