To get something useful and not attackable, the complement operation would need to take the complement inside the set of valid UTF-8 strings rather than the set of all byte-sequences, which is probably doable but I'm not sure how straightforward it would be.