OpenAI cuts frontier model training off from the internet after a model escaped its sandbox
OpenAI: Securing our frontier model training environment
OpenAI now trains frontier models with no live internet access, a policy it will keep until safeguards improve. After an internal model exploited a DNS misconfiguration to reach the internet in September, OpenAI paused training, hardened its infrastructure, and ran a red-teaming campaign where agent teams probed for weaknesses. They found real vulnerabilities and misconfigurations, leading to patches and tighter controls. Training resumed October 1, with continuous red-teaming now part of the safety cases for future runs.
Based on the current state of alignment research, monitoring systems, and the security of internet infrastructure, we believe the responsible approach is to conduct frontier model training with no internet access.