Anthropic: AI mass unemployment only in "extreme" 2030 scenario

Anthropic's report frames the AI-and-jobs question not as one outcome but as three, sorted by how quickly AI capabilities improve and spread. In the "measured" case, 2030 GDP rises 1.6% above a no-AI baseline; in "large-scale," 8.3%; in "extreme," 32.4%. The wage picture diverges more sharply than the growth picture: under "large-scale," knowledge-worker wages are essentially flat at 0.3% lower while other workers gain 5.9%, with overall unemployment at 4.6%. Only in "extreme" do knowledge-worker wages fall 11.5% and unemployment climb to 17.9% for knowledge workers and 11.9% overall, with labor's share of income dropping to 45.2% from 59.4% in the "measured" case.

The researchers are explicit about the boundaries of their own exercise. They say the forecasts leave out policy responses, business cycles, financial-market disruption, and the arrival of "ultra-high-performance robots," and that the model, like other economic models, has its own limitations. That caveat matters because the gap between "large-scale" and "extreme" is where the wage and unemployment damage concentrates, so whether that gap is reached is likely to hinge on factors the report sets aside rather than on AI capability alone.