A Distinguishing Attack with a Neural Network

This paper proposes a new distinguishing-type attack to identify block ciphers. This attack utilises a multidisciplinary approach to the problem. It is grounded in a neural network, by means of a linguistic and an information retrieval approach, from patterns found on a set of cipher texts. This result is possible due to the existence of intrinsic properties in the mathematical basis of ciphers, which create signatures in the cipher texts. Experiments were performed on a set of cipher texts, which were encrypted by the finalist algorithms of AES contest: MARS, RC6, Rijndael, Serpent and Two fish, with a unique 128-bit key. The processes of clustering and classification were successful, allowing the formation of well-defined groups, where cipher texts encrypted by the same algorithm stayed close to each other, from a topological standpoint, which allow the identification of the cipher.