Anthropic’s Economics team has published a working paper, “Economic Scenarios for Transformative AI”, alongside an interactive scenario explorer examining how AI capabilities and adoption could affect GDP, wages, employment, and the labor share through 2030.

The framework, developed using a task-based macroeconomic model, is not intended to make predictions. Instead, it translates a small set of measurable variables into comparable economic outcomes. These variables include the share of tasks AI can perform, the extent of deployment, productivity gains per task, and the balance between automation and augmentation. That structure makes the paper’s assumptions especially important: changes in observed AI use, task capabilities, worker transitions, and the supply of capital can produce materially different outcomes.

The paper presents three scenarios. In the modest scenario, AI’s economic impact resembles that of the internet. GDP in 2030 is 1.6% above its no-AI trajectory, annual growth reaches 2.4%, and unemployment increases by only one-tenth of a percentage point.

The substantial scenario reflects forecasts circulated by financial institutions in 2023. Under this path, AI performs roughly half of knowledge work by 2030, GDP is 8.3% higher than on the no-AI path, and annual growth reaches 5.4%. Knowledge-worker wages remain essentially flat, while wages in other areas increase.

The extreme scenario assumes that recursively self-improving AI is adopted rapidly. GDP growth reaches 15.4% per year and the economy becomes 32% larger, leaving society significantly wealthier. However, the labor share falls from 60% to 45%, cognitive wages decline 11.5% below trend, cognitive unemployment reaches 17.9%, and overall unemployment rises to 11.9%—above postwar records.

The model also finds that AI’s contribution to innovation-driven growth remains limited even in the extreme scenario because research continues to be constrained by physical tasks.

Anthropic’s Economics team is sharing a new model of how AI might affect economic growth, jobs, wages, and more by 2030. Explore the scenarios, tell us what you think will happen, and see how your answers compare to more than 10,000 Americans. — Anthropic (@AnthropicAI) September 9, 2026

Public Expectations Align With the Substantial Scenario

To complement the model, Anthropic surveyed 10,980 US adults in August about AI capabilities, adoption, productivity effects, and prospects for re-employment. The median respondent expects AI to perform six of eight benchmark tasks by 2030, ranging from routine business correspondence to software development. Respondents also expect AI to be deployed on 40% of feasible tasks and estimate that a displaced worker would need approximately eight months to find employment in a new occupation.

When these median responses are entered into the model, they generate outcomes close to the substantial scenario: GDP reaches 8.6% above the no-AI path and unemployment stands at approximately 4.6%. Expectations vary considerably, however. About 10% of respondents express views consistent with the extreme scenario, while 40% say Nobel-level discoveries driven by AI will never occur.

The authors note that nearly all of the divergence among the scenarios emerges after 2027 because the paths share current measurements of AI use. The sensitivity analysis identifies two major variables. The elasticity of capital supply affects whether workers or capital owners receive the gains, while wage rigidity affects whether cognitive workers bear the costs through lower wages or increased joblessness. As the scenarios develop, these measures—and the time required for displaced workers to move into new occupations—are the main conditions to watch when comparing outcomes with the model’s paths.

In the extreme scenario, total labor income remains roughly unchanged even though GDP is one-third larger. As a result, nearly all gains accrue to capital. Compensating knowledge workers would require transfers equivalent to about 9% of GDP, approximately the combined scale of Social Security and Medicare. The paper says there is no historical precedent for technology-driven redistribution on that scale.

According to Anthropic, the framework is intended to inform the company’s research funding and policy proposals aimed at ensuring that AI’s economic benefits are broadly shared.