Breaking the collusion detection mechanism of MorphMix

MorphMix is a peer-to-peer circuit-based mix network designed to provide low-latency anonymous communication. MorphMix nodes incrementally construct anonymous communication tunnels based on recommendations from other nodes in the system; this P2P approach allows it to scale to millions of users. However, by allowing unknown peers to aid in tunnel construction, MorphMix is vulnerable to colluding attackers that only offer other attacking nodes in their recommendations. To avoid building corrupt tunnels, MorphMix employs a collusion detection mechanism to identify this type of misbehavior. In this paper, we challenge the assumptions of the collusion detection mechanism and demonstrate that colluding adversaries can compromise a significant fraction of all anonymous tunnels, and in some cases, a majority of all tunnels built. Our results suggest that mechanisms based solely on a node's local knowledge of the network are not sufficient to solve the difficult problem of detecting colluding adversarial behavior in a P2P system and that more sophisticated schemes may be needed.