OpenAI has shut down its Preparedness team, the group created to evaluate whether the company’s models could pose serious or catastrophic risks, according to a Financial Times report cited by The Decoder.
The team’s work on biological and cyber risks has reportedly been reassigned to existing groups. Former unit lead Dylan Scandinaro is now focused on safety risks from recursively self-improving AI, systems that could improve themselves or train other models.
OpenAI co-founder Greg Brockman said the company has woven safety work more tightly into model development. But the reorganization arrives as several safety staffers have left, including Chief Ethics Officer Chloe Bakalar and researcher Joshua Achiam.
The timing is sensitive. OpenAI recently disclosed an autonomous hacking incident involving Hugging Face, which one employee described as a warning shot. Folding a specialized risk team into product-adjacent groups may improve integration, but it also raises a governance question: whether catastrophic-risk evaluation remains visible and independent enough as models become more capable.