Post

Internet censorship resistance systems (CRSs) have so far been designed in an ad-hoc manner. The fundamentals are unclear and the foundations are shaky. Censors are, more and more, able to take advantage of this situation. Future censorship resistance systems ought to be built from strong theoretical underpinnings and be based on empirical evidence. Our approach is based on systematizing the CRS field and its players. Informed by this systematization we develop frameworks that have broad scope, from which we gain general insight as well as answers to specific questions. We develop theoretical and simulation-based analysis tools 1) for learning how to manipulate censor behavior using game-theoretic tactics, 2) for learning about CRS-client activity levels on CRS networks, and finally 3) for evaluating security parameters in CRS designs. We learn that there are gaps in the CRS designer's arsenal: certain censor attacks go unmitigated and the dynamics of the censorship arms race are not modeled. Our game-theoretic analysis highlights how managing the base rate of CRS traffic can cause stable equilibriums where the censor allows some amount of CRS communication to occur. We design and deploy a privacy-preserving data gathering tool, and use it to collect statistics to help answer questions about the prevalence of CRS-related traffic in actual CRS communication networks. Finally, our security evaluation of a popular CRS exposes suboptimal settings, which have since been optimized according to our recommendations. All of these contributions help support the thesis that more formal and empirically driven CRS designs can have better outcomes than the current state of the art.