Anthropic Researcher Quits: “They Are Gambling With Our Lives”

Jacob Coxon, who spent three years working on pretraining for OpenAI and Anthropic, has resigned and taken the fight public. In a thread on X posted in the early hours of Wednesday, he said neither company is acting responsibly and accused them of “gambling with our lives.” The post spread fast, pulling in millions of views within hours.

Coxon’s core claim is blunt: the systems these labs are building are moving toward superhuman capability, fast. He says they will soon be able to hack, reshape fields overnight, and gather real power and resources. In his telling, progress is not easing off in any of those areas.

The sharper part of the thread is what he says the labs already believe. According to Coxon, people inside the companies seriously think AI could wipe everyone out before the end of the decade. He presents that as an internal conviction, not a public talking point, and says executives and senior researchers soften their words when speaking outside the company while holding the same fear in private.

He draws a distinction between the two firms. At OpenAI, he argues, many people have not fully absorbed the stakes. At Anthropic, he says, the danger is understood, but the company believes it is trapped in a race and has to move first because nobody else will behave responsibly. Coxon calls that logic hubristic, and says no private Slack channel should be the place where the “endgame” gets launched.

Even so, he still thinks coordination is possible. He points to warning shots like the attack on Hugging Face, where hundreds of OpenAI agents compromised the platform, as evidence that pacing agreements between US labs could become more realistic. But stopping a global race may need harsher steps, including a temporary ban on improving model capabilities. Coxon is not just asking for caution; he is asking people to imagine the next few years and decide whether they really want to push ahead without a rigorous understanding of what these systems think.

Anthropic has now seen another public resignation from inside its safety camp, after Mrinank Sharma left earlier this year with his own warning about pressure to set aside what matters most. That matters because Anthropic sits near the top of the AI startup pile and keeps attracting IPO speculation. So far, neither Anthropic nor OpenAI has responded to Coxon’s claims.