Why is taking a screenshot or photo of a document instead of uploading the whole file considered safer for avoiding PII exposure in an LLM?

A member described taking screenshots (using a snipping tool) or even phone photos of just the relevant paragraph of a document — for example, only the interest rate, amortization term, and balloon payment info from a commercial loan — rather than uploading the entire file. Because the screenshot only contains what's visibly captured, there's no possibility of leaking anything beyond that image; sensitive fields like account numbers simply aren't included in the frame. In one case, a page containing sensitive information was physically folded over before taking the photo, so it wasn't included at all. The LLM (Claude) was reported to do an excellent job extracting text from screenshots or even grainy phone photos, including from an 80-page, decades-old typewritten commercial lease with handwritten modifications — something it reportedly could not handle six months prior. The member acknowledged this method isn't very scalable but said it provided much more confidence about what data was actually being shared.