There is a “potential vandalised” detection for IP users already but bot’s behaviour can be improved a bit. One idea is a delay before fixing redirects that human users can react in meantime. Another idea is to look for the last editor and check whether his edit looks trusty enough. Also checking the block log would help as you can see in this reported case.
I guess it does not otherwise I would assume that such a scoring is available for abuse filters but it isn’t. A scoring of trusty edits could be derived from user rights and edit counts and account age. This is a way most abuse filters take into account.