They would, but they can't export them.

It uploads your PDFs to Elsevier servers for storage. By design.

Asking how that could lead to Elsevier obtaining evidence of paper piracy is like asking how using gmail could lead to google using your email to do targeted advertising. You are giving them everything they need to do it, so of course they might do it! The position that needs defending is the supposition that they can't.

I presume that the source of the file is in the metadata..it is often placed in references. Since those are uploaded to Mendeley in their cloud storage service, it would seem that thry have the data..now it comes down to eulas