In test of six sets of identical twins, none fool Windows Hello face recognition feature
Windows Hello: can identical twins fool Microsoft and Intel? — CAN TWINS TRICK FACIAL RECOGNITION? — Windows 10 paired with Intel's RealSense camera gives users a new way to log in to their computers- facial recognition.
Context & Ripple Effects
When Windows 10 shipped with Intel's RealSense camera, the obvious stress test was the hardest one in biology: six sets of identical twins, none of whom could trick the Windows Hello login. The pass mattered because rival approaches were about to look fragile by comparison.
The surrounding coverage frames this as round one of an escalating spoofing contest: researchers later fooled many AI systems with printed paper eyeglass frames, Samsung's Galaxy S8 face unlock fell to a plain photograph, and by 2019 attackers were using 3D masks against Alipay, WeChat Pay, and airport gates.
First-order effects
- Microsoft and Intel get third-party validation that RealSense's depth-based capture resists the closest possible lookalikes, strengthening Hello's pitch as a password replacement at Windows 10 launch.
- Samsung's photo-vulnerable face unlock stands in contrast: the twin result hands depth-sensing vendors a security talking point over RGB-camera-only implementations.
Second-order effects
- Device makers shipping face recognition are pushed toward depth cameras or layered biometrics — Samsung itself steered S8 users toward fingerprint and iris scanning after its photo incident.
- Payment and transit operators relying on facial verification, later hit by 3D mask attacks on Alipay and WeChat Pay, have grounds to demand hardware-level anti-spoofing rather than software-only checks.
Third-order effects
- If the attack curve holds — photos, then glasses, then masks, then nine computer-generated "master key" faces impersonating nearly half a benchmark set — biometric authentication becomes an arms race where liveness detection is table stakes and no single modality is trusted alone.
The trend: Consumer facial recognition is splitting into two tiers — depth-hardware systems that survive physical impersonation attempts and cheap camera implementations that keep falling to ever-cheaper spoofs.