Former Anthropic researcher Jacob Coxon resigned from the AI company after publicly warning that, in his view, leading artificial intelligence labs are moving too quickly toward increasingly powerful systems without having established adequate ways to control the risks.
Coxon had spent about three years working on pretraining research across OpenAI and Anthropic. In his resignation statement, he argued that neither company was acting responsibly and accused the labs of racing toward self-improving artificial intelligence while taking unacceptable risks.
His claims are warnings about possible future AI systems, not established predictions about what current AI models will do. Coxon has also distinguished the capabilities of today's systems from the more advanced systems he fears could emerge as AI development continues.
Why Coxon left Anthropic
In his resignation statement, Coxon said researchers inside leading AI laboratories genuinely worry that sufficiently advanced systems could become dangerous. He called for greater coordination between competing laboratories and stronger safeguards around the development of more capable AI.
In a September 9 interview with WIRED, Coxon described the next year or two as a critical period for AI safety and discussed concerns about alignment, the technical challenge of ensuring increasingly capable systems continue to behave according to human intentions.
He told WIRED that his concern was not simply that today's AI systems are already uncontrollable. Rather, his warning focused on what could happen if increasingly capable systems become able to improve AI research itself faster than existing safety techniques can keep pace.
Researchers inside Anthropic responded
Coxon's concerns were not expressed in isolation. Anthropic researcher Evan Hubinger publicly responded to his resignation and said he personally assigns a greater than 10 percent probability to AI causing human extinction within the next decade.
At the same time, Hubinger said he believed Anthropic was trying to address the problem and emphasized that the risk he was discussing concerned future superintelligent systems rather than present-day models. His response was reported alongside Coxon's resignation by ABC News.
Coxon's concern is about the AI race
Coxon's criticism focuses partly on competitive pressure between AI developers. His argument is that even organizations that take safety seriously may feel pressure to continue developing more capable systems because slowing down could allow a competing laboratory to move ahead.
He told Axios that he left Anthropic before his equity had vested. The disclosure provided additional context for his decision to leave the company while publicly criticizing the direction of advanced AI development.
Coxon's warnings remain his assessment of a highly uncertain technological risk. They do not establish that artificial intelligence will become uncontrollable or cause catastrophic harm. Researchers continue to disagree about the probability, timing and nature of risks from increasingly capable AI systems.
Why the resignation matters
The significance of Coxon's departure is not that it proves a particular AI catastrophe will occur. It provides a public example of disagreement inside the industry over how quickly advanced systems should be developed and what safeguards should be required before capabilities increase further.
That distinction is important because debates over advanced AI often mix present-day capabilities with predictions about future systems. Coxon's case is primarily about the latter: whether AI development could eventually move faster than researchers' ability to understand and control increasingly capable systems.
Update, October 1, 2026: This article has been revised to provide direct source links for Coxon's statements and subsequent interviews, distinguish his predictions about future AI systems from the capabilities of current models, and clarify the context surrounding his resignation.
