This is exactly how FR is designed to operate (author of an FR system here). The FR system's gallery is loaded with persons of interest, and then video acquired

This is exactly how FR is designed to operate (author of an FR system here). The FR system's gallery is loaded with persons of interest, and then video acquired from various sourced is analyzed. When a person of interest is identified, that "identification" is in the form of "here is person N with a probability of X% accuracy". That "notification" is then presented to a human, who then makes the human decision "yes, that is our person or no, that is not our person" and then whatever organic response that person's presence triggers occurs. FR is like having a human lookout who is really good with faces. Beyond that, it is all humans using this capability for good or bad purposes.