Practical Traffic Analysis: Extending and Resisting Statistical Disclosure

Abstract

Abstract. We extend earlier research on mounting and resisting passive long-term end-to-end traffic analysis attacks against anonymous message systems, by describing how an eavesdropper can learn sender-receiver connections even when the substrate is a network of pool mixes, the attacker is non-global, and senders have complex behavior including generating padding messages. Additionally, we describe how an attacker can use extra information about message distinguishability to speed the attack. Finally, we simulate our attacks for a variety of scenarios, focusing on the amount of information needed to link senders to their recipients. In each scenario, we show that the intersection attack can still succeed, albeit more slowly-in some cases, so slowly as to be impractical.

Understanding Statistical Disclosure: A Least Squares Approach

“…Danezis considers that in each round where Alice sends a message, the recipient anonymity set of this message is uniform over the receivers present in the round (and zero for the rest of users). The SDA was subsequently extended to more complex mixing algorithms [7], to traffic containing replies [5], to consider other users in order to improve the identification of Alice's contacts [14], and to evaluate more complex user models [15].…”

“…We note that previous authors evaluated the attacks either from mostly a de-anonymization of individual messages perspective (e.g., [8,21]), or from the point of view of the number of rounds necessary to identify a percentage of Alice's recipients (e.g., [14,15]). In this work we are interested in the accuracy with which the adversary can infer the sender (respectively, receiver) profile of Alice, i.e., we not only seek to identify Alice's messages receivers, but also to estimate the probability that Alice sends (or receives) a message to (from) them.…”

“…Statistical variants [3,7,21] rely on heuristics to operate, while the Bayesian inference-based method by Danezis and Troncoso [8] use Bayesian sampling techniques to recover accurately users' profiles in more complex systems than its predecessors. For any of these attacks it is difficult to obtain analytic results that characterize the dependence of their effectiveness on the parameters of the system, and hence they (as well as their sequels [5,14,15,18]) have been mostly evaluated through simulation.…”

Data Retention and Anonymity Services

“…Note that previous work in this field (cf. [11,12]) has focused on calculating anonymity sets resulting from intersection attacks using probabilistic algorithms, whereas our contribution is to model linkability in a probabilistic sense.…”

Tor HTTP Usage and Information Leakage

“…Another analysis has been conducted by Dan Egerstad in 2007 [6] who published a list of 100 sensitive email accounts including passwords from embassies that apparently used Tor incorrectly. Other published attacks on Tor aimed at decreasing or defeating the users anonymity by means of traffic analysis [7,8,9,10] as well as attacks on unique aspects such as path selection [11] or the "hidden services" of the Tor network [12,13].…”