The use of openly readable datasets by machine learning models raises questions on the boundaries of fair use and limitations in creating derivative works on future platforms.