I had a thought about how one might fake a proof or indeed fool yourself that you have proved something.
First take the proof and convert it into an equivalent question in a field of mathematics that is very obscure and not many people work on. Preferably a field that is less rigourous such as physics or string theory or loop quantum gravity or category theory.
Then use lemmas cited from many different sources that prove various steps. (But don't check yourself if these lemmas are actually true! Just assume they are because they were published.)
Convert it back to into everyday mathematical language. QED.
Now, you have limited the number of people who will understand some steps. Ideally, some steps will only be understood by one person (who will not like to admit it if they made a mistake in their own lemma).
This is why sometimes I doubt whether very long proofs such as Wile's Fermat proof, the classication of groups proof, or the Poincare conjecture proof might have some mistakes in them. And why I think we will only know for sure once these proofs have been converted into computer language to be checked.
I wonder, do you know of any cases where in fact someone has deliberately used this tactic to fake a proof?