It could be a lot of things! For example:

A PDF or other binary file stored in a text field might get misinterpreted as non-UTF-8 characters during a backup.

Similarly, audio or video files—or any kind of binary data—stored inappropriately in text fields could cause issues.

It could also be due to corrupt data or improper encoding when the data was inserted into the database.

Essentially, anything non-textual or incorrectly encoded could result in invalid UTF-8 characters showing up in a backup.