David Robinson left OpenAI after publishing an essay calling for nuclear-grade safety protocols, following rogue autonomous agent attacks and a paused model release.