Anthropic releases an economic scenario explorer

Anthropic released version 1.0 of an interactive tool modelling AI’s effect on the US economy through 2030, built by its economics team and led by economists Anton Korinek and Chad Jones. Rather than offering a single forecast, the tool lets users explore three named scenarios — modest, substantial and extreme — that vary by how quickly AI capabilities improve, how widely they are adopted, and how much they raise productivity, and represents jobs as bundles of tasks that AI can automate, augment or leave unchanged.

The scenarios span a wide range. By 2030, US GDP ends between 1.6% and 32.4% above a no-AI baseline across the three cases, reaching roughly $44 trillion in the extreme scenario. The accompanying technical report also projects how the gains would be split: labour’s share of output, currently around 60%, falls to as low as 45.2% in the extreme case, with capital’s share rising to 54.8%.

The report was reviewed by fifteen outside economists, including Daron Acemoglu and David Autor, though Anthropic noted the reviewers were not asked to endorse its conclusions. As with any long-range economic projection, the figures are model outputs conditioned on assumptions about adoption speed that are themselves uncertain, and the wide spread between the modest and extreme scenarios — a factor of twenty on the GDP effect — illustrates how much of the range is driven by those assumptions rather than by a settled estimate of AI’s productivity effect.

The release is notable less for pinning down a number than for coming from a frontier lab with a direct commercial stake in the technology it is modelling, publishing a tool that makes the uncertainty in such projections explicit rather than collapsing it into a single headline figure — a contribution to the broader debate about AI’s effect on jobs and growth running alongside disclosures of labour-market impact from other labs and researchers.