An Alarming Exit: Ex-Anthropic Researcher Warns AI Could 'Kill Us All' This Decade

Imagine dedicating your career to the cutting edge of artificial intelligence, only to walk away in fear, convinced that the very technology you helped build could spell humanity's doom within years. That’s precisely what Jacob Coxon, a former pre-training researcher at Anthropic and previously OpenAI, has done. He’s not just left his job; he’s exited the entire AI industry, carrying with him a deeply unsettling message: those building the most advanced AI models "earnestly believe that it could kill us all by the end of the decade."

Coxon, who resigned from Anthropic on September 9, 2026, didn't mince words. He publicly accused both Anthropic and OpenAI of acting irresponsibly, likening their current trajectory to a perilous sprint towards self-improving superintelligence. "They are racing straight to self-improving superintelligence and gambling with our lives," he declared, painting a grim picture where, by the end of the following year, things could spiral "out of control already." It’s a stark accusation, suggesting that behind the polished press releases, a deeper, more profound anxiety festers within these industry giants.

What makes these warnings particularly chilling is that Coxon isn't alone in his dire assessment. In a startling confirmation that sent ripples through the AI community, Evan Hubinger, Anthropic's Alignment Science Lead, publicly affirmed Coxon's fears. "Jacob is correct here — we really do earnestly believe AI could kill all humans!" Hubinger stated, even offering his own personal, unnerving estimate: a "greater than 10%" chance of AI causing human extinction within the next decade. He highlighted a critical flaw: Anthropic, despite its rapid progress, "do not yet have a plan to solve alignment for superintelligence and are not clearly on track to." The fear, he elaborated, stems from "superintelligence arising from recursive self-improvement, as we have said is happening faster than we thought."

This isn't an entirely new sentiment in the upper echelons of AI research. Back in July 2026, the "Pacing the Frontier" letter, signed by over 1,100 AI company employees—including Anthropic CEO Dario Amodei—warned of a "real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems." While Anthropic’s official spokesperson maintains the company has "always been transparent that AI will bring both enormous benefits and unprecedented risks," and that they "continue to build models with some of the strongest safeguards in the industry," Coxon’s insider perspective suggests that intense competition makes genuinely cautious trade-offs incredibly difficult to prioritize.

Of course, not everyone shares this apocalyptic outlook. Some within the field dismiss such warnings as little more than "science fiction-inflected doom-mongering," a dramatic overestimation of current capabilities. Yet, the gravity of these claims has captured the attention of policymakers. Figures like Senator Bernie Sanders have called for new restrictions on AI, and Representative Greg Casar has even announced legislation aimed at banning artificial superintelligence outright. As the debate rages, and the technology advances at breakneck speed, humanity finds itself at a pivotal, perhaps perilous, crossroads, grappling with the profound questions posed by its own ingenuity.