New site-developer focus is required. Steps in that direction are being made with, for example, video blob analysis of hands and movement for detecting movement that is too quick for detection by the human eye (100% multiple state classification accuracy but with a limited sample of only five interviews); detection of image manipulation through inconsistencies in compression artifacts (30%–100%, depending on type of image, compression, and tampering method); machine learning detection using audio and transcribed text to identify patterns that signal deception due to deviations from a baseline (66.4% accuracy, when baseline is at 60.2%); and computerized voice stress analysis to identify variations in an individual’s speech patterns (56.8%–92.8% accuracy, depending on context).