https://www.npr.org/2026/09/09/nx-s1-5961443/ai-anthropic-economyIn the extreme scenario Anthropic is now putting in front of the public, nearly 14% of American workers lose their jobs to AI, fewer than half find new ones, and GDP grows at more than seven times its current pace. That last number is the one policymakers should probably be thinking about more than they are.
Anthropic, the company behind the Claude AI assistant, has released an interactive tool that lets users model how AI might affect the U.S. economy over the next several years. Rather than issuing a single forecast, the tool gives users a set of variables to adjust: how capable the technology becomes, how fast it spreads, whether it complements workers or replaces them, and whether displaced workers find new roles. The outputs range from a gentle productivity bump with minimal disruption all the way to an economic shift the authors describe as going “well beyond any event in history.”
The tool is paired with a survey of nearly 11,000 people, which found that public expectations sit somewhere in the middle. Most respondents anticipated meaningful productivity gains alongside real disruption in AI-sensitive fields. That ambivalence tracks with what experts in the field are saying. Anthropic notes that AI researchers tend to expect faster adoption while economists are more skeptical about how quickly the technology actually moves through the broader economy.
Co-founder Jack Clark put it plainly in comments to NPR: the technology will keep improving fast, but getting it into everyday business processes will be harder than the optimists expect. So it becomes very capable, but the economic impact arrives more slowly than the capability curve suggests.
That diffusion gap matters a lot. Anton Korinek, Anthropic’s head of transformative AI economic studies, said it simply: if the AI can do impressive things but nobody actually uses it, there is no economic impact. Adoption rate is the variable that controls almost everything else in the model.
But the upside scenario is worth sitting with. If GDP growth really did hit seven times its current rate, governments would have fiscal resources that are hard to imagine today. Clark’s point is that policymakers should be preparing now to redistribute those gains, because the speed of change in the high-end scenario would leave little time to improvise.
This matters beyond Anthropic specifically. OpenAI, Google DeepMind, and others have published economic research, but an interactive scenario tool aimed at general users is a different kind of move. It shifts the conversation from expert debate to something the public and policymakers can engage with directly. Whether that produces better decisions or just better-informed anxiety is an open question. Still, putting the uncertainty on the table honestly is more useful than false precision.
