Fault Attacks on AES with Faulty Ciphertexts Only

Classical Fault Attacks often require the ability to encrypt twice the same plaintext, in order to get one or several pairs of correct and faulty cipher texts corresponding to the same message. This observation led some designers to think that a randomized mode of operation may be sufficient to protect block cipher encryption against this kind of threat. In this paper, we consider the case where the adversary neither chooses nor knows the input messages, and has only access to the faulty cipher texts. In this context, we are able to describe several attacks against AES-128 by using non uniform fault models. Our attacks target the last 4 rounds and allow to recover the correct key with practical time complexity, using a limited number of faulty cipher texts. This work highlights the need for dedicated fault attack countermeasures in secure embedded systems.