Two more examples examples for arXiv and one for Figshare. (1) Article in journal has a flaw. For some reason the journal is slow or unhelpful concerning a correct version, so the author posts on arXiv a revised and perhaps much bigger version. Filters (will these be on arXiv?) detect a copyright infringement. (2) I translate an Euler article and post it on arXiv (with all attributions). Or an English translation of an obscure German or Russian article by a great mathematician. (3) I post on Figshare (and GitHub?) the meat of a published article (like data, procedures, results of experiments, programs) in order to make the research Open Science. I explain the context but maybe I don't pass the automated filter which detects a copyright infringement. For all examples, same questions as before.
If something was a copyright violation yesterday, it will still be a copyright violation tomorrow. And vice versa.
If you fear that arXiv will implement filters that will recognize that what you are uploading is a translation of an obscure Prussian research paper and block the upload, I don't think that's going to happen. (But I will be really impressed if it happens!)