The Challenge of Identifying the Origin of Black-Box Large Language Models