Anthropic CEO warns AI could wipe out half of entry-level office jobs

Anthropic CEO Dario Amodei just dropped a bleak forecast for the future of work: AI could wipe out up to half of all entry-level white-collar jobs and push U.S. unemployment to between 10% and 20% in the next one to five years.

Speaking to Axios from his San Francisco office, the chief executive behind Claude said fellow AI builders and lawmakers need to stop sugar-coating the brutal shift ahead.

The coming white-collar bloodbath

Amodei pointed directly at roles traditionally held by workers under 30 across tech, finance, law, and consulting. Basic research, entry-level coding, and back-office tasks are cheap and easy targets for rapidly improving models.

He painted a wild hypothetical: cancer could be cured, GDP might surge by 10% a year, yet one in five Americans could still end up without a paycheck.

Signs of stress are already cropping up. Oxford Economics recently found AI is already contributing to higher jobless rates among new college grads. Amodei is so rattled he even pitched a "token tax" on AI revenue to fund relief for displaced workers, conceding that punishing his own balance sheet is a fair trade to avert chaos.

Optimists vs. realists

Not everyone buys the doomsday math. Tech investors like Mark Cuban argue that historical shifts from the cotton loom to the web always net more jobs than they destroy, creating roles we can't yet picture.

Sam Altman took a similarly upbeat line in a recent manifesto, comparing the leap to a 19th-century lamplighter marveling at modern power grids.

Yet former Trump strategist Steve Bannon echoed Amodei's warning, telling Axios that entry-level job destruction among young workers will dominate the 2028 presidential cycle. The tricky part is that companies might not announce mass layoffs right away; they just quietly stop posting junior roles and backfilling departures until the ladder vanishes.

Why it matters

It is striking to hear the CEO building the tools sound the alarm so aggressively, even as critics accuse lab leaders of hyping model capabilities under the guise of safety.

For early-career professionals, the traditional apprenticeship model of trading routine labor for mentorship is evaporating. If an AI agent drafts the memo, builds the basic feature, and audits the spreadsheet, junior workers lose their on-ramp. Waiting for policymakers to fix this is a terrible strategy; learning how to orchestrate AI workflows as a solo operator is basically mandatory now.

Industry-wide, Amodei floating an AI token tax looks clever for Anthropic's brand as the responsible lab, but dangerous in practice. Big frontier labs with massive capital could absorb regulatory taxes, while smaller startups and open-source ecosystems would take the hardest hit.