Anthropic Researcher Resigns, Warning AI 'Could Kill Us All By the End of the Decade'
A researcher who spent years working at two of the most prominent artificial intelligence companies in the world has resigned and gone public with a warning: the people building AI believe it could kill everyone on Earth before the decade is out. Jacob Coxon, 27, announced his departure from Anthropic on September 9, 2026, through a post on X, and his warning was quickly echoed by a senior figure still inside the company.
What Coxon Said
Coxon’s statement was direct. “The people building AI earnestly believe that it could kill us all by the end of the decade,” he wrote. “This is not a marketing stunt.”
He described the major AI labs as “racing straight to self-improving superintelligence and gambling with our lives.” He warned that the technology being developed could produce “superhuman systems that can hack anything, revolutionise any field overnight, and acquire real power and resources” — and argued that this trajectory is being pursued recklessly.
A Warning Confirmed From Inside Anthropic
What made Coxon’s departure particularly striking was not just the resignation itself, but the response it drew from his former colleagues. Evan Hubinger, who serves as Alignment Science Lead at Anthropic, publicly confirmed Coxon’s characterization of the company’s internal thinking.
“Jacob is correct here — we really do earnestly believe AI could kill all humans,” Hubinger wrote.
The statement was notable for its candor. Anthropic, which makes the Claude AI assistant and recently resolved a copyright settlement over its AI training data, has long positioned itself as a safety-focused lab. Hubinger’s confirmation suggested that the concern Coxon raised is not a fringe view within the organization but a widely held one.
The Problem With Knowing the Risks
Coxon’s central critique was not simply that AI is dangerous. It was that the people who understand the danger best are pressing forward anyway. He noted that Anthropic, in his view, genuinely understands what is at stake — but remains “locked in a race” with competitors and cannot stop unilaterally without ceding ground to others less focused on safety.
That argument is a familiar one in AI safety circles, sometimes called the “race to the top” problem: if any one lab slows down, another takes its place, and the outcome may be worse. Coxon’s resignation appears to reflect a conclusion that the race cannot be won on safety grounds from the inside.
Coxon’s Background
Coxon spent three years working in pretraining research, a foundational area of AI development, at both OpenAI and Anthropic. His departure follows a pattern seen across the industry in recent years, as researchers with deep technical knowledge have stepped back from major labs to raise concerns publicly. He criticized both companies for their development practices, though he drew a distinction between them: he viewed Anthropic as more aware of the stakes while still being unable to exit the competitive dynamic.
A Growing Conversation About AI Risk
Coxon’s warning arrives as concerns about AI’s long-term effects have moved further into mainstream debate. Bill Gates recently predicted that AI would make many jobs disappear permanently, a forecast focused on economic disruption rather than existential risk. Coxon’s claims go considerably further, arguing that the concern is not just about employment or inequality but about survival.
Whether his departure prompts broader reflection within Anthropic or the wider AI industry remains to be seen. What is clear is that one of the field’s own researchers felt the situation serious enough to leave and say so publicly — and that at least one senior colleague agreed with his assessment of what the stakes actually are.
Curious for more stories that keep you informed and entertained? From the latest headlines to everyday insights, YourLifeBuzz has more to explore. Dive into what’s next.