An Anthropic researcher, Jacob Coxon, has said he is leaving the AI industry,
citing refusal to participate in a sector-wide race to develop self‑improving
systems that he fears could get out of control. Coxon, who works on training
large models with massive datasets, said colleagues now use terms like
“crunchtime” and “endgame” to describe progress toward self‑improvement. He
warned the sector is moving toward its most aggressive scenarios and said the
situation could be out of control by the end of next year, adding that
competitive pressure between firms makes safety compromises likely.