First of all I don't think they can do face recognition(i. e. this guy is Johnny), the are probably only doing face detection(i. e. this rectangle contains a face). To recognize faces they would need a database of training data which contains several images for each face(more images -> better recognition), which I expect they don't have yet.
Commercial face recognition is no where near 1% false positive, it's more like 5-10% in extremely favorable cases. It's important to note here we're not talking about a controlled environment like an access point where a person is filmed and his face is recognized in a controlled light, face tilt, etc. Also, trying to decrease false negatives(i. e. face not associated with a name) will increase the rate of false positives and vice versa. For such a system, it's important to have a low false negative rate else it's useless.
I work for a company that does image processing, including face detection/recognition.