Blockwise-adaptive attackers - Revisiting the (In)security of some provably secure encryption modes: CBC, GEM, IACBC

In this paper, we show that the natural and most common way of implementing modes of operation for cryptographic primitives often leads to insecure implementations. We illustrate this problem by attacking several modes of operation that were proved to be semantically secure against either chosen plaintext or chosen ciphertext attacks. The problem stems from the simple following fact: in the definition and proofs of semantic security, messages are considered as atomic objects that cannot be split; however, in most practical implementations, messages are subdivided into smaller chunks than can be easily manipulated. Depending on the implementation, each chunk may consist of one or several blocks of the underlying primitive. The key point here is that upon reception of a processed chunk, the attacker can now adapt his choice for the next chunk. Since the possibility of adapting within a single message is not taken into account in the current security models, this leaves room for unexpected attacks. We illustrate this new paradigm by attacking three symmetric and hybrid encryption schemes based on the chaining mode in spite of their security proofs.