Nature probes the science behind AI extinction warnings

When an Anthropic researcher quit on 8 September telling the Wall Street Journal that the systems his employer builds could 'spiral out of control and destroy humanity,' the warning did not stay inside the industry. Jacob Coxon's follow-up on X — 'The people building AI earnestly believe that it could kill us all by the end of the decade' — drew more than 100 million views in 24 hours, and Evan Hubinger, who leads Anthropic's alignment science, amplified it with an estimate that the risk of human extinction was '>10% within the next decade.'

Nature's news explainer, by Elizabeth Gibney, is a sober attempt to ask what, if anything, backs the numbers. The piece sets Anthropic chief executive Dario Amodei's call for 'a slowdown — but not a halt — in AI development,' echoed by Sam Altman and Elon Musk, against the researchers who think the extinction frame is doing more rhetorical than scientific work.

Michael Vermeer of the RAND Corporation is the explainer's sharpest skeptic. The extinction case, he tells Gibney, 'involves so many untestable claims that you just end up with a conversation that is really more like faith than something scientific or empirical.' A 2025 RAND study he cites found complete nuclear extinction not feasible, while leaving biotechnology and atmospheric modification scenarios open — pathways that would still require significant real-world capability.

Heidy Khlaaf of the AI Now Institute goes further, calling the extinction framing 'fear-mongering' and pointing instead to 'AI's low reliability and accuracy rates in critical environments with life-or-death consequences.' Four of the AI researchers we track in our Who's Who directory shared the Nature piece after it ran, a sign that the near-term harms camp is treating this reframing as useful ammunition rather than a sideshow.