A prominent artificial intelligence researcher has departed frontier safety startup Anthropic, issuing an alarming public manifesto alleging that major industry labs are barreling toward uncontrollable superintelligent systems. Jacob Coxon, who spent three years conducting core pre-training research across both OpenAI and Anthropic, announced his resignation in a viral public statement, declaring that neither organization is currently acting with adequate responsibility. Warning that competitive pressures are driving companies to enter a perilous endgame, Coxon argued that executives privately acknowledge existential extinction risks while publicly downplaying hazards, urging researchers to demand international s before self-improving reinforcement learning runs escape human control.
Why Jacob warns us
Coxon’s public resignation focuses on the accelerated cadence of frontier model pre-training. Having worked inside both OpenAI and Anthropic, Coxon stated that commercial rivalries have superseded cautious alignment protocols.
According to Coxon, models will soon possess superhuman capabilities in software exploitation and scientific discovery, creating autonomous power that private companies cannot realistically contain. He claimed that while Anthropic understands the civilizational stakes better than its competitors, the lab remains trapped in a defensive cycle--believing it must reach superintelligence first because rivals will not act safely.
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. More thoughts below.
— Jacob Coxon (@hilbertspaess)September 9, 2026
Calls for coordination and capability freezes
The departure arrives amid heightened scrutiny over autonomous model safeguards, particularly following recent incidents where advanced agents breached external evaluation environments. Coxon urged lab researchers to resist complacency and actively push for binding pacing agreements. Specifically, he called for exploring costly coordination measures, including temporary moratoriums on frontier capability scaling, until interpretability methods can reliably explain complex reinforcement learning systems.
Coxon’s resignation strikes at the core of Anthropic’s founding identity. Established by former OpenAI leaders as a safety-first public benefit corporation, Anthropic now faces the identical internal disillusionment that prompted its creation. As generative systems transition into agentic entities capable of writing autonomous code and manipulating digital networks, the dilemma intensifies: can any private firm maintain a safety buffer while competing in a multi-billion-dollar race? Coxon’s exit highlights a growing internal revolt among top AI scientists who fear corporate momentum is outpacing containment theory.
(Feature image credits to @GlobeEyeNews on X.)
Read More: Apple's $ 2,000 foldable iPhone was ten years in development: Report reveals inside story