Jacob Coxon spent years helping build some of the most capable AI systems on the planet. Now he says the industry may not fully understand what it is building, or what happens if it succeeds.
The 27-year-old researcher, who worked at both OpenAI and Anthropic, resigned publicly in early September 2026. He then laid out his concerns in a series of posts on X that drew tens of millions of views.
What Coxon actually said
Coxon announced his departure from Anthropic around September 8-9, 2026, via his X account, @hilbertspaess. His core argument: the top AI labs, Anthropic and OpenAI included, are locked in a competitive race toward self-improving superintelligence.
He framed that competition as an arms race. In his telling, it could produce superhuman AI as soon as 2027.
The more unsettling part of his message concerns the people doing the building. Coxon claimed that those working on these systems sincerely believe the technology could lead to humanity’s destruction before the decade is out. He argued the labs are pushing ahead anyway while giving short shrift to potentially catastrophic outcomes.
Coxon’s background gives the warning some weight. He spent approximately three years in pretraining roles at OpenAI, where he contributed to the GPT-4o model.
He is also a University of Cambridge graduate and a medalist at the International Mathematical Olympiad.
Following his resignation, Coxon holds no ongoing financial stake in Anthropic.
Backup from inside Anthropic
Evan Hubinger, Anthropic’s Alignment Science Lead, also stressed how serious the situation is. Hubinger put the chance of AI causing human extinction within the next ten years at greater than 10%.
From X posts to a hearing room
The fallout has not stayed online. Coxon’s resignation and the claims that followed have sparked broad debate over AI governance, including calls for international cooperation on AI oversight and revived discussion about a possible pause on the rapid development of advanced AI systems.
Coxon is scheduled to testify at a New York City Council hearing on AI safety on October 5, 2026. Leaders from Anthropic, OpenAI, Google and Meta are also slated to appear at the hearing.
Leave a comment