The Catastrophic-Risk Team Is Gone
One of OpenAI's safety teams is gone. It was the team that evaluated catastrophic risks, named Preparedness. The FT reported its dissolution.
At 6:55 AM (KST) on August 17, 2026, Watcher.Guru posted about it.
The post says "OpenAI disbands its catastrophic risk evaluation team," citing the FT report. It has 212,000 views.
The team was dissolved in late July. Its biological and cyber risk work was redistributed to existing teams.
How the Safety Organization Was Shaken
The Preparedness team evaluated failure modes. Its scope was the cyber, biological, and autonomy risks of frontier models.
Dylan Scandinaro is the former team lead. His role changed to recursive self-improvement AI risk.
Co-founder Greg Brockman offered an explanation: safety work was integrated more tightly with model development.
Safety staff departures also continued. Chief ethics officer Chloe Bakalar left. Joshua Achiam and many others left the company. Unease was reported internally.
The continuity of the evaluation team is in question. That points to a structural change in how safety work is done.
How Should Enterprises Read This Signal?
Companies that use OpenAI models should take note. The evaluation structure has changed. The party doing safety evaluation has moved into the development organization.
The strength of independent verification may weaken. There is a possibility that evaluation gets tied to development schedules. This is an interpretation resting on reporting, though.
What companies should do is clear. Check the vendor's safety evaluation documents yourself. Re-review your standards with your security and governance leads.
Here are the changes in a table.
| Change | Signal for enterprises |
|---|---|
| Preparedness team disbanded | Safety evaluation org structure changed |
| Bio and cyber work redistributed | Need to confirm who owns evaluation and by what standards |
| Many safety staff departed | Check for governance gaps |
| Integrated into model development | Reassess the strength of independent verification |
The Internal Story Rests on Reporting
This news is based on press reports. It relies on the original FT report and a summary from the-decoder. OpenAI's official position hasn't been confirmed yet.
The departures and unease are statements based on reporting, not interviews with the people involved. No specific figures have been released either.
The trend is still clear. The center of safety evaluation has moved toward development. Companies should read this as a signal to check their governance.
Sources: Watcher.Guru post (2026-08-17 06:55 KST), original FT report, the-decoder summary (2026-08-16).