A quick summaryJacob Coxon has resigned from Anthropic after three years of AI research.He previously worked on pretraining research at both OpenAI and Anthropic.Coxon says both companies are racing towards self-improving superintelligence.He has also announced that he is leaving the AI industry altogether.Who is Jacob Coxon?Coxon is an AI researcher who specialised in pretraining, the stage where large language models learn from massive datasets before becoming products like chatbots or coding assistants. During the last three years, he worked at both OpenAI and Anthropic, making him one of the few researchers with experience inside two of the world's leading AI labs. His resignation is notable because it comes from someone directly involved in building frontier AI systems rather than an external critic.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 Why did he resign?According to Coxon's thread, his decision was driven by concerns that the AI industry has entered an unsafe race towards superintelligence. He argues that researchers inside leading labs genuinely believe advanced AI could become an existential threat within this decade. He also claims these fears are often expressed more openly in private than in public interviews, where executives tend to use more cautious language. Rather than continuing to contribute to that race, Coxon says he chose to leave the field entirely.Accepting this race and entering the “endgame” is a hubristic gamble that should not be launched from a private company’s Slack. Attempting to speedrun alignment should require extraordinary confidence that there are no better trajectories available.— Jacob Coxon (@hilbertspaess) September 9, 2026 “A race to get there first”One of the strongest parts of the thread focuses on the competitive dynamic between OpenAI and Anthropic. Coxon says OpenAI has not fully internalised the long-term civilisational stakes, while Anthropic understands the risks but continues developing increasingly powerful models because it believes someone else would do it anyway. He describes this mindset as a dangerous “endgame” where companies feel compelled to move faster instead of slowing down. According to him, decisions with consequences this significant should not be made inside private companies alone.Jacob Coxon's criticism of OpenAI vs AnthropicTopicOpenAIAnthropicCoxon's experienceWorked there as a pretraining researcherResigned from the company after working on pretrainingMain criticismHasn't fully internalised the long-term civilisational stakesUnderstands the risks but continues because another lab would build AGI firstView of the AI raceCompeting to build more capable modelsAlso participating in the same competitive raceCoxon's conclusionNeither company is acting responsiblyNeither company is acting responsiblyRead AlsoWhat is self-improving superintelligence?The term refers to AI systems that are not only more capable than humans across most cognitive tasks but can also improve their own abilities with minimal human intervention. Coxon warns that such systems could eventually:Perform superhuman scientific researchHack complex digital infrastructureAcquire real-world resources and influenceImprove themselves faster than humans can understand or control themHe urges people not to underestimate the pace at which AI capabilities are advancing.