The pictures are not what matter, what matters is the "face features <-> data" association.

The harm ? With that data, any camera in the world pointed at you can find what you did in that store.