FOCI '19 Workshop Program

The prevalence of Internet censorship has prompted the creation of several measurement platforms for monitoring filtering activities. An important challenge faced by these platforms revolves around the trade-off between depth of measurement and breadth of coverage. In this paper, we present an opportunistic censorship measurement infrastructure built on top of a network of distributed VPN servers run by volunteers, which we used to measure the extent to which the I2P anonymity network is blocked around the world. This infrastructure provides us with not only numerous and geographically diverse vantage points, but also the ability to conduct in-depth measurements across all levels of the network stack. Using this infrastructure, we measured at a global scale the availability of four different I2P services: the official homepage, its mirror site, reseed servers, and active relays in the network. Within a period of one month, we conducted a total of 54K measurements from 1.7K network locations in 164 countries. With different techniques for detecting domain name blocking, network packet injection, and block pages, we discovered I2P censorship in five countries: China, Iran, Oman, Qatar, and Kuwait. Finally, we conclude by discussing potential approaches to circumvent censorship on I2P.

With the increasing use of TLS encryption over web traffic, censors start to deploy SNI filtering for more effective censorship. Specifically, a censor can identify the web domain being accessed by a client via the SNI extension in the TLS ClientHello message. In response, in August 2018, a new extension called ESNI (Encrypted-SNI) is proposed for TLS 1.3, aiming at fixing this server name leakage.

In this paper, we first characterize SNI-based censorship in China by measuring its prevalence and effectiveness. We outline its assisting role in censorship by comparing it with other commonly used censorship methods. We then measure the deployment prevalence of ESNI and further analyze its current and potential effectiveness in censorship circumvention. We also monitor the censorship associated with ESNI from 14 areas all around the world. Based on our analysis, we discuss the key factors to the success of ESNI and potential problems in a post-ESNI era. We hope our work will make ESNI a more promising and effective censorship circumvention strategy.

The challenge of deniability for sensitive data can be a life or death issue depending on location. Plausible deniability directly impacts groups such as democracy advocates relaying information in repressive regimes, journalists covering human rights stories in a war zone, and NGO workers hiding food shipment schedules from violent militias. All of these users would benefit from a plausibly deniable data storage system. Previous deniable storage solutions only offer pieces of an implementable solution. Artifice is the first tunable, operationally secure, self repairing, and fully deniable steganographic file system.

Artifice operates through the use of a virtual block device driver stored separately from the hidden data. It uses external entropy sources and erasure codes to deniably and reliably store data within the unallocated space of an existing file system. A set of data blocks to be hidden are combined with entropy blocks through erasure codes to produce a set of obfuscated carrier blocks that are indistinguishable from other pseudorandom blocks on the disk. A subset of these blocks may then be used to reconstruct the data. Artifice presents a truly deniable storage solution through its use of external entropy and erasure codes, while providing better durability than other deniable storage systems.

As the internet becomes increasingly crucial to distributing information, internet censorship has become more pervasive and advanced. Tor aims to circumvent censorship, but adversaries are capable of identifying and blocking access to Tor. Meek, a traffic obfuscation method, protects Tor users from censorship by hiding traffic to the Tor network inside an HTTPS connection to a permitted host. However, machine learning attacks using side-channel information against Meek pose a significant threat to its ability to obfuscate traffic. In this work, we develop a method to efficiently gather reproducible packet captures from both normal HTTPS and Meek traffic. We then aggregate statistical signatures from these packet captures. Finally, we train a generative adversarial network (GAN) to minimally modify statistical signatures in a way that hinders classification. Our GAN successfully decreases the efficacy of trained classifiers, increasing their mean false positive rate (FPR) from 0.183 to 0.834 and decreasing their mean area under the precision-recall curve (PR-AUC) from 0.990 to 0.414.

This talk will consider how Internet companies respond to rule violations by their users. Typically, the responses are viewed as binary: remove the content or keep it; terminate the user's account or not. However, there are a wide range of responses in between those binary options. The talk will identify a wide range of intermediate options and explore some design principles for choosing appropriate responses to rule violations. These design principles should be useful both to Internet companies and those who regulate them.