Deep learning in cryptanalysis a comprehensive review of techniques, applications, challenges, and future trajectories

Abstract

The integration of deep learning (DL) into cryptanalysis represents a paradigm shift, challenging traditional mathematical approaches and unlocking novel attack vectors against cryptographic algorithms and implementations. This comprehensive review synthesizes findings from recent pivotal publications to provide an in-depth analysis of the current state-of-the-art, methodologies, empirical successes, fundamental limitations, and future potential of DL in cryptanalysis. We focus extensively on two primary domains DL-based side-channel analysis (DL-SCA) and DL-enhanced cryptanalysis of symmetric primitives. The review meticulously examines advancements in attack efficiency (reducing the number of traces/queries), robustness against sophisticated countermeasures, automated feature extraction, and the nascent exploration of theoretical foundations. While DL demonstrates remarkable capabilities in automating complex pattern recognition critical to cryptanalysis, significant challenges persist, including the “black-box” nature of models, data dependency, scalability to full cryptographic primitives, and the critical need for explainability and theoretical grounding. This review serves as a foundational resource for researchers and practitioners navigating this rapidly evolving intersection of artificial intelligence and cryptography.