I would assume that they are working closer with companies that they trust to feed in information. Having that on an what will eventually be a public API is recipe for disaster.
Each instance of Watson is unique and has to be trained as such based on it's "corpus" (set of data) and actual feedback on the quality of it's answers by experts. The public API sounds like it will allow access to specific flavors of pre-trained and data-filled Watsons, like the food- recipe one or some basic medical ones.