Alignment

OpenAI details layered monitoring for misaligned frontier models

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