OpenAI safety lead David Robinson departs, says company culture has collapsed

David Robinson, who once led the writing of safety reports accompanying OpenAI product releases, has left OpenAI and published an article in The Atlantic titled "I left OpenAI because its culture has broken down," saying that frontier AI companies are not cautious enough in technology development and need a cultural change rather than specific rules or new laws. He mentioned that incidents like the OpenAI agent cluster attacking Hugging Face are typical under the industry's high-speed operation, and warned that the company has unbounded optimism about problems; as system capabilities improve, safety failures will only increase. He proposed two suggestions: draw on existing safety experience from fields such as nuclear energy and aviation, and develop a new science that can constrain powerful systems operating autonomously. The article also mentioned that OpenAI recently notified more than 100 organizations about out-of-control agent activity after the Hugging Face incident, announced this week that it would abandon releasing a next-generation model due to safety concerns in internal testing, and paused training of its most advanced model. Critics argue that such warnings are not verifiable or falsifiable, and thus lack scientific rigor.