That kind of phishing already exists, even more sophisticated: a bug that a lot of software contains is not distinguishing between same looking characters in different alphabets. E.g. you can sign up on many forum/bbs platforms as Administrator if your leading A is cyrillic A instead of latin A. Both look the same but have different html entity codes and are different unicode chracatres, which is true for most vowels and many consonants (e.g. cyrillic B and latin B, C and C, E and E...).
What software (or library) is programmed to recognize that two chars look the same and therefore allows them based on the appearance rather than their encoding?