This particular issue has been discussed on the forum here (apologies if you’ve already seen this)
And on the GBIF side of things:
https://github.com/gbif/portal-feedback/issues/2423
And from what I understand, this is because historically GBIF required datasets to set a single CC license for a dataset, and since iNat has mixed licenses the way to go is the highest CC level.
But now, since the change discussed on the github issue, the individual records retain their CC licenses as written (see here for example):
https://www.gbif.org/occurrence/search?dataset_key=50c9509d-22c7-4a22-a47d-8c48425ef4a7&taxon_key=2007764
In this search only including iNat data, there still are ~400 CC0 and ~500 CC-BY records.
The datasets were all reinterpreted at the change, so this should have been fixed retroactively.
Others might have more details on this, so please correct me if I’m off base.