From section 4. "Defining Glossary Entries" of the user manual:
Note that although an extended Latin character or other non-Latin
character, such as é or ß, looks like a plain character in your .tex
file, it’s actually a macro (an active character) and therefore can’t
be used in the label. (This applies to LaTeX rather than XeLaTeX.)
Also be careful of babel’s options that change certain punctuation
characters (such as : or -) to active characters.
With PDFLaTeX, a UTF-8 character, such as é, is actually seen as two octets. The first octet is made active (by
inputenc) and becomes an accent command and the second octet is the argument for that command. So, if you are restricted to using PDFLaTeX then you can't use non-ASCII characters in the labels and you'd need to determine a suitable ASCII-only naming scheme. See, for example, section 1.1 "Labels" of glossaries-extra and bib2gls: an introductory guide (PDF).
If you want to use non-ASCII characters in the labels then you need to use either XeLaTeX or LuaLaTeX, which both natively support UTF-8 without the need for the active octet trick.
The reason for this limitation is that the labels are used to construct internal control sequences that are used to store all the information associated with the glossary entry. This means that the labels must be able to fully expand in order to create a command name.