Anthropic alignment lead puts AI extinction odds above 10%

Nature has run a science-desk explainer weighing the actual evidence behind claims that AI could wipe out humanity. The framing is unusually direct for a general-science outlet: Elizabeth Gibney's piece features the recent resignation of Anthropic researcher Jacob Coxon, who left the company on September 8 saying that "the people building AI earnestly believe that it could kill us all by the end of the decade."

Inside Anthropic, alignment science lead Evan Hubinger puts extinction risk above 10% within the next decade. CEO Dario Amodei has called for a slowdown rather than a halt, and OpenAI's Sam Altman and xAI's Elon Musk have backed the suggestion. Nature also gives space to critics who see the framing as unfalsifiable. Michael Vermeer, a science and technology policy researcher at RAND, tells the magazine the extinction question "involves so many untestable claims that you just end up with a conversation that is really more like faith than something scientific."

The article's spine is a 2025 RAND study that worked through specific pathways: nuclear weapons, biotechnology, and atmospheric modification. Complete nuclear extinction was judged not feasible; the biotech and atmospheric routes could not be ruled out, but any such effort "would almost certainly take time and be detectable by humans, who might then stop their eradication." Heidy Khlaaf, chief AI scientist at the AI Now Institute, calls the existential focus "fear-mongering" and argues that AI's low reliability in life-or-death environments is the more pressing hazard.

Gibney sets the piece against a hardening political backdrop: more than 1,400 AI-sector employees signed an open letter after recent cybersecurity incidents, US legislators have proposed a bill banning "artificial superintelligence," and in September 2026 a false AI-generated report nearly caused the US military to board a Chinese ship.

Within a day of publication, two of the researchers in our Who's Who directory had the piece in circulation.