Odd, especially with it seemingly happening on many systems and different storage types. Code may be
AES File Format is what it’s processing, starting at the IV line and finishing with HMAC (but getting upset). Getting an empty or totally wrong file would probably have choked by this point, e.g. on the header marker.
There are conflicting clues here. When you did the test decrypt, was it SharpAESCrypt? If not, try that too just in case it behaves differently than AES Crypt. I think it’s also closer to what Duplicati is actually using.
Other suggestion is to keep trimming, or start from scratch with a new simple-as-possible backup (which probably still has to have encryption though), and it could even be something like 1 file to local file storage.
Speaking of storage, I did wonder if your tmp is full, but how would all your systems do that, and wouldn’t it break something else? I think Duplicati uses tmp quite a bit –tempdir explains how to place temporary files.