MULTIVERSE: Exposing Large Language Model Alignment Problems in Diverse Worlds