‘Unemployed or Very-Low-Wage Underclass’ Looms

The CEO of Anthropic—the AI company valued at about $350 billion—warns that AI could create a permanent underclass of workers.

Dario Amodei wrote in a 20,000-word essay out Monday that the AI systems his company is helping to build could leave less-skilled workers with nowhere to go—”an unemployed or very-low-wage ‘underclass.'”

AI’s takeover of jobs will advance, he wrote, “from the bottom of the ability ladder to the top,” making scores of jobs obsolete. The changes will thus leave these workers nowhere else to go, he argued.

A ‘General Labor Substitute for Humans’

“If AI does to white-collar work what globalization did to blue-collar, we need to confront that directly,” BlackRock CEO Larry Fink warned as he opened the World Economic Forum at Davos last week.

Fink told leaders at Davos to stop with the pablum about “the jobs of tomorrow” and start making concrete plans for sharing AI’s gains. Amodei’s essay warns that such efforts won’t be enough.

When mechanized farming displaced agricultural workers over generations, those workers moved into factories. When globalization gutted manufacturing starting in the 1990s, some workers moved into service and knowledge work. But Amodei argues that AI increasingly matches the full range of human cognitive abilities, so it will take away jobs drafting memos, reviewing contracts, and analyzing data that might otherwise emerge. A customer service rep who retrains as a paralegal would find AI waiting there, too.

“AI isn’t a substitute for specific human jobs but rather a general labor substitute for humans,” he wrote.

Early Warning Signs

Amodei’s timeline—losses of up to half of entry-level white-collar jobs within five years—sounds dire even as CEOs compete to make the most apocalyptic predictions about the future of work.  “Probably none of us will have a job,” Elon Musk told a Paris tech conference in 2024, describing a future where work becomes “optional.” Asked in August 2025 how workers would survive, OpenAI’s CEO, Sam Altman, said, “I don’t know [and] neither does anyone else,” before gesturing at the need “to have some new economic model.”

But those looking at the data aren’t panicking yet.

Employment among workers aged 20 to 24 in the most AI-exposed occupations has declined since ChatGPT’s release in November 2022, according to an analysis released earlier this month by the Federal Reserve Bank of Dallas. The researchers noted that no technology has ever massively disrupted the workforce within just a few years of its release. So far, they found, AI’s overall impact has been “small and subtle,” adding only about 0.1 percentage point to the unemployment rate.

An ongoing study by the Yale Budget Lab on AI’s effects on jobs has reached similar conclusions. The share of workers in the most AI-exposed occupations has remained stable since ChatGPT launched, and jobs are shifting only slightly faster than with the rise of the internet in the late 1990s.

“It would be unprecedented if a new technology had massively disrupted the workforce in three years,” Martha Gimbel, executive director of the Yale Budget Lab, told Investopedia in December. “These kinds of things take time.”

Lessons From Past Technological Disruptions

That doesn’t mean workers should be complacent, Gimbel said. The Yale Budget Lab’s research shows that when technological disruptions arrive, the workers they displace rarely recover.

“Workers who get displaced do not get the benefits of the technology in the same way,” she said. “We do not have a good track record of figuring out how to help those workers.”

Telephone operators displaced by automation were more likely to be underemployed—or to leave the workforce entirely. And while data from the Industrial Revolution is sparse, Gimbel noted “that it was devastating for the weavers, and they never really recovered.”

Amodei argues that AI’s disruptions will unfold far faster than past tech shifts, when telephone operators, weavers and factory workers across the Rust Belt never recovered. Neither might today’s paralegals and customer service reps, and the window to prepare is closing.

“It cannot possibly be more than a few years before AI is better than humans at essentially everything,” Amodei wrote. “I can feel the pace of progress, and the clock ticking down.”