How we detect frontier model misalignment

Source: OpenAI

What was announced

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