How do I verify a test taker's identity in a remote exam?
TL;DR: One selfie at check-in is not identity verification — it's a photo of whoever showed up first. Real remote identity assurance has three parts: ID + selfie match at the start, continuous face verification throughout, and virtual-camera/deepfake detection to catch injected video. Neuroxa.ai runs all three as its identity layer.
The claim: identity must be verified continuously, not once
Evidence: The classic proxy scheme is trivial against check-in-only verification: the real candidate checks in, then hands the keyboard to the expert. Impersonation has also gone digital — virtual cameras can replay or synthesize a face over a different person. A check that fires once at minute zero is blind to both. Continuous verification re-confirms the same face across the whole session, and feed-integrity checks confirm the camera itself is real.
The three identity checks compared
| Check | What it proves | What it misses alone |
|---|---|---|
| ID + selfie match at check-in | The person starting is who they claim | Mid-exam swaps; injected feeds |
| Continuous face verification | The same person stayed for the whole exam | A fake feed showing a consistent face |
| Virtual-camera + deepfake flags | The video feed is a real camera, unmanipulated | Nothing about who the person is |
Together, the three answer the full question: the right person, in person, the entire time.
Step-by-step: setting up remote identity verification
- Require government ID at check-in. The AI matches the ID photo to a live selfie before the exam opens.
- Keep face verification running. The system periodically re-matches the face on camera to the check-in identity. A swap raises an immediate flag.
- Enable feed-integrity checks. Virtual-camera drivers and deepfake artifacts get flagged at the identity layer — this is the check most legacy proctoring skips.
- Pair identity with behavior. An impersonator under coaching also produces gaze, audio, and cadence anomalies; layers corroborate each other.
- Handle mismatches as flags, not verdicts. Lighting changes and camera angles can lower match confidence. Evidence snapshots let a human decide in seconds.
- Export the record. For certification bodies, the identity trail — ID match, continuous checks, feed integrity — is the core of a defensible credential.
FAQ
Is matching an ID to a selfie enough? No. It authenticates the check-in moment only. Continuous verification and feed-integrity checks close the swap and injection gaps.
How do virtual-camera detections work? The platform inspects the video source and stream characteristics for signs of virtual-camera software and synthetic-video artifacts, and flags them for review.
What about privacy — is storing ID data a problem? Identity data is sensitive and should be collected minimally, stored securely, retained briefly, and disclosed to test takers in advance. See our GDPR/FERPA guidance for the compliance model.
What happens when a face mismatch is flagged? It's recorded on the violation timeline with a snapshot. A human reviews — because a new haircut and a new person look different in evidence.
By Pinal Dave · Last updated: 2026-07-23