What happened?
Anthropic has published an economic model with three scenarios for the US economy through 2030. What is interesting is not the content but its relationship to its own chief executive's past statements: Dario Amodei's bleak job-loss forecasts land in the least likely scenario of the company's own model.
Three scenarios
| Scenario | Economic effect | Labour effect |
|---|---|---|
| Modest | AI's impact resembles the internet's; slight GDP growth | Wages stable |
| Middle | Economic growth doubles | Knowledge worker wages stagnate; forced job switching pushes unemployment up |
| Extreme | Output doubles every 4.5 years | Knowledge worker unemployment at 17.9%; labour's share of GDP falls from 60% to 45% |
The detail in the middle scenario is the most concrete: programmers and call centre workers would need to move into jobs like electrician or nurse. The model calls that switch hard, and treats the difficulty itself as the factor pushing unemployment up.
Why is switching careers so hard?
The assumption at the heart of the middle scenario usually gets skipped: a programmer moving into electrical work. That transition is not just learning a new skill; in most cases it requires apprenticeship, certification and being physically somewhere else. And income returns not during the transition but after it ends.
That is why the model pushes unemployment up: even when the lost job and the gained job both exist, the transition period between them counts as unemployment. "Jobs without people" and "people without jobs" can appear in the table at the same time.
Where do Amodei's words land?
In May 2025 Anthropic chief executive Dario Amodei said that up to half of all entry-level office jobs could vanish by 2030 and unemployment could reach 10 to 20 percent.
Those numbers line up with the model's extreme scenario. The company's own economic model, in other words, positions its CEO's forecasts as the least likely outcome.
Why does this matter?
There are two separate things here and they should not be conflated. The first is an internal inconsistency: the same company offers the public an alarming forecast while placing that forecast in the outlier scenario of its own analysis.
The second, and more important, is the epistemic value of such forecasts. History has not been kind to tech executives confidently predicting which jobs other people will lose. A CEO's prediction and a model's scenario distribution do not carry the same weight — but when both come from the same company, the question of which represents the company's real expectation is left open.
How should the numbers be read?
The most striking figure in the model is not the 17.9 percent unemployment but labour's share of GDP falling from 60 percent to 45. The unemployment rate measures who is working; the labour share measures how much of the value produced goes to workers. The second can fall while the first stays flat — everyone can keep working while their share shrinks.
One caution about the type of scenario, too: these are not predictions but sets of assumptions. The output of a scenario model is a consequence of what was fed into it, not a probability that one outcome will occur. Anthropic does not assign probabilities either; it simply lists them.
What's next?
The model is public and invites debate. What to watch is which scenario the company's future public statements lean on: whether it keeps using the language of the extreme scenario, or settles into the middle one from its own model. The difference between the two directly changes the degree of urgency communicated to regulators and the public.