In my case, these errors started on a few endpoints after upgrading them from 9.5.5 to 9.5.8. I opened up a case with IBM (TS000130487) which resulted in APAR IJ05780 (identified a failure in transcoding the additional data that 9.5.8 collects to prevent computer duplication). I think that fix has made it into 9.5.10 (but not sure).
The troubleshooting process is to “walk” the IP address to the client. In other words, the IP shown on your top level relay (TLR) will be one of your child relays, the child relay will have the same error with another IP. That IP will either be another child relay or the client itself. What is worse, is that the client will jump from relay to relay when it has this symptom, so even a single problem client can cause many log entries on the TLR.
In my case at least, once finding the problem client, I downgraded them to 9.5.5 and the issue was gone.