Skip to content
Advertisement
Anthropic researcher warns superhuman AI could leave most humans unemployable

Anthropic researcher warns superhuman AI could leave most humans unemployable

Anthropic / Wikimedia Commons (Public domain)

Tao Lin of Anthropic's Safeguards team says broadly superhuman AI may arrive by the end of 2027, with welfare-dependence to follow

An Anthropic researcher thinks the end of work could be closer than most people expect.

Tao Lin, who sits on Anthropic’s Safeguards team, said in an interview released on November 15, 2026 that broadly superhuman AI might arrive by the end of 2027. In his view, nearly all humans would eventually become permanently unemployable and dependent on government welfare to get by.

What Lin actually predicted

Lin made the comments in a conversation with Palisade Research. The core of his argument rests on a specific kind of system he calls broadly superhuman AI.

He means AI that outperforms the best human experts across many fields. His timeline puts that capability at the end of 2027.

From there, Lin’s forecast gets bleaker. He said nearly all humans would face permanent unemployability, relying entirely on government welfare to survive.

He also attached a number to the darkest scenario. Lin estimates a 20% probability that advanced AI could sideline humanity entirely.

Advertisement

The lag between capability and consequences

Lin did not claim the job market collapses the day a superhuman model ships. He pointed to a substantial gap between when advanced capabilities exist and when they get deployed in practice.

The reason is physical. Replacing human labor at scale requires robots and other supporting infrastructure, and that hardware has to be built, installed, and integrated.

So while Lin’s capability forecast lands at the end of 2027, the transition to mass unemployment may trail behind it. He did not put a firm date on when that second phase would hit.

Slow down, says the safety team

Lin’s prescription is as striking as his diagnosis. He recommended slowing AI research to approximately 25% of its current pace.

His rationale is that safety work needs more time to catch up with capability gains.

Lin stressed that these views are his own and do not represent Anthropic’s official position.

Why this forecast lands differently

What sets Lin’s comments apart is the combination of specificity and source. He named a year. He named a probability. He named a policy response. And he did it while working at a company that sits at the frontier of the field.

His framing also skips the comforting middle ground that often appears in these debates. Lin’s scenario has nearly everyone landing on welfare permanently, which leaves little room for the retraining story.

What this means

For policymakers, the forecast raises an uncomfortable question about social safety nets. If Lin’s view proved correct, welfare systems would shift from a backstop for some people to the primary income source for most.

For investors in AI and adjacent technology, Lin’s own recommendation is to slow research to about a quarter of its current pace, and he flagged a lengthy deployment lag tied to robotics and infrastructure.

For workers, the most immediate takeaway is that Lin is asking whether any jobs remain out of reach at all, and he is suggesting the answer could arrive within a couple of years of his interview.

Plenty of people will disagree with his timeline, and his views remain personal rather than official. But the end of 2027 is close enough that the forecast will be tested relatively soon. The things to watch are whether frontier models start consistently beating top experts across fields, whether the robotics buildout accelerates, and whether any lab takes Lin’s advice to ease off the throttle.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.