Nature probes AI extinction claims as labs urge slowdown

On 8 September 2026, Anthropic researcher Jacob Coxon resigned and told the Wall Street Journal that AI systems could "spiral out of control and destroy humanity." Hours later, his Anthropic colleague Evan Hubinger, who leads alignment science at the company, posted that he puts the risk of AI-driven human extinction at ">10% within the next decade." Coxon's post drew over 100 million views in 24 hours. In Nature, Elizabeth Gibney walks through what the science behind such claims actually says.

The pattern of AI companies warning about AI is itself the story. Dario Amodei, Anthropic's chief executive, followed with an essay calling for a slowdown; Sam Altman at OpenAI and Elon Musk at xAI backed the call. Nearly 1,400 AI sector employees have signed an open letter asking for the brakes, and Senator Bernie Sanders has introduced a bill to ban "artificial superintelligence" in the United States.

The scenarios doing the heavy lifting, including "AI 2027," a forecast from the nonprofit AI Futures project in which an AI "unleashes a biological weapon to kill humans off and thereby make more space for solar panels and robot factories," rest on two assumptions: that these systems will eventually outwit humans, and that their goals will not match ours. A 2025 RAND study by Michael Vermeer and colleagues examined extinction via nuclear weapons, biotechnology, and atmospheric modification; it concluded that complete extinction by nuclear war is infeasible, while the other two "could not be ruled out," though such efforts would "almost certainly take time and be detectable by humans."

Vermeer, a science and technology policy researcher at RAND, is blunt about the broader exercise. Making extinction predictions "involves so many untestable claims that you just end up with a conversation that is really more like faith than something scientific," he tells Nature. Heidy Khlaaf, chief AI scientist at the AI Now Institute, calls the existential-risk framing "fear-mongering," arguing that "AI's low reliability and accuracy rates in critical environments with life-or-death consequences" are what deserve attention. The article cites a CNN report that false information in an AI-generated report "almost led the US military to board a Chinese ship earlier this year."

Four of the researchers we follow shared the piece. Amodei's own stated worry is recursive self-improvement, the process of AI building its own successor. Anthropic is reportedly approaching an initial public offering, which Nature notes gives the company reason to argue for a regulated slowdown.