Neural Cryptanalysis: Metrics, Methodology, and Applications in CPS Ciphers

Abstract

Many real-world cyber-physical systems (CPS) use proprietary cipher algorithms. In this work, we describe an easy-to-use black-box security evaluation approach to measure the strength of proprietary ciphers without having to know the algorithms. We quantify the strength of a cipher by measuring how difficult it is for a neural network to mimic the cipher algorithm. We define new metrics (e.g., cipher match rate, training data complexity and training time complexity) that are computed from neural networks to quantitatively represent the cipher strength. This measurement approach allows us to directly compare the security of ciphers. Our experimental demonstration utilizes fully connected neural networks with multiple parallel binary classifiers at the output layer. The results show that when compared with round-reduced DES, the security strength of Hitag2 (a popular stream cipher used in the keyless entry of modern cars) is weaker than 3-round DES.