Automatic detection of weak cipher usage in aircraft communications

The Aircraft Communications Addressing and Reporting System (ACARS) allows aircraft to communicate with entities on the ground via short messages. To provide confidentiality for sensitive information communicated via the ACARS network, some operators started to deploy proprietary cryptography to encrypt message contents. This is highly problematic, however, as all of the observed approaches in practice offer next to no communication security but give a false sense of it. Authorities hence would like to filter out weak ciphers at the network level to alert operators of their risks. This project explored the use of deep convolutional neural networks (CNN) for automatic detection of weakly encrypted message contents on the ACARS data link network. We constructed a labelled dataset of plaintext and ciphertext messages and experimentally evaluated the performance of a deep CNN for message classification. We find that a model trained on a sufficiently diverse set of ciphertexts is able to detect weakly encrypted messages with high accuracy.