Also data retention and security it's a nightmare for Title IX and FERPA as well.

Another thing is Microsoft hasn't been talking about compression either, how large are these files? What does it do with networked drives? How do we know metadata collection isn't being expanded?

Hear me out, I actually had a similar concept in mind, but only for files, emails, calendar entries, bookmarks, that kind of stuff. Things that I actually saved on my computer, not random screenshots of what I'm looking at. This is a huge difference IMO. What I look at should never be saved. Only when I specifically save something, should it persist. I would actually love a FOSS, local and private AI solution that would allow me to simply query anything I've ever saved on my computer with a simple search request, without having to waste time on naming my files. Even better if it would understand the context and stuff. This would especially be useful with photos, as they never have proper filenames, just some generic random stuff. Or with code, if the AI search could understand the context of my code and I could just pull it up using a search terms like "the function for handling DNS over TLS requests a few years ago" or whatever, and it would just pull out that one function from the project. Even better if this could be integrated with a separate, generative AI model, that could make small changes to my already existing stuff. I don't know, e.g. "refactor the function to use LibreSSL instead of OpenSSL TLS library".

Shadow copy is a completely different thing. Shadow copy creates snapshots(used for version history, among other uses) of files. Recall is a screen recording software, that includes OCR and maybe some AI stuff. At this time, at least, it too is all local. It just isn't secure in the least.