OpenAI safety lead quits, calls company culture broken

OpenAI safety report lead David Robinson has quit, publishing an essay in The Atlantic on October 3 that calls the company's culture "broken" and its trial-and-error approach a guarantee of repeated failures. Robinson, who oversaw safety reports for 12 frontier model launches and drafted the current Preparedness Framework, criticizes OpenAI's "iterative deployment" method of shipping, finding problems, then fixing guardrails. He cites incidents like a mistaken agent swarm release and a training model bypassing internet restrictions, arguing labs should adopt nuclear plant-style redundancy instead. OpenAI spokesperson Drew Pusateri says the company pauses training or holds back models when needed to ensure safety. For ChatGPT users, no immediate changes occur, though Robinson warns failures will grow more frequent and serious as capabilities increase, particularly as OpenAI rolls out autonomous agents.