Anthropic launches the Economic Scenario Explorer: three AI growth scenarios and their blind spots

On 10 September 2026 Anthropic published the interactive Economic Scenario Explorer on its Anthropic Institute site, along with the technical report “Economic Scenarios for Transformative AI” (Korinek et al., 2026). The model, designed by Anton Korinek, Chad Jones, Szymon Sacher, Tess Cotter and Peter McCrory, lets you vary assumptions about model capability, speed of adoption, autonomy and productivity gains, and then watch the simulated effects on US GDP, employment, wages and labour income in 2030.

Three scenarios come out of the analysis. The “modest” one puts US GDP at $34.1 trillion (+1.6% against a path without AI), an impact comparable to the internet's. The “substantial” one takes GDP to $36.3 trillion (+8.3%) and assumes AI does half of all cognitive work by 2030, with knowledge workers' wages essentially flat, unemployment rising by around 5% and an overall gain for non-cognitive workers. The “extreme” one pushes GDP to $44.4 trillion (+32.4%), but the share of GDP going to labour drops to 45.2% (from roughly 60%) while capital's share climbs to 54.8%, with knowledge workers' wages down by more than 10% and unemployment at levels typical of historic recessions.

The model rests on a survey of more than 10,980 Americans conducted in August 2026 (source: The Next Web) and on a representation of jobs as “bundles of tasks”. The full questionnaire does not appear to have been published, and it is not verified whether the calibration went through independent peer review — comments from outside economists on a draft are not a peer review, and Anthropic says so itself. Anthropic spells out the limits: no economic policy responses, no business cycles, no aggregate demand shocks, no catastrophic risks, no hyper-capable robotics, no data-centre impacts. GDP is computed from the supply side rather than from consumer spending, a methodological choice that is not compared with other macroeconomic approaches.

None of the three scenarios is a forecast: they are simulations conditional on the assumptions the user sets, and Anthropic assigns them no probability. Outside economists, among them Daron Acemoglu, David Autor and David Romer, commented on an early draft but did not endorse the conclusions. Ben Moll (LSE) and Alex Imas (University of Chicago) have publicly challenged the double-digit growth projections, calling them “extremely unlikely” for the next 10 to 15 years and pointing to a more reasonable 4-5% for the decade after that. According to The Next Web, they have also publicly bet that US GDP per capita growth will stay below 15% a year through 2033.

The report itself admits its setup is “a stark simplification of a complex reality”. The answers of the more than 10,980 people surveyed land closer to the “substantial” scenario than to the “extreme” one. In short, the Economic Scenario Explorer is a tool that makes the consequences of different assumptions visible, but its estimates depend on unverified assumptions and on data that is only partly published. The built-in conflict of interest — Anthropic sells the models whose impact it is estimating — calls for a critical reading from policymakers and researchers.