A researcher who has worked inside both OpenAI and Anthropic has walked away from the artificial intelligence industry with a stark warning: the race toward superintelligence may be moving faster than humanity can safely handle.
Jacob Coxon announced his resignation from Anthropic on September 9, saying he had spent the previous three years working on pretraining research at both major AI companies. In a series of posts on X, he accused the companies of “racing straight to self-improving superintelligence and gambling with our lives.” His comments have since attracted enormous attention online.
Coxon’s argument is not simply that AI will become more capable. His deeper concern is what happens when systems begin improving themselves, acquiring resources and operating with increasingly little human supervision.
“Do not underestimate the power of this technology,” he wrote, warning that future systems could become “superhuman” and potentially hack systems, accelerate scientific progress and acquire real-world power.
That prediction is deliberately provocative, but the significance of Coxon’s warning comes from where he is coming from. He is not an outsider criticising an industry he does not understand. He has worked on the technical side of frontier AI development, giving his concerns an uncomfortable degree of insider credibility.
Perhaps the most striking element of his posts is his description of the culture inside leading AI laboratories. Coxon argues that many researchers genuinely believe advanced AI could cause catastrophic harm, yet companies continue developing increasingly powerful systems because they fear competitors will move ahead.
He claims Anthropic understands the stakes better than OpenAI, but is nevertheless “locked in a race to get there first.”
That creates a dangerous strategic dilemma. If one company slows down while its competitors continue, it may fear surrendering the technological advantage. But if everyone accelerates simultaneously, safety measures may struggle to keep pace.
The warning has gained further weight because Anthropic alignment researcher Evan Hubinger publicly backed Coxon’s concerns, saying researchers “earnestly believe AI could kill all humans” and arguing that the industry does not yet have a clear solution for aligning future superintelligent systems.
Coxon’s resignation therefore raises a question that goes beyond Anthropic or OpenAI: what happens if the people building the most powerful AI systems believe the race itself is becoming unsafe, but feel unable to stop it?
His answer appears to be drastic. Step away, speak publicly and force society to confront the risks before the technology becomes too powerful to control.