Microsoft’s Azure AI Vision Face API is used to power the face detection and recognition. The software can also conduct a “liveness” check, which helps prevent the use of a static photo or 2D video to trick the verification system, Microsoft said, so deepfakes shouldn’t be effective.
Customer organizations can choose the level of confidence required to accept a Face Check login attempt. The higher the confidence score threshold, the less likely Face Check will incorrectly verify an impersonator. The default score is a 50% match, which equates to a one in 100,000 chance of getting a false positive; at 90%, the chances are one in a billion, Microsoft said. (A higher confidence score requirement also increases the likelihood a legitimate login attempt will be rejected.)
Changes in a user’s appearance compared to the verified photo — a different haircut, for example –—could lower the match score, as well as differences in surroundings, such as lighting.