Liveness detection is a set of techniques that verify a biometric sample — a face, a voice — comes from a live, physically present human at the moment of capture, rather than from a photo, a recording, a replay, or an AI-generated deepfake.

Why matching isn't enough

Biometric systems answer "is this the right face/voice?" Liveness answers the prior question: "is this a real face/voice, right now?" Without it, a printed photo, a screen replay, or a cloned voice can pass a matching check perfectly — because it genuinely does match.

Active vs passive liveness

Active livenessPassive liveness
MethodChallenge–response: blink, turn your head, read a phraseAnalyzes the sample itself: texture, depth, micro-movement, spectral properties
User frictionHigher — requires cooperationNone — invisible to the user
Replay resistanceStrong (response must match a fresh challenge)Good and improving
Typical useHigh-risk onboarding, step-up checksContinuous or high-volume verification

Voice liveness: the neglected channel

Almost all liveness investment has gone into faces, driven by KYC onboarding. But the highest-value fraud in 2026 happens on calls: a cloned executive voice requesting a wire. Voice liveness applies the same two families of technique — passive acoustic analysis of the live audio, and active challenges (unexpected questions, mid-call verification phrases) that pre-generated or streaming clones handle poorly.

The practical ceiling is the same as any detection layer: it estimates, it doesn't guarantee. That's why voice liveness belongs in front of a deterministic control — callback verification — rather than replacing it. Score the call; verify the payment.

Protocol tipIf your identity stack has face liveness for onboarding but nothing on the voice channel, your controls verify strangers more rigorously than the "CFO" asking for $40,000.