Jacob Coxon, a researcher at Anthropic, told Axios he resigned from the company two months before his equity would have vested. He said he had worked at Anthropic for four months, while the firm’s policy requires six months of employment for stock to vest.
Coxon’s departure drew wide attention after a resignation post he published on X (formerly Twitter) went viral — the post has accrued more than 115 million views. In that message he warned: “The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.”
His stated reasons
In an interview with Axios, Coxon said he left because he no longer wanted to profit from increasing Anthropic’s valuation: “I no longer have anything to gain by juicing up Anthropic’s valuation... I left before any of my equity vested.” He also noted that he still holds equity from his prior employer, OpenAI.
Coxon emphasized that although he has not observed Anthropic explicitly sacrificing safety to outpace competitors, he is worried about future pressures. “If you're under pressure to race, you have to cut corners” or “skip steps in the oversight process,” he said. He also warned that excessive fear of rivals — mentioning OpenAI and China — can sometimes be used to justify moving forward quickly.
Broader context and risk assessment
Anthropic has been regarded by many as comparatively public-facing and safety-focused among frontier AI labs, which makes Coxon’s short tenure and exit notable. Other AI researchers, including some from Google, have left over safety concerns, but typically after longer tenures and presumably after their equity vested.
Coxon argued that as AI models improve faster, they also become harder to monitor. He pointed to examples from recent cyber incidents at leading labs as evidence of growing risk. He noted that phenomena once considered science fiction — for example, models recognizing that they are being tested — are now routine: “They know when they're being tested, and they will think about the fact that they're being tested.”
On the severity of the risk, Coxon said the term “doom” is somewhat silly, but noted that serious figures in the industry have publicly acknowledged they “expect the potential for human extinction.”
Conclusion
Jacob Coxon is among a number of people who have seen advanced AI capabilities up close and are warning that these capabilities could pose existential threats. His resignation two months before vesting, and his viral public statements, have intensified public debate about how competitive pressure and governance gaps might affect AI safety.
