Former Anthropic AI researcher Jacob Coxon said he resigned after concluding that Anthropic and rival OpenAI are competing to develop autonomous superintelligence that could ultimately pose an existential threat to humanity.
Superintelligence and rogue AI once seemed like themes confined to science fiction. A recent warning from a former Anthropic researcher has nevertheless brought those risks into the center of a serious debate about the direction of current AI development.
Coxon’s claims allege that the world’s two leading AI laboratories are pursuing increasingly capable autonomous systems. He argues that a race toward self-improving superintelligence could outpace existing safeguards and create consequences far beyond the technology’s present capabilities.
A Former Anthropic Researcher Sounds the Alarm
In a lengthy X post, Coxon said he had resigned from Anthropic because of concerns over the company’s strategic direction rather than the prospect of better employment elsewhere.
Coxon stressed that he was not issuing the warning as part of a public-relations campaign or promotional effort. He said he had spent the previous three years conducting pretraining research at both OpenAI and Anthropic. According to his account, neither organization has acted responsibly in its pursuit of the first superintelligent system, exposing society to what he describes as an unacceptable collective gamble.
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
Coxon cautioned against underestimating how rapidly AI capabilities could advance. He said the technology has not reached its limits and that its progress may be accelerating. Without effective oversight, he warned, it could soon produce systems capable of surpassing human performance in cybersecurity, scientific research, infrastructure control, and other critical areas.
In his assessment, uncontrolled superintelligence could emerge by the end of the decade. Coxon said he has heard similar concerns expressed privately by executives and senior officials at leading AI organizations, even as companies continue to reassure the public that genuinely autonomous AI remains a distant prospect.
According to Coxon, the principal obstacle to slowing development is competitive pressure. Labs may believe that reducing their own efforts will not stop rival companies, governments, or other actors from achieving comparable capabilities first. That fear can drive every participant toward faster deployment despite the growing risks.
An Appeal to Researchers and Industry
Coxon called for coordinated action across laboratories, industries, and civil-society groups before the risks become difficult to control. He acknowledged that both approaches carry danger: halting or restraining development could concede an advantage to less cautious actors, while continuing toward superintelligence could allow the technology to escape existing controls.
“If you are a lab researcher, I urge you to consider what the next few years will actually feel like. Do you want to kick off a superintelligent RL run without a rigorous understanding of its mind?” Coxon wrote. “Should you put your head down because it’s happening anyway or take this moment to call for different conditions?”
His comments have since triggered widespread discussion about whether AI regulation should be tightened or whether major companies should continue operating without greater restrictions. The central dispute remains whether a coordinated pause would reduce the danger or whether an uncontrolled race would simply shift the advantage to governments, hostile groups, or laboratories that refuse to slow down.
