Deep Learning-Based Cryptanalysis of Different AES Modes of Operation

With the advent of machine learning and the development of powerful machines, the problem of decryption takes on a new light and opens new avenues for research. The remarkable rise of technologies and algorithms contributes to the widespread use of machine learning in various cases. One of the uses is in cryptanalysis and attack of algorithms used in cryptographic processes. In this paper, we elaborate the idea by using the deep neural network to perform the known-plaintext attack on AES to restore as many bits as possible, on the given plaintext. Moreover, we perform our experiments on different key sizes and different modes of operation on AES. The results show that the deep neural network can restore the bits in the whole data set with a probability of more than 98%, restore two consecutive bytes with more than 70%, and more than half of the plaintext bytes with a probability of 99%.