Or even vulnerabilities. Imagine a scenario where a parser for an authentication engine reads a different value for a given key than the value the authorization logic reads.
I don't know if there's a name for this class of problem. I'd be interested to know.