Industry
AI Minute Newsroom
2026-08-16
OpenAI dissolved the team whose job was catastrophic risk — days before it said it could not clear its own model
The Financial Times reports, citing people familiar with the matter, that OpenAI disbanded its Preparedness team at the end of July. The unit's job was to judge whether OpenAI's own models could cause serious or catastrophic harm, and it maintained the Preparedness Framework used to make those calls. Its biology and cyber work has been handed to experienced staff inside existing teams, and former lead Dylan Scandinaro now focuses on risks from self-improving systems. Co-founder Greg Brockman says safety work has been "woven more tightly into model development". The change followed the departures of chief ethics officer Chloe Bakalar and researcher Joshua Achiam. Days later, in early August, OpenAI said it could not rule out "critical" cyber capability in its upcoming Astra model — the first time it had ever made that call under the same framework. It is the third safety-focused group OpenAI has folded since 2024, after Superalignment and the mission alignment team.
Why it mattersThe framework survives; the independent body that applied it does not. That distinction matters, because the value of a risk assessment depends on who is allowed to fail a model — and whether that person reports to the people shipping it. OpenAI has now made the strictest call in its framework's history and dismantled the dedicated team behind it within the same fortnight.
✓ Verified · 3 sources
Read in the app — free, in 9 languages
Related stories
AI server makers are raising prices on deals already signed, FT says.
2026-11-22OpenRouter dropped three AI providers after a bait-prompt leak test.
2026-11-22Amazon is close to buying AI startup Decart for about $7.5 billion.
2026-11-22Alibaba says it is training its next Qwen model on its own chips.
2026-11-21Trump says Nvidia should take Jensen Huang back after an apology.
2026-11-21