Liveness Detection: Stop Deepfakes
Liveness detection asks a simple question: is there a real person present right now, or is the system seeing a replay, mask, face swap, or generated video? As deepfakes improve, liveness detection has become a core defense in identity verification.
What Liveness Detection Checks
Liveness detection looks for signals that are hard to fake in real time: natural blinking, head movement, depth, lighting response, skin texture, device motion, and challenge responses. In identity verification, these checks help separate a live applicant from a prerecorded or AI-generated presentation attack.
Passive vs Active Liveness
Passive liveness works in the background without asking the user to perform a task. Active liveness asks for an action, such as turning the head, reading numbers, or blinking on command. Passive checks are smoother for users, while active checks can raise the cost of an attack. Many verification flows combine both.
How Deepfakes Try to Bypass Liveness
Attackers may replay a recorded video, use a face-swap filter during a live call, hold a screen in front of the camera, or generate a talking-head video that follows prompts. Strong liveness systems check timing, texture, depth, and interaction patterns instead of relying on one facial cue.
Where AI Video Detection Helps
AI Video Detector is not a replacement for a full identity verification platform, but it can support manual review. Enterprise platforms like Sensity integrate directly into identity verification pipelines. If a submitted onboarding video, appeal clip, or suspicious support recording looks manipulated, scanning it can surface evidence frames, confidence labels, and reason codes for the risk team.
Review Signals to Document
Document whether the face boundary shifts, whether eyes and mouth move naturally, whether lighting changes match the scene, and whether audio matches lip movement. Save timestamps and screenshots. A clear evidence trail matters when an account is approved, rejected, or escalated.
Practical Risk Controls
Use device checks, rate limits, replay detection, human review for high-risk cases, and clear user consent. When provenance metadata is available, Content Credentials can verify the source before the video reaches the liveness check. Treat liveness detection as one layer in a broader identity verification system, not as a single magic test that stops every deepfake.
Sources
How This Article Was Created
This article was written by the AI Video Detector product team based on publicly documented identity verification workflows and internal testing of detection signals.