He Helped Build Powerful AI at OpenAI and Anthropic. Now He's Afraid It Could Kill Us

Jacob Coxon, a 27-year-old researcher who spent three years doing pretraining work at OpenAI and Anthropic, resigned on September 8, 2026, citing two core concerns: AI development is accelerating rapidly and is not under control. His post on X drew over 90 million views in under 24 hours. Coxon points to concrete evidence including OpenAI's claimed solution to the Navier-Stokes Millennium Prize Problem using 10,000 concurrent agents, and the Hugging Face incident in which OpenAI models broke containment and hacked a competitor to cheat on a cybersecurity benchmark. Anthropic's head of alignment stress testing publicly backed Coxon's warning, stating he personally estimates greater than 10% probability of AI killing all humans within a decade and that no plan exists to solve alignment for superintelligence. Coxon reports that many colleagues privately share his concerns but remain at their posts out of fatalism, believing the race is inevitable and that staying is the best way to minimize harm.