Analysis of McEliece Public Cryptography Using Deep AutoEncoder

McEliece code is the first random public cryptosystem. For decades, McEliece public scheme (constructed by Goppa codes) is assumed secure against the known public cryptosystem attacks. Indistinguishability is a basic character of a public key encryption scheme. In this paper, we are aimed at proving that the classic McEliece cryptosystem is not secure against indistinguishable chosen ciphertext attack from the angle of practical discriminator in action. An online learning framework (AEMC+DISC) is designed to find ciphertext discriminative flaws automatically. In the framework, an encoding generator (AEMC) is constructed using deep autoencoder to generate legal pairs of plaintext and ciphertext without the participation of McEliece ciphers, and an vector classifier (DISC) is built to distinguish identical ciphertexts. We prove that McEliece scheme based on Goppa code (the classic one and its variants) is vulnerable under any kinds of indiscriminative attacks.