Longtime OpenAI employee resigns over the company’s safety culture

David Robinson resigned after three and a half years at OpenAI, where he says he led safety-report writing and helped draft the current Preparedness Framework. In a first-person Atlantic essay, he argues the company’s trial-and-error culture is inadequate for advanced AI and calls for redundancy modeled on nuclear plants or airports, outside safety expertise, and new science for systems operating without human oversight.