There's no need for a data hoarder to rely on platform specific facial recognition - they'll presumably ingest all photos and tag/group them themselves. This of course includes images where you've been photographed in a group/public place. I imagine it's pretty trivial to "walk" from a single known-good photo (eg: booking / surveillance photo taken in connection to a "suspicious" public gathering, atm cctv, or where a person is figuring in the same photo as a known "person of interest") - to ~99.9% of all photos in the dataset...