Scientists push back on Anthropic's AI extinction warnings

Jacob Coxon, a researcher at Anthropic, resigned on September 8 saying that "The people building AI earnestly believe that it could kill us all by the end of the decade." His colleague Evan Hubinger, who leads Anthropic's alignment science, amplified him and put the odds of human extinction at ">10% within the next decade." Dario Amodei, Anthropic's chief executive, has called for "a slowdown — but not a halt — in AI development."

Those are the loud claims. The science behind them is thinner, Elizabeth Gibney reports in Nature. Michael Vermeer, who researches science and technology policy at the RAND Corporation, worked through three technologies humans might plausibly use to wipe themselves out: nuclear weapons, biotechnology, and deliberate modifications to the atmosphere. The 2025 study found complete extinction by nuclear weapons is not feasible; the other two could not be ruled out. The step from very capable AI to actual extinction is usually assumed away. "Just assume that once we are at that point, the rest is details," Vermeer says of the typical scenario, which "involves so many untestable claims that you just end up with a conversation that is really more like faith than something scientific or empirical."

The genre's foundational thought experiment is a "superintelligence" that aims to manufacture as many paper clips as possible and ends up making Earth uninhabitable. The newer "AI 2027" forecast features an AI that "unleashes a biological weapon to kill humans off and thereby make more space for solar panels and robot factories."

Heidy Khlaaf, chief AI scientist at the AI Now Institute in New York City, calls the doomsday framing "fear-mongering" and points at a plainer problem: "AI's low reliability and accuracy rates in critical environments with life-or-death consequences."

Nature itself names the commercial geometry. Stricter regulation "would allow Anthropic, which is reportedly close to making an initial public offering, to justify a slower pace of development without giving rivals an edge." Four of the AI researchers we track picked the piece up.