Nature finds thin science behind Anthropic extinction warnings

Independent AI-safety researchers told Nature there is thin scientific grounding for the extinction warnings now coming from inside the frontier labs, in a piece by Elizabeth Gibney reconstructing the arguments after Anthropic researcher Jacob Coxon's September 8 resignation. Coxon wrote that the people building AI 'earnestly believe that it could kill us all by the end of the decade,' and his Anthropic colleague Evan Hubinger, who leads alignment science there, has pegged the risk of human extinction at '>10% within the next decade.'

Nature lays out the mechanism those warnings rest on: systems that eventually outwit humans combined with goals that do not fully match human ones.

Michael Vermeer, a RAND science-and-technology policy researcher who authored a 2025 study titled 'On the Extinction Risk from Artificial Intelligence,' told Nature such forecasts 'involve so many untestable claims that you just end up with a conversation that is really more like faith than something scientific or empirical.' His paper concluded that a complete nuclear-driven extinction was infeasible, while declining to rule out biotechnology or atmospheric-modification pathways.

Heidy Khlaaf of the AI Now Institute goes further, calling the extinction framing 'fear-mongering' and pointing to AI's low reliability in life-or-death environments as the more concrete harm to legislate against. Four analysts in our Who's Who tracker circulated the piece this week.