If a remote file read fails for some reason, Reader will try to copy the file to the working directory and open it there. I think this is fine for single files, but if you're working with large datasets it is potentially dangerous. I just had an incident where it copied 11 GB worth of files from dCache without me realising what was happening.
pyread.py#L82-L90