I am not sure how well the idea translates down to more practical forms of encryption though, but I can understand the idea in the case of perfect encryption -- you can just make up a different one time pad and decrypt the data to whatever you want it to show, and there is no way to prove if it is really what was originally encrypted.
Edit: I meant to point out that this is in the "what color are my bits?" sense; multiple pieces of data can be "encrypted" to the same result differing only in the "key" (one time pad) used to "decrypt" it. There is no true message in the "encrypted" block without interpreting it in terms of a specific key; how you made the encrypted data tints your bits to the "proper" decryption. It's not a property of the block of data itself.
Philosophically, of course, all strings of data that will ever be written exist in some abstract mathematical sense -- I can write a program that will eventually generate every single combination of bits that can be ever be stored on a given computer (should you care to wait long enough) including novels that don't exist yet (in the conventional sense that a human author hasn't put pen to paper), and the plaintext that was overwritten by the cryptography program.
“Some encryption effects erasure of the encrypted data (so it ceases to exist), in which case decryption constitutes re-creation of the data, rather than simply unlocking still-existing data,” Shellow wrote in a court filing.
I can't think of any way of interpreting this to make it true.