Impersonation is what several of the 20 integrity signals exist for. There are four which are relevant and work on different parts of the interview.
Multiple faces. Computer vision runs on the video feed and counts the faces in frame. More than one is flagged - this is the signal that catches a second person sitting beside the candidate, feeding them answers.
Face out of view. The same vision layer tracks whether the candidate's own face stays in frame. A face that repeatedly drops out of view is flagged, because a candidate who is out of frame while answering cannot be matched to the voice on the audio.
Multiple voices. Working on the audio rather than the video, this flags when more than one voice is detected in the room. Any off-camera prompting will immediately be registered.
Lip-sync verification. This one connects the two. It checks whether the words on the audio track line up with the candidate's mouth movements. When they do not then it indicates that someone else is speaking off-camera, or audio is being laid over the candidate's video - it is flagged for you.
Alongside these, plagiarism and AI-answer detection flags answers that look copied or machine-generated, which covers the scripted-answer version of the same problem. Eye-gaze patterns sit in the same group and are worth reading next to the four above: a gaze that repeatedly settles on a fixed point off-screen is seen differently from a candidate looking away while they think.
Every one of these is just a signal, not a verdict. They appear on the report with the recording and the transcript beside them, so you can go to the exact moment and judge it yourself. No detection layer replaces watching the tape on a candidate you have real doubts about - the signals exist to tell you where to look.
