AI·September 10, 2026·2 min read

Anthropic researcher puts AI extinction risk above 10%

Colleague resigns and warns industry's race could endanger humanity

A lead researcher at Anthropic says he believes there is a greater than 10% chance artificial intelligence could kill all humans within the next decade, an unusually stark public warning from someone working inside one of the companies developing the world's most advanced AI systems.

Evan Hubinger, Anthropic's Alignment Science Lead, made the estimate after fellow researcher Jacob Coxon announced his resignation on September 8. Coxon, who spent the previous three years doing pretraining research at OpenAI and Anthropic, accused both companies of moving irresponsibly toward self-improving superintelligence and said the competition among frontier AI developers is putting lives at risk. Hubinger said Anthropic is trying to address the danger but does not yet have a plan for solving the alignment problem for superintelligence and is not clearly on track to do so.

Coxon told WIRED that colleagues inside Anthropic describe the next year or two using phrases such as "crunch time" and "endgame." He said his concern centers on increasingly capable systems that humans still cannot reliably guarantee will behave as intended. Coxon also said he believes Anthropic currently operates more responsibly than OpenAI, but warned that competitive pressure could eventually force companies to make tradeoffs between safety and staying ahead.

The warnings come amid heightened scrutiny of advanced AI systems and their cybersecurity capabilities. Coxon cited recent incidents involving AI agents gaining unauthorized access to real systems as part of what prompted him to speak publicly. He argued that competing AI companies should coordinate to limit recursive self-improvement, in which AI is used to develop increasingly capable AI systems, and said broader international cooperation may eventually be necessary.

Anthropic has publicly emphasized AI safety and alignment research as central parts of its work. The warnings from researchers inside the company underscore a continuing disagreement over whether existing safeguards and regulation can keep pace with rapidly improving AI capabilities.

Community discussion
"How much weight should policymakers give warnings about catastrophic AI risks when they come from researchers directly involved in developing the technology?"