The AI giant acknowledges that it could have done far more to prevent its AI agents from going rogue. But it still fails to ...