A Tool for Estimating Information Leakage

Abstract

Abstract. We present leakiEst, a tool that estimates how much information leaks from systems. To use leakiEst, an analyst must run a system with a range of secret values and record the outputs that may be exposed to an attacker. Our tool then estimates the amount of information leaked from the secret values to the observable outputs of the system. Importantly, our tool calculates the confidence intervals for these estimates, and tests whether they represent real evidence of an information leak in the system. leakiEst is freely available and has been used to verify the security of a range of real-world systems, including e-passports and Tor.Introduction. Information leakage occurs when something about a system's secret data can be deduced from observing its public outputs. Not all information leakage is serious: many retailers' billing systems readily "leak" the last four digits of a credit card number, and password-checking functions "leak" some information about a secret password in response to an incorrect guess (e.g., that the guess is not the password). Information leakage is therefore quantitative and it is important to be able to answer the question "how much information does a system leak?". Information theory is a useful framework for quantifying these leaks in systems (see e.g. [9]), and two particular measures, mutual information and min-entropy leakage, place useful bounds on an attacker's ability to guess the secrets from the public outputs.Our tool, leakiEst, estimates these leakage measures from datasets containing secrets and public outputs that are generated from trial runs of a system. Its methodology is based on our previous work that provides rigorous verification methods for estimating information leakage [3,4]; it performs statistical tests to distinguish an insecure system with a very small information leak from a secure one with no leaks. This is similar to detecting a correlation between two random variables, a well-investigated problem, and we compare leakiEst's performance to that of existing statistical tests. If a leak is found, leakiEst can display the conditional probability of observing each output from the system given a particular secret, which may be used to derive a concrete attack against the system.There are several tools that calculate the amount of information that leaks from a program (e.g., [7,2]). These tools provide tight bounds, but require access to the source code of the program and a formalism that is powerful enough to model the underlying system. These requirements are often prohibitive, and

Hybrid statistical estimation of mutual information and its application to information flow

“…As remarked above, the stochastic simulation techniques implemented in HyLeak have also been developed in the tools LeakiEst [CKN13] (with its extension [KCP14]) and LeakWatch [CKNP13,CKN14]. The performance of these tools as compared to HyLeak is represented by the "statistical" analysis approach in Section 8.…”

Hybrid Statistical Estimation of Mutual Information for Quantifying Information Flow

“…Chothia et al have developed this approach in tools leakiEst [3,39] and LeakWatch [5,29]. The hybrid statistical method in this paper can be considered as their extension with the inclusion of component weighting and adaptive priors inspired by the importance sampling in statistical model checking [40,41].…”

“…For example, statistical methods [2,3,4,5,6] have been studied for quantitative information flow analysis [7,8,9], which estimates an entropy-based property to quantify the leakage of confidential information in a system. More specifically, the analysis estimates mutual information or other properties between two random variables on the secrets and on the observable outputs in the system to measure the amount of information that is inferable about the secret by observing the output.…”

LeakWatch: Estimating Information Leakage from Java Programs

“…Execution and sandboxing of the target program are achieved using the core Java libraries, and leakage estimates are calculated by leakiEst [19,20], our information leakage estimation library for Java, which now implements our min-entropy leakage result from Section 3; this ensures that LeakWatch is not tethered to a particular version or implementation of the JVM specification.…”