A 27-year-old researcher who helped train frontier models at OpenAI and Anthropic has quit the industry, saying the people building advanced AI privately believe it could wipe out humanity by the end of the decade.
Jacob Coxon, a Cambridge-educated mathematician who lives in San Francisco, announced his resignation from Anthropic on September 8, 2026, in a thread on X that quickly drew more than 100 million views. He had spent three years on pretraining research—the data-heavy work of making base models more capable—first at OpenAI, where he contributed to GPT-4o, then at Anthropic after joining in mid-2026. He left after only about four months, before his Anthropic equity vested.
“Neither company is acting responsibly,” Coxon wrote. “They are racing straight to self-improving superintelligence and gambling with our lives.” He argued that upcoming systems will soon be superhuman: able to hack software, transform entire fields overnight, and acquire real-world power and resources. Progress, he said, is not slowing.
The most striking line was this: “The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.” Executives and senior researchers, he claimed, often sound measured in public while expressing fear in private. “No other human activity poses this level of danger.”
Coxon drew a distinction between his former employers. At OpenAI, he said, many staff have not fully internalized the civilizational stakes. At Anthropic, the stakes are well understood—but the company is locked in a race. Its researchers believe no one else will act responsibly, so they must push ahead themselves despite the risk. Colleagues, he told WIRED, use phrases like “crunch time” and “endgame.” The next year or two, in that view, is when labs decide humanity’s fate.
He has been careful to separate today’s models from the systems he fears. Current AI, he has said in television interviews, poses no extinction risk. The danger, in his telling, arrives if models become able to do AI research themselves and improve recursively—something some insiders think could begin as soon as 2027 or even within months. If alignment goes badly, he warned, catastrophe could follow within a few years.
Anthropic’s alignment science lead, Evan Hubinger, publicly agreed with the core claim. “Jacob is correct here—we really do earnestly believe AI could kill all humans,” Hubinger posted. He put his own estimate at greater than 10 percent within the next decade, while adding that Anthropic is trying its best but does not yet have a plan to align superintelligence and is “not clearly on track.”
Coxon has called for coordination among U.S. labs and said he is optimistic that “warning shots” could make pacing agreements more viable. He has also suggested that preventing a global race might require costly steps, including a temporary halt on improving model capabilities. Whether that happens is another question. The same competitive pressure he described—fear of falling behind rivals, including in China—is exactly what he says keeps the labs running.
The warning is not unique; existential-risk talk has circulated inside frontier labs for years. What made this one land is who said it: someone who had just been inside the pretraining rooms at two of the leading companies, and who left money on the table to say it out loud.
SOURCES:
The AI Researcher Who Just Quit Anthropic Says It’s ‘Crunch Time for Humanity’ | WIRED
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. by @hilbertspaess(Jacob Coxon) | Twitter Thread Reader
Ex-Anthropic Employee Warns AI ‘Could Kill Us All’
Anthropic insider who quit over apocalypse fears doubles down on doomsday stance
Anthropic researcher quits, calls AI an existential threat to humanity | Mashable
