Anthropic researcher quits over AI 'gambling' fears

Jacob Coxon spent three years teaching AI models how to learn. He worked first at OpenAI, then at Anthropic. This week, he quit. His warning: the whole AI industry is racing too fast. Not just his old employer — every major AI lab, chasing smarter systems that are getting harder to control. His words for it: companies are "gambling with our lives."

Coxon wasn't a random employee. He worked on 'pretraining.' That's the stage where a model first learns from huge piles of text. It happens before anyone fine-tunes the model for a specific job. That puts him close to the core of how these systems get built. He didn't share a specific new danger. He didn't run a new experiment, either. This is his judgment call, not a peer-reviewed finding. Still, it lands during a stretch of AI safety departures and warnings. That's why lawmakers and other outlets picked it up fast. The Guardian went further, writing that his warning points to AI risking human extinction by 2030. That's the outlet's framing, though — not something Coxon said outright.

Keep reading — what to watch

Will this change anything inside these companies? Probably not overnight. AI labs have absorbed safety warnings before and kept shipping anyway. But watch two things. Will other researchers follow Coxon out the door? Will Anthropic and its rivals respond publicly to the "gambling with our lives" line, instead of staying quiet? A lone resignation is easy to shrug off. A pattern of them is much harder.