That is surely a violation of several different usage agreements. Certainly. Institutions don't hate this as much as publishers, but it endangers institutions'

That is surely a violation of several different usage agreements. Certainly. Institutions don't hate this as much as publishers, but it endangers institutions' access to publisher content and there are some who will work with publishers to find violators. That said, it may still be pretty hard to track if sci-hub's backend retrieval process that uses institutional accounts was written with care. I think that sci-hub may already be removing watermarks from papers now. I know that I've seen watermarked papers there before ("Downloaded from XYZ University at yyyy-MM-dd hh:mm:ss"). I just tried retrieving several recent papers in Elsevier, Springer, and Wiley journals from sci-hub. None of them showed watermarks.