Cryptanalysis of Cipher Machines with the Help of Artificial Neural Networks
In this talk, we present our results on applying Artificial Neural Networks (ANNs) for cryptanalysis of rotor cipher machines. Rotor cipher machines were used until the 1970s. They are electro-mechanical stream cipher devices used for encryption and decryption of messages. In particular, we consider two different attack models: known-plaintext attack and ciphertext-only attack. In the both cases, our goal is to recover the key of a machine. To get each specific bit of the key we train a separate ANN to determine its value. Thus, each ANN solves a binary classification problem. In the case of the known-plaintext attack, the networks get as an input the keystream generated by the machine. Whereas in the ciphertext-only-attack scenario, the input of the networks is the ciphertext itself. To illustrate both attacks, we apply them against Hagelin M-209 which was used by the US Navy and the US military during the WWII and until after the Vietnam War.
Robert Landrichinger ist IT-Security Masterstudent hier in Hagenberg, IT-Systemadministrator und Trainer. Seit dem Bachelorstudium beschäftigte er sich mit künstlichen Neuronalen Netzen. Das Interesse für Kryptographie lebt er derzeit mit der Arbeit an einer Rotor-Chiffriermaschine des US-Militärs, der Hagelin M209, aus.
„Vasily Mikhalev received his PhD in symmetric cryptography at the University of Mannheim in 2019. His thesis focused on the design and analysis of cryptographic solutions suitable for the constrained environments. Since 2020, he has been employed at the University of Siegen and his primary area of research involves the application of machine learning techniques for breaking classical and modern cryptographic algorithms.“