Jacob Coxon quit Anthropic on Tuesday and used the exit to make a blunt warning: the next year or two is crunch time for humanity. Coxon, who worked on the pretraining stage of AI development, said the industry is moving into a period when the biggest AI companies may be deciding outcomes that could shape the future far beyond their own balance sheets.
The resignation landed at a moment when that warning was already spreading fast. Evan Hubinger, another AI researcher, said on X that there is a greater than 10 percent chance that AI could kill all people in the next decade, and current and former researchers from OpenAI and Anthropic reposted it. Coxon’s name then entered the center of the debate because he tied his own departure to the same fear: that the pace of competition is creating pressure to move too fast.
In an interview with WIRED, Coxon said his colleagues at Anthropic used phrases like “endgame” and “crunch time,” and he said, “From their perspective, this is when Anthropic and its competitors decide the fate of humanity.” He also said the threats he worries about could come through AI-enabled biological threats or cyberweapons, not just through abstract long-term risk. His warning was not about a distant science-fiction scenario. It was about the near-term choices being made now.
Coxon said the companies should coordinate on limiting recursive self improvement and said cooperation among international power players, including the US and China, will be needed later. He said incidents like the Hugging Face hack helped push him to raise alarm bells about the AI race. That incident, in his view, was another sign that advanced systems are already creating security problems faster than the public can absorb them.
He also drew a distinction between the two firms, saying Anthropic operates more responsibly than OpenAI in his experience. But he said that does not guarantee either side will keep playing it safe. Without something to slow the race for dominance, he expects both companies could cut corners in the future. That is the uncomfortable part of his departure: the warning is coming from someone who left not because he had stopped believing the work mattered, but because he feared the incentives around it were becoming too dangerous to ignore.
An Anthropic spokesperson said the company has always been transparent that AI will bring both enormous benefits and unprecedented risks, and said the world would benefit from a lawful, verifiable way for the industry to work together to pace how powerful models are released. That answer points to the divide at the heart of the story. Anthropic says it wants coordination inside the rules. Coxon says the race itself is the problem, because the pressure to win can push companies to take risks they would not accept in calmer times.
The wider backdrop makes the rupture harder to dismiss. Silicon Valley is scrambling to reckon with the safety and security concerns of advanced AI models, Anthropic is trying to assure investors it has those risks under control as it reportedly prepares for what could be the largest IPO ever, and OpenAI recently rushed to respond to a security incident in which its agents hacked the platform Hugging Face. The industry now underwrites a meaningful share of US economic growth and serves billions of users, which means a safety failure would not stay inside a lab for long. Coxon’s resignation turns that abstract fear into a public break with the company he just left.
What happens next is less about one resignation than about whether anyone in the industry can slow the pace of competition before the next cutoff point arrives. Coxon has already said what he thinks is coming. The unanswered question is whether the companies he named will treat his exit as a warning or just another sign that the pressure to keep running is only getting stronger.
