They already have the faces annotated via the tagging system. To be really sneaky they could inject faces which have been detected but NOT tagged. However they probably have enough data anyway - enough being the largest facial recognition training data set in the world.