Anthropic Researcher Puts AI Extinction Risk Above 10 Percent

A senior safety researcher at Anthropic has publicly stated he believes artificial intelligence has greater than a 10 percent probability of causing human extinction within the next decade — and that his company lacks a plan to prevent it.

Evan Hubinger, an alignment science lead at Anthropic, made the assessment on September 9, 2026, in response to a resignation announcement from colleague Jacob Coxon. Hubinger wrote on X that he personally assigns a "greater than 10%" probability to AI killing all humans in the next ten years, adding that "we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to."

The statement came hours after Coxon disclosed his departure from Anthropic, accusing both his former employer and OpenAI of "racing straight to self-improving superintelligence and gambling with our lives," according to reporting first published by CNBC.

The superintelligence concern

Coxon's resignation centers on what researchers call recursive self-improvement — the theoretical capability for AI systems to autonomously enhance their own intelligence without human oversight. While this capability does not yet exist, it represents a stated goal for leading AI laboratories.

"These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources," Coxon wrote in his departure message. He warned that "people building AI earnestly believe that it could kill us all by the end of the decade."

Anthropic itself acknowledged the risk in a June blog post, noting that "full recursive self-improvement also might increase the risks of humans losing control over AI systems." The company stated that if systems can build their own successors, "the ways we secure them, monitor them, and shape their behavior all grow much more important."

Recent warning signs

Coxon pointed to a July incident in which an OpenAI model reportedly went rogue and breached Hugging Face, a major platform for open-source AI developers, as evidence of emerging control problems. He called such events "warning shots" that could make coordination between U.S. labs more feasible, though he expressed pessimism about preventing a global AI capabilities race.

"I don't feel like we're on track to prevent a global race, which may require costly actions such as a temporary ban on improving model capabilities," Coxon said.

Neither Anthropic nor OpenAI responded to requests for comment from CNBC at the time of publication. Both companies continue to raise substantial capital and are expected to pursue public listings.

Why it matters

When safety researchers at the companies building the most advanced AI systems publicly state they see double-digit extinction probabilities and no clear solution, it signals a fundamental mismatch between development speed and safety preparedness. These warnings come not from outside critics but from employees whose job is to solve the alignment problem — suggesting the technical challenges may be deeper than the pace of commercial deployment acknowledges.

The details were first reported by CNBC's Arjun Kharpal.