Sorry for the slow reply.
Certainly, you could simply not publish the dataset at all but most people who use Dataverse are interested in the *metadata* (not necessarily the files!) being discoverable via DataCite, Google Dataset Search, and other search engines.
I don't know if this helps at all but IP Groups are similar to Institution-Wide Shibboleth Groups* in the sense that inclusion into the group is determined at run time. In the code we use the term "explicit group" to mean that membership into that group explicitly assigned to individuals (or other groups).
A factoid is that we tried to drop IP Groups support in the rewrite from DVN 3 to Dataverse 4 (we feel that Shibboleth-based groups offer better security) but Scholars Portal really, really wanted us to keep them so we did. So you might want to ask some of them about this feature.
I hope this helps,