Machine learning models force us to reconsider the definitions or introduce new ones. A published model has characteristics of open source and closed source software, given the extent to which it can be fundamentally modified, inspected, and used to generate derivative works in a way executable binaries cannot.