Many people don’t think there is a moral case against training a model on copyrighted data without obtaining a license to do that specifically.