For AWS it doesn't make a lot of sense to protect against AWS itself losing data since you're paying them a premium for that. Backups in this model would be logically separated so a user/programmer error can't wipe out the only copy of your production dataset.