Composing protocols with randomized actions

Abstract

Recently, several composition results have been established, showing that two cryptographic protocols proven secure against a Dolev-Yao attacker continue to afford the same security guarantees when composed together, provided the protocol messages are tagged with the information of which protocol they belong to. The key technical tool used to establish this guarantee is a separation result which shows that any attack on the composition can be mapped to an attack on one of the composed protocols running in isolation. We consider the composition of protocols which, in addition to using cryptographic primitives, also employ randomization within the protocol to achieve their goals. We show that if the protocols never reveal a secret with a probability greater than a given threshold, then neither does their composition, given that protocol messages are tagged with the information of which protocol they belong to.