Dopple-ganging up on Facial Recognition Systems

This McAfee Advanced Threat Research report demonstrates a novel adversarial machine-learning attack that uses CycleGANs combined with FaceNet embeddings to generate photorealistic, morphed passport-style images that can be misclassified by facial-recognition systems. The authors describe data collection, model training (white-box and gray-box), transferability concepts, and demo scenarios showing how an attacker could bypass automated passport verification; they highlight the security implications for biometric authentication and call for standards and defense-in-depth.