From OpenAI Blog

How we detect frontier model misalignment

OpenAI uses layered monitoring to detect frontier model misalignment and unauthorized agent behavior so responders can intervene before incidents escalate.

Read the full report at OpenAI Blog ↗