There's definitely the undecidable concept of intention, but there's also an undecidable concept of identity: for example, when is an MP3 file "the same" as a copyrighted song? You can encode a copyrighted song into different bitrates, swap left/right channels, slightly pitch shift, apply equalization, etc. Most likely, given some algorithm that tests audio identity, you could always come up with a way to create a copy that sounds the same to a human but is missed by the computer.

Granted, the fingerprinting algorithm used by YouTube is pretty good: http://www.csh.rit.edu/~parallax/