Nature audits AI extinction warnings, finds thin science

Independent AI-safety researchers say the empirical floor under warnings that artificial intelligence could kill all humans within a decade is thin, according to a feature by Elizabeth Gibney in Nature.

The latest round was ignited on September 8, when Anthropic researcher Jacob Coxon resigned with a post declaring that 'the people building AI earnestly believe that it could kill us all by the end of the decade.' It drew more than 90 million views in a day, and four researchers in our Who's Who directory passed the Nature piece around once it ran.

Evan Hubinger, who leads Anthropic's alignment-science work, puts the chance of AI killing all humans within the decade at greater than 10 percent. Researchers have not spelled out how AI would do it, Nature notes; a speculative forecast called AI 2027 imagines 'an AI unleashes a biological weapon to kill humans off and thereby make more space for solar panels and robot factories.'

Michael Vermeer, a science-and-technology policy researcher at RAND, authored a 2025 study on the extinction risk from AI that found a complete nuclear-driven extinction infeasible while declining to rule out biotechnology or atmospheric-modification pathways. He told Nature the predictions 'involve so many untestable claims that you just end up with a conversation that is really more like faith than something scientific or empirical.'

Heidy Khlaaf, chief AI scientist at the AI Now Institute in New York, is blunter: she calls the extinction framing 'fear-mongering' and argues that AI's low reliability in life-or-death environments is the more concrete harm to legislate against.