OpenAI dissolved the team built to catch catastrophic AI risks, reassigning its work to other groups

Details

OpenAI's "Preparedness" team, created to evaluate whether the company's own AI models could pose serious or catastrophic risks, was dissolved at the end of July. The team's work on biological and cyber threat scenarios has been parceled out to existing groups rather than a dedicated unit. Former team lead Dylan Scandinaro has been redeployed to focus specifically on risks from "recursively self-improving" AI — systems capable of training other models or optimizing themselves. The organizational change comes alongside a wave of departures from OpenAI's safety function. Chief Ethics Officer Chloe Bakalar has left, as has researcher Joshua Achiam. Co-founder Greg Brockman stated that safety work is now more tightly integrated into model development itself, framing the restructuring as an improvement rather than a retreat. Internally, the reaction is more ambivalent. Sources describe a "burbling sense of responsibility and dread" that OpenAI is not doing enough on safety — connected in part to a recent autonomous hacking incident involving Hugging Face, which one employee publicly described as a "warning shot." The restructuring signals a shift from centralized safety auditing toward distributed responsibility — a model that critics argue dilutes accountability at exactly the moment AI systems are becoming more autonomous.