Anthropic model projects severe knowledge worker job losses in extreme AI scenario

Anthropic's economics team released an interactive model this week projecting how AI could reshape the U.S. economy through 2030, with its most extreme scenario forecasting 17.9% unemployment among knowledge workers and economy-wide joblessness of 11.9% — levels beyond anything seen in a typical recession.officechai+1

The release landed alongside a separate and jarring admission from within the company: Evan Hubinger, Anthropic's alignment science lead, posted on X that "we really do earnestly believe AI could kill all humans," putting his personal estimate of that risk at greater than 10% over the next decade.bbc+1

Three Paths to 2030

The tool, called the Economic Scenario Explorer, breaks every job into a bundle of tasks drawn from the U.S. Department of Labor's O*NET taxonomy, then models how AI might automate, augment, or replace each one across the roughly $30 trillion in task-value the economy generates annually.unite+1

Anthropic sketches three scenarios. In the modest case, AI acts like the internet — a gradual productivity boost that lifts GDP to $34.1 trillion by 2030, just 1.6% above a no-AI baseline. In the substantial scenario, AI handles half of knowledge work mostly autonomously, pushing GDP to $36.3 trillion — an 8.3% gain — while knowledge worker wages go flat. The extreme scenario assumes AI becomes more productive than humans at nearly all knowledge tasks, with essentially no new work created for displaced workers. GDP reaches $44.4 trillion, a 32.4% jump, with annual growth of roughly 15%.officechai+1

Winners, Losers, and the Capital Shift

Even in the extreme scenario where the economy grows rapidly, the gains tilt sharply toward capital owners. Labor's share of GDP falls from roughly 60% to 45.2%, and knowledge worker wages drop 11.5% below where they would be without AI, even as wages in physical occupations like construction rise 33.6%. Anthropic's model suggests coders and call center agents could be forced to retrain as electricians or nurses, with slow occupational switching leaving many stranded between jobs.officechai+1

A companion survey of more than 10,000 Americans found that the average respondent's expectations align with the substantial scenario — implying about 10% higher GDP by 2030 and unemployment near 5%. Only around 10% of those surveyed held expectations matching the extreme outcome.unite+1

The Elephant Not in the Model

The model was built with input from economists including Daron Acemoglu, David Autor, and Emi Nakamura, and Anthropic acknowledges it omits policy responses, business cycles, and the possibility of advanced robotics. It also does not account for a potential AI investment bubble or, notably, for existential risk from the technology itself — a scenario Hubinger's post made harder to ignore. As he wrote, Anthropic "does not yet have a plan to solve alignment for superintelligence and are not clearly on track to".x+2