Proper clinical design
I think that negative controls are always needed, unless there is a preliminary evidence or a well sounded hypothesis of a clear advantage of the new treatment over the gold standard positive treatment. In that case it could be reasonable to plan a clinical trial with only positive controls, taking in mind that if the results tell us that there is not a clear difference between treatments and the efficacy is not high, we can't rule out the assay sensitivity problem.
Without both, how would you be able to confirm that the model itself has sufficient ability to differentiate? In other words, if all we see is something that is better than nothing, is that sufficient to provide perspective, or do you really require both in order to be able to accurately assess overall performance?