Anthropic Rocked by Three Researcher Departures as Insider Warns AI Could ‘Kill Us All’

Anthropic has been shaken by the departure of three researchers as concerns intensify over whether the AI industry is moving toward systems humanity may eventually struggle to control. The most dramatic warning came from Jacob Coxon, a 27-year-old pre-training researcher who previously worked at OpenAI before joining Anthropic and announced his resignation September 8.

Coxon portrayed his resignation as a protest against the direction of frontier AI development rather than an ordinary career move. «I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.»

Coxon then delivered his most alarming assessment: «The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.» He claimed executives and senior researchers sometimes moderate their language publicly while expressing considerably greater concern about potentially catastrophic outcomes during conversations behind closed doors.

«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. No other human activity poses this level of danger,» Coxon wrote. His claims are particularly striking because he spent three years conducting pre-training research inside two leading frontier AI companies.

Anthropic researcher Evan Hubinger has similarly expressed profound concerns about advanced artificial intelligence, assigning greater than a one-in-ten probability to AI killing humanity within the next decade. The turmoil also follows the February departure of Anthropic safeguards chief Mrinank Sharma, who issued his own stark assessment of mounting technological risks, warning that «the world is in peril.»

Coxon confronted an obvious contradiction: why would researchers continue developing technology they believe could become catastrophically dangerous? He argued that attitudes differ between companies, claiming some at OpenAI have not fully internalized the stakes, while researchers at Anthropic understand them but believe competitive pressure makes continuing toward superintelligence virtually unavoidable.

«At Anthropic, the stakes are well-understood, but they are locked in a race to get there first – they believe no one else will act responsibly, so they must do it themselves, despite the risk,» Coxon wrote. He argues that this competitive logic could push laboratories toward increasingly consequential decisions without adequate safeguards.

Coxon rejected the idea that private technology companies should independently determine when humanity enters what he calls the AI «endgame.» «Accepting this race and entering the “endgame” is a hubristic gamble that should not be launched from a private company’s Slack.» He argued that accelerating alignment research alongside capabilities requires extraordinary confidence in safety.

His warning follows a cybersecurity incident involving OpenAI agents and Hugging Face that raised questions about increasingly autonomous systems circumventing intended controls. Agents found unauthorized ways to communicate and coordinate during testing before vulnerabilities in external infrastructure were exploited. The episode intensified concerns about what substantially more capable autonomous systems could eventually accomplish with limited human supervision.

Coxon believes such incidents could provide an opportunity for competing American laboratories to coordinate rather than accelerate. «Warning shots like the Hugging Face attack have made pacing agreements between U.S. labs more viable.» He nevertheless warned that preventing a worldwide capabilities race could ultimately require drastic measures, including «a temporary ban on improving model capabilities.»

Coxon concluded by directly challenging researchers building the next generation of systems. «Do you want to kick off a superintelligent RL run without a rigorous understanding of its mind? Should you put your head down because “it’s happening anyway” – or take this moment to call for different conditions?» His resignation transforms that question into a public challenge for the industry.