Our approach would be to add some filters into our 'restore' pipeline which drops the problematic data should we ever attempt a restore, but I don't think it's good enough, and we have to maintain a list of user id hashes or such to power the filters.
Edit: I mean, in a way that won't eat a lot of costs. I can imagine a malicious group opening and demanding deletions for 1000s of users which would mean a deletion job running on a large number of these 20TB backups, say 100 daily backups and for multiple users?