No right to remain silent: Isolating Malicious Mixes

Mix networks are a key technology to provide network anonymity, used for messaging, voting and private lookups. However, simple mix networks are insecure against malicious mixes, which can drop or delay packets to facilitate traffic analysis attacks. Mix networks with provable robustness address this by using complex and expensive proofs of correct shuffling, which come with a cost and make assumptions that are unnatural to many settings in which mix networks are deployed. We present Miranda, a synchronous mix network mechanism, which is provably secure against malicious mixes – yet retaining the simplicity, efficiency and practicality of mix network designs. Miranda uses first-hand experience of unreliability by mixes and clients, to derive a mix ‘reputation’, and to ensure that each active attack – including dropping of packets – results in reduction in the connectivity of the malicious mixes, thus reducing their ability to attack. Besides the practical importance of Miranda itself, our results are applicable to other mix networks designs and anonymous communication, and even unrelated settings in which reputation could provide effective defense against malicious participants.