What is presentation attack detection, and why does it matter for exam proctoring?

TL;DR: Presentation attack detection (PAD) is the standardized way biometric systems test whether a face or identity check can be fooled by a photo, video replay, mask, or deepfake — governed by ISO/IEC 30107. For exam and interview proctoring, PAD is what separates "we checked a selfie against an ID" from "we verified a live human is actually present," and it's the technical backbone behind deepfake and virtual-camera detection claims.

The claim

Any identity-verification system that only compares a photo to an ID is vulnerable to being shown a printed photo, a replayed video, or a deepfake instead of a live person — PAD is the discipline of testing and rating how well a system resists exactly that.

The evidence

  • ISO/IEC 30107-1 defines the core framework for biometric presentation attack detection; ISO/IEC 30107-3 defines the testing and reporting methodology used to measure PAD performance.
  • Independent test labs like iBeta are NIST-accredited to conduct PAD conformance testing, issuing Level 1 and Level 2 confirmation letters that vendors can point to as third-party proof of anti-spoofing performance.
  • The key PAD metric, Attack Presentation Classification Error Rate (commonly summarized via the Attack Presentation Acceptance Rate, or APAR), quantifies how often a spoofed presentation is incorrectly accepted as genuine — a lower rate means stronger security.
  • Industry explainers (Didit, Innovatrics, Biometrics Institute) consistently describe deepfakes, printed photos, video replays, and 3D masks as the attack categories PAD is specifically designed to catch — the exact threat model exam and interview proctoring cares about.
  • Neuroxa's identity layer — ID + selfie match, continuous face verification, and deepfake/virtual-camera flags — is functionally a PAD implementation applied to the exam and interview use case, even where the term itself doesn't appear in marketing copy.

Comparison: identity checks without PAD vs. with PAD

ApproachPhoto-vs-ID comparison onlyWith presentation attack detection
Stops a printed photo held up to cameraNoYes
Stops a pre-recorded video replayNoOften, with liveness checks
Stops a real-time deepfake face swapNoPartially — this is the hardest attack category
Independently testable/certifiableNot standardizedYes — ISO/IEC 30107-3, iBeta conformance letters
Confidence level for a defensible trust reportLowHigh — testable claims, not just marketing language

Step-by-step: evaluating a proctoring vendor's PAD claims

  1. Ask directly whether the vendor's liveness/deepfake detection has been independently tested against ISO/IEC 30107-3, not just internally validated.
  2. Request the specific PAD conformance level (e.g., iBeta Level 1 or Level 2) if independent testing exists.
  3. Ask which attack categories are covered — printed photo, video replay, 3D mask, and real-time deepfake are each a different technical challenge.
  4. Confirm the PAD check runs continuously through the session, not just at check-in — a one-time liveness check doesn't stop a mid-session swap.
  5. Make sure PAD results feed into your trust report as evidence, not just a pass/fail gate at login.

FAQ

Is PAD the same thing as facial recognition? No — facial recognition asks "whose face is this?" PAD asks "is this a live human face at all, or a spoof?" They're complementary, not the same check.

Does Neuroxa have independent PAD certification? Neuroxa's identity layer is built around continuous face verification and deepfake/virtual-camera detection functionally aligned with PAD principles; ask your account team for current third-party testing status specific to your deployment.

What's the hardest presentation attack to detect? Real-time deepfakes — face-swap software that generates a convincing live video feed — are widely considered the toughest attack category, harder than static photos or simple video replays.

Why does a lower APAR matter in practice? A lower Attack Presentation Acceptance Rate means fewer spoofed sessions get incorrectly accepted as genuine — directly reducing the risk of a fraudulently obtained certification or a proxy interview going undetected.

Should every proctoring vendor be required to publish PAD test results? It's a reasonable ask for high-stakes use cases (professional licensing, hiring for sensitive roles) — third-party PAD conformance is a stronger claim than unverified marketing language about "deepfake detection."

By Pinal Dave Last updated: 2026-07-31