Anthropic built an economic model that asks a single question: if AI keeps getting more capable, what happens to jobs, wages, and growth by 2030. The answer is not one number but three branching scenarios, and the company that sells the technology is the one drawing the branches.
The modest scenario adds 1.6 percent to GDP relative to a world without AI, landing at $34.1 trillion. The substantial scenario adds 8.3 percent, for $36.3 trillion. The extreme scenario, built around rapid adoption and recursive self-improvement, adds 32.4 percent, reaching $44.4 trillion. Anthropic frames all three as modeled outcomes, not predictions, and the gap between the modest and extreme cases is the entire story: identical framework, wildly different endings depending on assumptions fed in.
Growth is not the variable that should worry a knowledge worker. Distribution is. Anthropic’s model has labor’s share of GDP, currently around 60 cents of every dollar, sliding toward capital as automation spreads. Wages for knowledge workers stay flat under the substantial scenario. Under the extreme scenario, those wages fall more than 10 percent by 2030 even as the overall economy is far larger. Workers in occupations AI barely touches, construction among them, see wages rise as productivity gains elsewhere spill into demand for their labor.
Unemployment tells a similar story, split by occupation rather than smooth across the workforce. In the extreme scenario, Anthropic’s model has unemployment among knowledge workers spiking to levels unseen in modern US history, driven by workers unable to quickly retrain into occupations less exposed to AI, such as electrician or nursing roles. The model estimates that by 2030, only 37.8 percent of workers remain in occupations untouched by AI automation, down from 62.2 percent of the workforce classified as knowledge workers today.
Anthropic paired the model with a survey of more than 10,000 Americans conducted in August. The typical respondent’s expectations track closest to the substantial scenario: a 10 percent GDP lift and unemployment near 5 percent by 2030. Roughly one in ten respondents hold expectations consistent with the extreme case. That gap between public sentiment and the company’s most dramatic scenario matters because it is Anthropic, not an independent economics department, that built the assumptions driving the extreme outcome.
The model has a documented list of what it leaves out: business cycles, policy responses, financial-market shocks, and any scenario involving advanced robotics. Anthropic’s own external reviewers, a group that included MIT’s David Autor and Stanford’s Pete Klenow, pushed back on parts of the framework, with some arguing the extreme scenario reads better as a thought experiment than a forecast and others saying the modest scenario understates effects already visible in the data. Anthropic incorporated some of that feedback, including a channel where returns to capital rise as AI adoption spreads, but left other criticisms, such as the model’s inability to track individual displaced workers, unresolved.
The company says this modeling will shape the research Anthropic funds on labor-market disruption and the policy proposals it plans to advance. That is worth watching alongside the numbers themselves: a company building frontier models is also building the analytical case for how those models should be regulated, and the assumptions baked into its own scenario explorer will shape which interventions get proposed. Operators evaluating AI adoption timelines should treat the extreme scenario as a stress test for their own workforce plans, not a baseline expectation, and revisit it as Anthropic’s team updates the model with future evidence.
Reported by Anthropic’s Economics team via the Anthropic Institute, published September 2026 as version 1.0 of its Economic Scenario Explorer, drawing on the technical report “Economic Scenarios for Transformative AI” (Korinek et al., 2026).
