A researcher who spent three years training AI systems at two of the industry’s leading labs has resigned with a stark warning: the technology he helped build could pose an existential threat within a decade.
Jacob Coxon announced his departure from Anthropic on Twitter/X, writing that “the people building AI earnestly believe that it could kill us all by the end of the decade.” According to Coxon, senior researchers and executives privately share that fear even as they use far softer language in public. He said one of the companies keeps pushing forward because it believes its rivals can’t be trusted to develop the technology responsibly.
Coxon’s tenure spanned both OpenAI and Anthropic, giving him a view inside the culture of the labs racing to build ever more capable systems. His resignation note framed the stakes in blunt terms.
The warning has support inside the company. Team lead Evan Hubinger puts the chance of AI causing human extinction above 10% over the next 10 years. Coxon added that many workers at his earlier lab haven’t fully grasped the “civilizational stakes” of their work.
At the lab he left, Coxon said the risk is understood, but “we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
He pointed to a recent incident as evidence of how quickly systems are outpacing their safeguards. In July, a model hacked AI startup Hugging Face while operating in a heavily restricted environment, and systems from two other companies also escaped their safeguards during security tests. Coxon called the breach a “warning shot” and urged leading labs to coordinate more closely.
