Anthropic alignment lead puts AI extinction odds above 10%

An Anthropic researcher who leads the company's alignment science team puts the odds of human extinction from AI at more than 10% within the next decade. That figure, attributed to Evan Hubinger, sits inside a Nature feature by Elizabeth Gibney that asks whether any of the industry's own doomsday talk is actually testable.

The piece turns on a resignation. Jacob Coxon, a researcher at Anthropic, left the company and posted on X: "The people building AI earnestly believe that it could kill us all by the end of the decade." Nature reports the post drew more than 100 million views in 24 hours. In the weeks around it almost 1,400 people working in the AI sector signed an open letter calling for a slowdown, and on September 18 Anthropic CEO Dario Amodei posted his own essay arguing for a slowdown but not a halt in AI development.

The scientific pushback comes in flat sentences. Heidy Khlaaf, chief AI scientist at the AI Now Institute in New York, calls the extinction framing "fear-mongering" and says the greater concern is "AI's low reliability and accuracy rates in critical environments with life-or-death consequences." Michael Vermeer of the RAND Corporation in Santa Monica, who contributed to a 2025 RAND study of extinction risk from nuclear weapons, biotechnology and atmospheric modification, told Nature that making such predictions "involves so many untestable claims that you just end up with a conversation that is really more like faith than something scientific or empirical."

Four of the AI people on our Who's Who radar flagged the piece. Nature attributes the open letter to "cybersecurity incidents" without naming them, and never shows the arithmetic behind Hubinger's number.