Does this really matter if the OS AI tools are watching what's written to your display?

Interesting point. If you're using iOS, in theory they're using similar encryption for anything they store of yours. I suppose you're having to trust them too (although them violating that trust would be a major hit to their brand, though certainly not impossible). I suppose the only solution here would be to use an audited open source OS?

That's the only one I can imagine. Otherwise this just seems like buzzword marketing. To each their own. Everyone has a different threat model.