They mention "not malicious", but I wonder if current controls are strong enough to prevent malice if the objective is to "interpret intentions disastrously". Isn't this irresponsible?