It's not about the downloading of the data, it's about its use in training models, which is dubious from a copyright perspective.