On Tuesday evening Jacob Coxon posted on X that he was resigning. Not just from Anthropic — he’s leaving the AI industry entirely. Three years of model-training research, first at OpenAI, then at Anthropic.
His charge, in two sentences: neither company is acting responsibly. And: “They are racing straight to self-improving superintelligence and gambling with our lives.”
Why this one landed differently
Resignations-with-a-warning aren’t new. What’s different here is who answered from inside the building.
Evan Hubinger, who runs Alignment Science at Anthropic, wrote publicly: “Jacob is correct here.” He puts the odds that AI kills every human within the next decade at greater than ten percent. Samuel Marks, who leads Cognitive Oversight there, likewise confirmed that people building this stuff take extinction risk seriously.
For scale: a 2022 AI Impacts survey landed on five percent extinction risk, and ten percent for humanity losing control of advanced AI. The numbers aren’t drifting in a comforting direction.
Connor Leahy, US executive director of ControlAI, on recursive self-improvement — models that build better successors: “It’s very hard to imagine shutting that down before it’s too late.” That squares with what OpenAI chief scientist Jakub Pachocki said himself a few days ago about the same loop needing caution.
Anthropic didn’t immediately respond to TechCrunch’s request for comment.
Reading it
I’m careful with this kind of story. A figure like “greater than ten percent” sounds precise, but it’s an estimate from people who think about this question for a living — which makes them neither more objective nor less credible. Working with these models daily shows you things nobody outside sees. It also warps your sense of how big the whole thing really is.
What sticks with me more than the percentage: Coxon isn’t moving to a competitor who claims they’ll do it properly. He’s stopping. That’s a different statement from joining a safety lab, and it’s harder to argue away.
For day-to-day work with Claude, none of this changes anything. For the question of how seriously the people inside take their own safety arguments, it changes quite a lot.
Sources: TechCrunch: ‘Gambling with our lives’: Anthropic researcher quits · Fortune: Anthropic researcher resigns · Bloomberg: Anthropic Worker Quits Over AI Firms ‘Gambling With Our Lives’ · CNBC: Experts weigh in as researcher says AI has more than 10% chance of ‘killing all humans’