← Catalog
Research high discovered 2024-10-01

Live Deepfake Injection to Evade Mobile KYC Verification

A red team defeated a mobile facial-authentication / KYC service by injecting a live, face-swapped deepfake video into the camera feed — impersonating a target and passing liveness checks to open fraudulent accounts.

What happened

Catalogued by MITRE ATLAS as AML.CS0033, the iProov Red Team defeated a mobile facial biometric / KYC service by injecting a live, face-swapped deepfake into the camera feed. Rather than presenting a photo to the lens (a presentation attack), the technique injects synthesized video directly into the capture pipeline, letting the attacker impersonate a target identity and satisfy the system’s liveness checks — enough to pass identity verification and open fraudulent accounts.

It targets exactly the AI meant to stop fraud: the face-matching and liveness models. As generative face tools have gotten cheap and real-time, injection attacks against onboarding and KYC have become a practical threat to any app relying on selfie verification.

Why it maps to ATLAS

  • AML.T0088 — Generate Deepfakes: a synthetic face video of the target is produced.
  • AML.T0016.002 — Obtain Capabilities: Generative AI: off-the-shelf generative tools create the fake.
  • AML.T0073 — Impersonation: the deepfake impersonates a specific identity.
  • AML.T0015 — Evade AI Model: the liveness/face-matching model is fooled.
  • AML.T0048.000 — External Harms: Financial Harm: the outcome is account fraud.

Detection notes

Presentation-attack detection alone doesn’t stop injection attacks. Harden the capture path: device attestation, secure/verified camera pipelines, and virtual-camera detection, plus liveness challenges tuned to surface face-swap artifacts. For high-stakes KYC, use risk-based step-up — secondary factors or human review — rather than trusting a single selfie.

Detection

Log sources

  • Camera pipeline integrity / virtual-camera detection
  • Presentation-attack-detection (PAD) + liveness telemetry
  • Device attestation + injection-detection signals

Signals

  • Virtual-camera or injected video sources feeding the verification app
  • Face-swap / synthesis artifacts under liveness challenges
  • Onboarding attempts inconsistent with device/environment attestation

Mitigations

  • Inject-resistant capture: device attestation, secure camera pipeline
  • Advanced liveness + deepfake-artifact detection challenges
  • Risk-based step-up (human review, secondary factors) for KYC