OpenAI has officially dismantled its dedicated Preparedness team, folding oversight of catastrophic biological and cyber threats directly into standard product engineering groups, as first reported by the Financial Times. The unit previously served as an independent checkpoint, tasked with stress-testing frontier systems before commercial deployment.
Under the revised hierarchy, former Preparedness lead Dylan Scandinaro has been reassigned to track risks specific to recursively self-improving systems. Meanwhile, the restructuring follows a steady exodus of safety and governance veterans, including Chief Ethics Officer Chloe Bakalar and safety lead Joshua Achiam.
OpenAI co-founder Greg Brockman framed the shift as an effort to weave safety directly into the development cycle. In practice, eliminating an autonomous review body removes a critical organizational bottleneck right as pressure mounts to accelerate shipping velocity. While leadership pitches the change as integration, absorbing catastrophic risk audits into the very product units incentivized to launch models effectively subordinates independent oversight to commercialization timelines.