It's not quite the same... maybe even reversed. Big companies can "secretly" use private datasets with copyrighted or legally questionable data with a wink and

It's not quite the same... maybe even reversed. Big companies can "secretly" use private datasets with copyrighted or legally questionable data with a wink and a nod, without anyone being able to prove it in court. Open source models generally do not have that option. Maybe there are some exceptions like Meta using some private data in future Llamas, but I question how sustainable those efforts are.