Ok, looking at the B&R logs, it seems that a recently added SSH username and password was the culprit. Deleting that credential now allows a successful config backup.
Odd that the credential worked fine for its purpose also strange that Veeam / SQL Server couldn't fix the corruption. Would also be nice if that error message from the logs was put into the job email.
Had a same case with support, #04318948.
Eventually resolved by recreating (create new, remap to new object, delete old) SSH credentials, we didn't exactly determine which one was broken, we just replaced them in bulk. Support has the database dump so maybe they can do some post-analysis.
Very strange case as all the credentials were being used daily by jobs/components and they worked - just configuration backup kept failing.
About a week ago, Linux FLR started failing because it couldn't find SSH credentials in database. Support fixed it with a SQL statement (#04423439). It just seems to be related to problems with credential handling in database.
Now, configuration backup broke again as I've had to change Linux credentials several times due to another case... ugh. Back to support again. Seems to be related to either of 2 actions
editing description (changed dummy descriptions from previous credential recreations to real descriptive ones)
switching from user+root -> root-only -> user+root authentication methods - as required during diagnosis of case.