OpenAI has disbanded its Preparedness team, moving frontier-risk responsibilities into other internal groups.

The Financial Times reports the change happened at the end of July, with senior staff taking ownership of areas such as bio and cyber inside existing teams.

OpenAI created the team to connect capability assessment, evaluations and internal red teaming across catastrophic risks including cybersecurity, biological threats and autonomous replication.

The reorganization does not mean OpenAI has abandoned its Preparedness Framework, which the company still describes as the foundation of its approach to severe frontier-model risks.

Instead, the work is becoming more distributed.

IMO, one possible upside of the reorganization is that bio and cyber safety experts can work directly with the groups building and operating the systems. The Hugging Face incident showed that safety depends on things like sandboxing, network controls, and monitoring, not just model evaluations. Embedding safety specialists closer to operational work could help OpenAI catch system-level risks earlier.

The tradeoff is that decentralized safety ownership may reduce independent resistance when product and safety priorities conflict.