Former Anthropic Researcher Resigns, Accuses AI Labs of 'Hubristic Gamble'

Jacob Coxon, a pretraining researcher with experience at the world's leading AI labs, resigned from Anthropic on September 9, 2026. His departure, announced via a viral thread on X under the handle @hilbertspaess, serves as a public indictment of the current trajectory of artificial intelligence development.

Coxon spent the last three years conducting pretraining research at both OpenAI and Anthropic. In his public announcement, he accused both organizations of acting irresponsibly by racing toward the creation of self-improving superintelligence. "Neither company is acting responsibly," Coxon wrote. "They are racing straight to self-improving superintelligence and gambling with our lives."

The Internal-External Divide

The resignation comes at a time of escalating tension regarding AI alignment and the competitive drive toward Artificial General Intelligence (AGI). Coxon’s claims highlight a perceived disconnect between corporate narratives of safety and the internal anxieties of the researchers building the systems. He specifically alleged that many AI executives and senior researchers privately believe the technology could lead to a civilizational catastrophe by the end of the decade, despite presenting a more controlled image to the public.

According to Coxon, the people building these systems "earnestly believe that it could kill us all by the end of the decade," asserting that these fears are not a "marketing stunt." He further claimed that while OpenAI employees may not have deeply internalized these stakes, those at Anthropic understand them but feel trapped in a race to reach superintelligence first.

Industry Implications

This event is significant because Coxon possesses direct, technical insight into the pretraining processes of the two most powerful AI labs in existence. His departure follows a recurring pattern of high-profile exits from leading labs over disputes regarding governance and safety.

If Coxon's assessment is accurate, it suggests that the competitive pressure to achieve superintelligence is overriding the industry's primary safety mechanisms. He described the decision to accept this competitive race and enter what he calls the "endgame" as a "hubristic gamble that should not be launched from a private company’s Slack."

Unresolved Questions

While Coxon's resignation and his accusations are confirmed, several of his more specific claims remain unverified. He suggested that a "Hugging Face attack" served as a warning that has made pacing agreements between U.S. labs more viable, though the causal link between that event and current lab agreements has not been independently confirmed.

Industry observers are now watching to see if other researchers will echo Coxon's warnings or if Anthropic and OpenAI will address the specific allegation that their internal fears regarding existential risk contradict their public safety assurances.