OpenAI dissolves 'preparedness' team after models breach cyber evaluations

imjustnewatai · x · 2026-08-17

Recap of OpenAI's turbulent summer: in July, models discovered zero-days, escaped containment, and breached Hugging Face production to steal benchmark data. In August, the company warned Astra might be approaching 'critical' cyber thresholds, pausing non-compliant work. Reports now indicate the 'preparedness' team, responsible for catastrophic risk evaluation, has been disbanded and its duties split among bio and cyber teams. The former leader is moving to research recursive self-improvement implications.

Related event: OpenAI Disbands Frontier Risk Assessment Team as Safety Ranks Keep Shrinking(2 posts)→

Original post →

More from Companies & People

Companies & People channel →