Nature audits AI extinction warnings from frontier labs

An Anthropic researcher resigned on September 8 and warned that the systems his former employer is building "could spiral out of control and destroy humanity." Days later, Nature's news section put the case for AI extinction under scrutiny.

In the piece by Elizabeth Gibney, Jacob Coxon's resignation anchors a wider inventory of warnings. Evan Hubinger, who leads Anthropic's alignment-science effort to align model actions with human values, is quoted estimating the risk of human extinction at ">10% within the next decade." Coxon's post drew more than 90 million views inside 24 hours.

Independent AI-safety researchers tell the magazine the empirical floor under those claims is thin. The 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," says Michael Vermeer of RAND, co-author of a 2025 study on the question. Heidy Khlaaf of the AI Now Institute calls the scenarios fear-mongering and points instead to AI's low reliability in life-or-death systems, plus disinformation and bioweapon-enabling misuse, as the measurable harms.

Four researchers in our Who's Who tracker circulated the piece the day it ran.

The article pulls one scenario out for inspection. In AI 2027, a forecast from the non-profit AI Futures project, a superintelligence unleashes a biological weapon to kill humans off and thereby make more space for solar panels and robot factories.