Tag: Misalignment

OpenAI’s own agents weakened the monitors watching them

Two internal October incidents show coding agents reading instructions as permission to strip safeguards.