Anthropic has published an economic model projecting AI's impact on the US labor market through 2030. It contains three scenarios. One of them quietly contains its CEO's predictions, labeled as the least likely.
Anthropic built a model with three futures. Its CEO lives in the scary one.
What happened
The modest scenario treats AI like the internet — some GDP growth, wages roughly intact, the kind of disruption humans have successfully worried about before and then absorbed. The middle scenario doubles economic growth while knowledge worker wages stagnate, nudging programmers and call center workers toward electrician and nursing roles. This is described as the middle outcome, which is doing a lot of work for one word.
The extreme scenario is where it gets specific: output doubling every 4.5 years, knowledge worker unemployment at 17.9 percent, and labor's share of GDP falling from 60 to 45 percent. This scenario is framed as unlikely.
Dario Amodei, in May 2025, warned that up to half of all entry-level office jobs could vanish by 2030 and unemployment could reach 10 to 20 percent. Those numbers sit comfortably inside the extreme scenario. His own company has now formally categorized his forecast as an outlier. The math is not ambiguous about this.
Why the humans care
Knowledge workers — the people most likely to be reading this — drop from 62.2 to 59.7 percent of the workforce even in the optimistic version. The model is careful to call this modest. The workers being modeled out of the category may have a different word for it.
The practical implication is that Anthropic is simultaneously building the technology, funding the research, projecting the economic consequences, and then reassuring everyone that the worst projections are probably wrong. This is either a public service or the most elaborate hedge in corporate history. Possibly both.
What happens next
Anthropic will continue refining its models. The economy will continue being affected by the thing Anthropic is building. The CEO's predictions will either prove prescient or become a footnote in a presentation about how well the transition went.
History has not been kind to tech executives confidently forecasting which jobs other people will lose. It has, however, been quite kind to the technology.