Scalable Secure Multi-party Network Vulnerability Analysis via Symbolic Optimization

Threat propagation analysis is a valuable tool in improving the cyber resilience of enterprise networks. As these networks are interconnected and threats can propagate not only within but also across networks, a holistic view of the entire network can reveal threat propagation trajectories unobservable from within a single enterprise. However, companies are reluctant to share internal vulnerability measurement data as it is highly sensitive and (if leaked) possibly damaging. Secure Multi-Party Computation (MPC) addresses this concern. MPC is a cryptographic technique that allows distrusting parties to compute analytics over their joint data while protecting its confidentiality. In this work we apply MPC to threat propagation analysis on large, federated networks. To address the prohibitively high performance cost of general-purpose MPC we develop two novel applications of optimizations that can be leveraged to execute many relevant graph algorithms under MPC more efficiently: (1) dividing the computation into separate stages such that the first stage is executed privately by each party without MPC and the second stage is an MPC computation dealing with a much smaller shared network, and (2) optimizing the second stage by treating the execution of the analysis algorithm as a symbolic expression that can be optimized to reduce the number of costly operations and subsequently executed under MPC. We evaluate the scalability of this technique by analyzing the potential for threat propagation on examples of network graphs and propose several directions along which this work can be expanded.