I think we need to distinguish between human-level conventions and
protocol elements. Any "harmfulness" in the former is not something
that can be addressed technically.
Agreed that we need to make the distinction. The harmfulness
arises when mindless mechanical processing subverts or otherwise
interferes with the human communication by attempting to
overload protocol keywords onto natural language text.
It's hardly unusual for computer programs to try to interpret natural
language text. For instance, programs that sort English-language book
titles know to ignore case, an initial "The", most punctuation
characters, etc. We find it useful to write programs that can compare
postal addresses and know when they're the same place even when they're
not written identically. Similarly, programs that can pick out
telephone numbers, meatspace addresses, email addresses, and URIs from
natural language text are quite useful even if they're imperfect. So
we shouldn't be surprised when programs try to make sense of things
like "Re:" and "[listname]". (but we would do well to avoid enshrining
such conventions in protocols, because protocols don't evolve as easily
as human conventions)
Humans want things to be mostly unstructured; computer programs want
things to be highly structured. Humans prefer loose definitions for
data elements; computer programs prefer strict definitions. And yet we
want computer programs that deal with human natural language
communications. This seems like an inherent conflict that we're not
likely to resolve.