An economy can become dramatically richer while leaving workers with a shrinking share of its prosperity.
SAN FRANCISCO, UNITED STATES
Anthropic has published an economic model exploring how artificial intelligence could transform growth, employment and income distribution in the United States by 2030. Its most extreme scenario produces annual economic growth of 15.4%, but also eliminates more than one-fifth of cognitive jobs and raises overall unemployment to levels associated with a severe recession.
The company stresses that its scenarios are not predictions and assigns no probability to them. Researchers modelled economic activity as groups of tasks that AI could ignore, assist, automate or create. Outcomes change according to assumptions about technological capability, business adoption, worker productivity and the time displaced employees require to find new occupations.
Under the modest scenario, AI performs only a limited share of economic activity. Gross domestic product would be 1.6% higher in 2030 than without AI, annual growth would reach 2.4% and cognitive employment—covering management, professional, sales and office occupations—would decline by only 0.5%.
The substantial scenario assumes that AI becomes capable of completing half of knowledge-based tasks, although most work remains primarily human. Annual growth would rise to 5.4%, GDP would be 8.3% above the baseline and cognitive employment would fall by 3.9%. Unemployment among office workers would reach 4.5%, while wages would begin diverging between occupational groups.
The extreme scenario assumes highly autonomous and potentially self-improving AI capable of performing almost all knowledge work without creating equivalent new roles for people. GDP would finish 32.4% above the no-AI trajectory and double approximately every four and a half years. Cognitive employment would fall by 21.5%, unemployment among those workers would reach 17.9% and total unemployment would rise to 11.9%.
Distribution represents the model’s most consequential finding. Office wages would decline by 11.5%, while earnings in less exposed occupations could rise by 33.6%. Labour’s share of national income would fall from 60% to 45.2%, while capital income would increase by more than 80%. Economic output would expand rapidly, but a growing proportion of the gains would flow to investors and asset owners.
Current labour-market evidence does not demonstrate displacement on anything approaching that scale. Anthropic’s earlier research found no statistically clear increase in unemployment among the most AI-exposed professions, although it identified tentative evidence of slower hiring for younger workers. The scenarios should therefore be understood as stress tests showing what could happen under particular assumptions, not as forecasts of an inevitable future.
The central policy challenge is not simply preparing people to use AI. Governments must consider how productivity gains will be distributed, how workers will transition between occupations and how social protection will function if employment no longer determines access to economic security.
Technological progress creates prosperity only when its benefits remain connected to human wellbeing.