Suppose that I am developing a service (DocBot) to aid the transfer of large mail attachments. Instead of attaching documents to her messages directly, Alice appends the hash of each document, and separately sends the hash to DocBot, which requests the full document if it does not recognise the hash.
When Bob receives the message, he asks DocBot for the the original document associated with the hash. For the sake of argument, DocBot does not require any additional information from Bob – if anyone knows the hash, that is taken as proof that they are authorised to see the original document.
My question is: assuming that the hashing scheme is sufficiently collision and pre-image resistant, and that no one can intercept Alice's messages to Bob, what new vulnerabilities are introduced by this scheme? That is, does DocBot create ways for a malicious user to learn information they shouldn't, or interfere with other people's communications?