Sure. I'm not entirely sure how PCIE works these days. But in it good old days we had methods to read pretty much arbitrary memory regions via PCIE or early Thunderbolt(?).

I just figured it'd be massively complicated to wait for the user to pull something on the screen, do computationally expensive OCR, some AI image detection to puzzle documents back together, and then you'd only get a fraction of what's really stored on the computer and you'd still need a way to send that information home... When you could just pick a plethora of easy options like read all the files from the harddisk and send just them somewhere. I think it's far more likely they do some easy and straightforward solution. And it'd be more effective as well.