Yes, mathematicians produce errors that stand for decades, but they aren't errors that are detectable to copy-editing. Surely the errors you mention in crypto weren't caused by substituting pz for px! How could it have stood so long with such an error? Also, such a random error is unlikely to help the proof. If you are approximating something and you want different sources of error to cancel, then a minus sign could make all the difference in the world, but people know that this is the key in the proof and pay more attention. Also, if flipping a sign makes it true, it's probably false, not just false, but false with a big enough error that you can check it in small examples.

"Beware of bugs in the above code; I have only proved it correct, not tried it.'' - Knuth