This paper presents a pilot study exploring lip movement collected from WiFi channel state information as a behavioral biometric identifier. We collected data for eight phrases of varying lengths from four volunteers, and, following pre-processing, extracted seven time-domain statistical features to train a SVM model per phrase length. We achieved up to 94.59% accuracy, with demonstrated advantages using phrase lengths of three to four words and the use of four PCA functions applied on each of the four 20 MHz sub-channels for dimensionality reduction prior to feature extraction.