Many HSMs do not force you to use unique labels for keys. So it's somewhat understandable if a human mixed up some keys during a non-standard or deviated process. This doesn't explain how so many different keys were "generated" so poorly or why no one cared about mismatched signatures.

And the discussion mentioning wrap/unwrap testing kind of feeds directly into the fear that either:

1. These weak keys were created outside the HSM and then imported.

2. These weak keys were meant to be the wrap keys somehow. Like, if they hadn't paid for some of the wrap/unwrap features on one of their HSMs and so were using easy-for-humans keys to handle the wrap/unwrap operations so they could more easily move keys between HSMs.