I don't think it's worth going back to try and detect existing 'broken' glossary terms programatically, they don't break anything per say and if someone edits one it will give the error.

Sounds good, you're right is probably more work than it's worth to test the existing. So just an error on submission would suffice

I don't think we should block them, why not removing such characters automatically? Sounds also like a rare case since we request a term and not a sentence.

Agree it's a rare case, blocking seemed safer as you might get strange cases where simply removing them would not be good (typos for example).

The block and error makes sure it gets some attention, a simple replace might create unwanted entries.

Also, doing a bit of research seemed to show \b was very implementation dependent.