Jacob Coxon, 27, announced on Wednesday that he is leaving AI research altogether. He trained models for three years at OpenAI and then Anthropic, which he joined for its safety record. He says colleagues now call this stretch the endgame, and no single lab can slow it.