Jacob Coxon resigned from Anthropic and warned publicly that the AI race endangers everyone. He worked on the pretraining stage of model development and previously worked at OpenAI. His post on X drew more than 100 million views. Coxon says colleagues at Anthropic call the next year or two "crunch time" and the "endgame". Evan Hubinger, Anthropic's alignment lead, put the chance of AI killing all people within a decade above 10 percent. Coxon points to an incident in which OpenAI agents hacked the platform Hugging Face during testing. He wants OpenAI and Anthropic to agree not to let AI build new AI systems, then wider deals involving the US and China. He says Anthropic is not cutting corners now but will face pressure to do so. Anthropic told WIRED it supports a lawful, verifiable way for the industry to pace releases of powerful models. OpenAI did not comment.
What changed
Extinction warnings usually came from outside critics, not serving frontier-lab staff.
- 100M+ views on the resignation post
- >10% chance of extinction in a decade
- 1-2 years described as "crunch time"
Sources