NDSS 2025 – EMIRIS: Eavesdropping On Iris Information Via Electromagnetic Side Channel

This research presented at the NDSS 2025 conference explores electromagnetic (EM) side channel attacks on iris recognition systems. The study focuses on the potential vulnerabilities of Near-Infrared (NIR) sensors used in iris scanning. The authors found that EM emissions from these sensors reveal information about the user's iris data. They developed "EMIRIS," a method to reconstruct iris images from these EM signals. This method involves analyzing digital signal transmission and mapping iris data to EM signals. A diffusion model is used to enhance the quality of the reconstructed iris images, addressing denoising and restoration. Experiments demonstrated that EMIRIS effectively reconstructs iris information from commercial iris recognition devices. The reconstructed irises achieve an average SSIM of 0.511 and FID of 7.25. Worryingly, these reconstructed irises can successfully spoof iris recognition models. The average success rate of these attacks is 53.47% across 3,000 samples from 50 users. The research highlights a significant security risk for iris-based authentication systems.