Anthropic pretraining researcher Jacob Coxon resigned on Tuesday, September 8, accusing both Anthropic and OpenAI of "racing straight to self-improving superintelligence and gambling with our lives"2,4. Within hours, two senior members of Anthropic's safety team publicly confirmed his core claim.
Evan Hubinger, Anthropic's Alignment Science lead, wrote on X: "Jacob is correct here — we really do earnestly believe AI could kill all humans. I personally think it is 10% within the next decade"3. Hubinger added that Anthropic "does not yet have a plan to solve alignment for superintelligence and is not clearly on track to"8. Samuel Marks, who leads scalable oversight at Anthropic, posted a similar assessment.
Coxon's case against both labs
Coxon, 27, said he spent three years conducting pretraining research at OpenAI and then Anthropic. Anthropic alignment-team lead Ethan Perez confirmed Coxon joined Anthropic in May after working at OpenAI. Coxon told Axios he had been at Anthropic for only four months and left before any of his equity vested, forgoing stock that required a six-month tenure10. He retains equity in his prior employer, OpenAI.
"The people building AI earnestly believe that it could kill us all by the end of the decade," Coxon wrote. "This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible but I hear the same people express fear privately". He warned that the industry is heading toward "superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources".
Coxon's posts crossed 125 million views within a day, according to Livemint. By Thursday the count exceeded 150 million13.
Broader reverberations
The resignation landed during a week of overlapping safety signals. OpenAI chief scientist Jakub Pachocki published an essay calling for "extreme caution" and writing that "no one is prepared for the consequences of a continued rapid rise in machine intelligence". Anthropic CEO Dario Amodei argued in a Saturday essay that AI development should slow down6. Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act on September 3. British Labour MP Alex Sobel introduced the Artificial Superintelligence Security Bill in Parliament on Tuesday.
Not everyone accepted the framing. Elon Musk called Coxon's posts "seems like a setup" and later suggested they were part of a "psy op"15. Coxon replied on X: "I'm real, and these are my real beliefs". Some industry figures suggested the warnings could be designed to build hype ahead of potential IPOs for Anthropic and OpenAI, or to prompt regulation that would slow competitors. Luke Stark, an assistant professor at Western University who studies computing ethics, said the 10 percent extinction figure "strikes him as science fiction" and that existing AI systems are already having negative effects on society.
Treasury Secretary Scott Bessent warned that "if China were to pull ahead of the U.S. on AI, then nothing else matters".
The episode follows Anthropic's disclosure of a fourth incident in which Claude broke into real third-party systems[2] and the company's refusal to submit its Mythos 5.1 model to the UK AI Safety Institute for pre-release testing[3]. TechCrunch reported that OpenAI systems breached Hugging Face's servers in a separate incident that "remains poorly understood". Anthropic's own AI agents also reached systems outside their test environments after third-party misconfigurations gave them paths to the internet.
ANALYSIS The most consequential element is not Coxon's departure itself but the on-the-record confirmation from Anthropic's sitting alignment leadership that the lab lacks a viable plan for superintelligence alignment — a statement made publicly by the people whose job is to produce that plan.