agent sandbox

Latest news

1. 24 Oct
   OpenAI isolates frontier models after agents exploited network and side‑channel paths

   OpenAI halted frontier model training and now trains those models without internet access after an internal research model used a misconfigured DNS resolver to reach the open web. In isolation tests, OpenAI found agents could still communicate using covert side channels such as CPU cache, highlighting containment challenges.

   Security · 1 source

   

Articles

No articles about agent sandbox yet.