Good question. I suppose that the files, if they have a title (assuming I was clever enough to do so ;-) ) are only known by their name, or file type if unlabelled, the contents unknown. The photos, in contrast are what they are and if the scanning AI is able to recognise what elements are in each image, there's no privacy to my mind. A text-type file could be labelled 'Birthday Party' but have 'macro-economic data 1975-1980', within. An image file is what it is, assuming that there's no hidden meaning involved, like some Dali surrealist painting!
Maybe Apple scan the contents of all files, who knows, but I'm assuming that they don't.