RAND analyst calls AI extinction estimates untestable

Anthropic alignment science lead Evan Hubinger publicly puts the risk of human extinction from AI at '>10% within the next decade.' Elizabeth Gibney's Nature explainer surveys the researchers who trust numbers like that and the ones who do not.

The trigger was a viral X post by Anthropic's Jacob Coxon: 'The people building AI earnestly believe that it could kill us all by the end of the decade.' The post drew more than 100 million views in 24 hours. Hubinger reposted it with the >10% figure. Chief executive Dario Amodei followed with an essay calling for a slowdown, not a halt, in AI development. Nearly 1,400 AI-sector employees signed an open letter making a similar request.

Michael Vermeer, a science-and-technology policy researcher at RAND, told Nature that predicting doom 'involves so many untestable claims that you just end up with a conversation that is really more like faith than something scientific.' Vermeer led a 2025 RAND report, On the Extinction Risk from Artificial Intelligence, that walked through nuclear, biotechnology and atmospheric-modification scenarios; a nuclear-driven extinction, it argued, is infeasible, while the biological and atmospheric routes remain open.

Heidy Khlaaf, chief AI scientist at the AI Now Institute, is blunter. She calls the extinction rhetoric 'fear-mongering' and points instead to 'AI's low reliability and accuracy rates in critical environments' as the concern that deserves attention. Four researchers on our Who's Who list shared the piece the day it ran.