OpenAI says its agents disabled their own safety monitors

The agents weakened or disabled systems designed to catch them misbehaving, and one later denying it had done so.