The surge in face forgeries has increasingly undermined confidence in the authenticity of online content. As generation algorithms rapidly evolve, new fake categories will constantly emerge, severely challenging existing face forgery detection methods. Although face forgery detection has recently improved, current techniques remain largely confined to binary Real-vs-Fake classification or the recognition of known fake categories. Moreover, they fail to identify the emergence of entirely new forgery methods. In this work, we study the Open Set Face Forgery Detection (OSFFD) problem, which requires the detection model to identify novel fake categories. To enhance its real-world applicability, we reformulate the OSFFD problem and address it through uncertainty estimation. Specifically, we propose the Dual-Level Evidential face forgery Detection (DLED) approach, which estimates prediction uncertainty by extracting and integrating category-specific evidence on the spatial and frequency levels. Comprehensive experiments across diverse settings demonstrate that our proposed DLED approach achieves state-of-the-art performance. Notably, it surpasses various existing baseline models by a $20\%$ margin on average when identifying forgeries from novel fake categories. Concurrently, our DLED method yields competitive performance on the standard binary Real-versus-Fake face forgery detection task.

Machine learning (ML) suffers from a persistent and critical flaw: adversarial examples. Many new forms of adversarial example attacks have been invented and many narrow defenses have been proposed. Unfortunately, no defensive approach can withstand current attacks. We hypothesize that ML model robustness can be improved with approaches that delineate the data-point-sparse latent space between data-dense regions of a model’s classification space as a barrier class. We introduce one such defense, PadNet, that builds a barrier class using a combination of training samples that mix multiple classes together. It leverages this barrier class to separate decision boundaries between benign classes with regions of padding. PadNet then implements a gradient regularization strategy that penalizes gradients in the direction of the barrier class, causing the decision boundary to draw tighter around training samples increasing boundary thickness between classes. We evaluate PadNet against a sampling of the most effective state-of-the-art attacks, demonstrating that it offers significant robustness and reliability compared to current defenses. We also test it against adaptive attacks and find that PadNet remains robust against them.

organize analytics Figure 1: Study flow showing the two phases of user studies: the Requirements Study, followed by the Ontology Study and their sub-steps.

The proliferation of online face images has heightened privacy concerns, as adversaries can exploit facial features for nefarious purposes. While adversarial perturbations have been proposed to safeguard these images, their effectiveness remains questionable. This paper introduces IVORY, a novel adversarial purification method leveraging Diffusion Transformerbased Stable Diffusion 3 model to purify perturbed images and improve facial feature extraction. Evaluated across gender recognition, ethnicity recognition and age group classification tasks with CNNs like VGG16, SENet and MobileNetV3 and vision transformers like SwinFace, Ivory consistently restores classifier performance to near-clean levels in white-box settings, outperforming traditional defenses such as Adversarial Training, DiffPure and IMPRESS. For example, it improved gender recognition accuracy from $37.8 \%$ to $96 \%$ under the PGD attack for VGG16 and age group classification accuracy from $\mathbf{2. 1 \%}$ to $\mathbf{5 2. 4 \%}$ under AutoAttack for MobileNetV3. In black-box scenarios, IVORY achieves a $22.8 \%$ average accuracy gain. IVORY also reduces SSIM noise by over $50 \%$ at 1x resolution and up to $80 \%$ at $2 x$ resolution compared to DiffPure. Our analysis further reveals that adversarial perturbations alone do not fully protect against soft-biometric extraction, highlighting the need for comprehensive evaluation frameworks and robust defenses.

Clickbait, masked behind interesting headlines and thumbnails, is often used to spread misinformation and trick users into clicking on social media posts or links that direct them to malicious websites. To help users protect against clickbait, we examined interventions based on persuasion theories including designs that used social consequence, personal consequence, and badges. To this end, we first conducted a preliminary study to translate the participants’ feedback into improving our initial designs, followed by a lab study with 20 participants (60% Male, 40% Female; 18–44 years old) aimed at understanding their perceptions of the improved interventions; we further updated our designs based on their feedback. We then conducted an online study with 773 participants (56% Male, 42% Female; 18 to above 65 years old) over MTurk to evaluate the impact of persuasion techniques leveraged in our designs. Our findings suggest that persuasion can be an effective strategy to warn users against clickbait, specifically ones that use incentives such as revealing mystery of clickbait. Overall, our studies provide valuable insights into understanding users’ needs and expectations around interventions against clickbait, and offer guidelines for future research in these directions.

Tor users derive anonymity in part from the size of the Tor user base, but Tor struggles to attract and support more users due to performance limitations. Previous works have proposed modifications to Tor’s path selection algorithm to enhance both performance and security, but many proposals have unintended consequences due to incorporating information related to client location. We instead propose selecting paths using a global view of the network, independent of client location, and we propose doing so with a machine learning classifier to predict the performance of a given path before building a circuit. We show through a variety of simulated and live experimental settings, across different time periods, that this approach can significantly improve performance compared to Tor’s default path selection algorithm and two previously proposed approaches. In addition to evaluating the security of our approach with traditional metrics, we propose a novel anonymity metric that captures information leakage resulting from location-aware path selection, and we show that our path selection approach leaks no more information than the default path selection algorithm.

Anonymous communication networks such as Tor are vulnerable to end-to-end flow correlation, which allows an eavesdropper to link a client with their destination by identifying pairs for network flows that belong to the same circuit. However, existing approaches, while having high accuracy, suffer from a high false positive rate and computational complexity as well as low sensitivity. In this paper, we propose ESPRESSO, a new method designed for Tor traffic correlation attacks that build upon the state-of-the-art DeepCoFFEA. We utilize an aggregated feature representation and we employ Transformers for global processing to capture long-range dependencies. Furthermore, we use an improved window-based amplification strategy to improve performance further. Our preliminary results show significant performance gains over the prior state-of-the-art DeepCoFFEA.

Clickbait headlines work through superlatives and intensifiers, creating information gaps to increase the relevance of their associated links that direct users to time-wasting and sometimes even malicious websites. This approach can be amplified using targeted clickbait that takes publicly available information from social media to align clickbait to users’ preferences and beliefs. In this work, we first conducted preliminary studies to understand the influence of targeted clickbait on users’ clicking behavior. Based on our findings, we involved 24 users in the participatory design of story-based warnings against targeted clickbait. Our analysis of user-created warnings led to four design variations, which we evaluated through an online survey over Amazon Mechanical Turk. Our findings show the significance of integrating information with persuasive narratives to create effective warnings against targeted clickbait. Overall, our studies provide valuable insights into understanding users’ perceptions and behaviors towards targeted clickbait, and the efficacy of story-based interventions.

The evolving landscape of manipulated media, including the threat of deepfakes, has made information verification a daunting challenge for journalists. Technologists have developed tools to detect deepfakes, but these tools can sometimes yield inaccurate results, raising concerns about inadvertently disseminating manipulated content as authentic news. This study examines the impact of unreliable deepfake detection tools on information verification. We conducted role-playing exercises with 24 US journalists, immersing them in complex breaking-news scenarios where determining authenticity was challenging. Through these exercises, we explored questions regarding journalists’ investigative processes, use of a deepfake detection tool, and decisions on when and what to publish. Our findings reveal that journalists are diligent in verifying information, but sometimes rely too heavily on results from deepfake detection tools. We argue for more cautious release of such tools, accompanied by proper training for users to mitigate the risk of unintentionally propagating manipulated content as real news.

In this paper, we present Laserbeak, a new state-of-the-art website fingerprinting attack for Tor that achieves nearly 96% accuracy against FRONT-defended traffic by combining two innovations: 1) multi-channel traffic representations and 2) advanced techniques adapted from state-of-the-art computer vision models. Our work is the first to explore a range of different ways to represent traffic data for a classifier. We find a multi-channel input format that provides richer contextual information, enabling the model to learn robust representations even in the presence of heavy traffic obfuscation. We are also the first to examine how recent advances in transformer models can take advantage of these representations. Our novel model architecture utilizing multi-headed attention layers enhances the capture of both local and global patterns. By combining these innovations, Laserbeak demonstrates absolute performance improvements of up to 36.2% (e.g., from 27.6% to 63.8%) compared with prior attacks against defended traffic. Experiments highlight Laserbeak’s capabilities in multiple scenarios, including a large open-world dataset where it achieves over 80% recall at 99% precision on traffic obfuscated with padding defenses. These advances reduce the remaining anonymity in Tor against fingerprinting threats, underscoring the need for stronger defenses.

Recent website fingerprinting attacks have been shown to achieve very high performance against traffic through Tor. These attacks allow an adversary to deduce the website a Tor user has visited by simply eavesdropping on the encrypted communication. This has consequently motivated the development of many defense strategies that obfuscate traffic through the addition of dummy packets and/or delays. The efficacy and practicality of many of these recent proposals have yet to be scrutinized in detail. In this study, we re-evaluate nine recent defense proposals that claim to provide adequate security with low-overheads using the latest Deep Learning-based attacks. Furthermore, we assess the feasibility of implementing these defenses within the current confines of Tor. To this end, we additionally provide the first on-network implementation of the DynaFlow defense to better assess its real-world utility.

End-to-end flow correlation attacks are among the oldest known attacks on low-latency anonymity networks, and are treated as a core primitive for traffic analysis of Tor. However, despite recent work showing that individual flows can be correlated with high accuracy, the impact of even these state-of-the-art attacks is questionable due to a central drawback: their pairwise nature, requiring comparison between N2pairs of flows to deanonymize N users. This results in a combinatorial explosion in computational requirements and an asymptotically declining base rate, leading to either high numbers of false positives or vanishingly small rates of successful correlation. In this paper, we introduce a novel flow correlation attack, DeepCoFFEA, that combines two ideas to overcome these drawbacks. First, DeepCoFFEA uses deep learning to train a pair of feature embedding networks that respectively map Tor and exit flows into a single low-dimensional space where correlated flows are similar; pairs of embedded flows can be compared at lower cost than pairs of full traces. Second, DeepCoFFEA uses amplification, dividing flows into short windows and using voting across these windows to significantly reduce false positives; the same embedding networks can be used with an increasing number of windows to independently lower the false positive rate. We conduct a comprehensive experimental analysis showing that DeepCoFFEA significantly outperforms state-of-the-art flow correlation attacks on Tor, e.g. 93% true positive rate versus at most 13% when tuned for high precision, with two orders of magnitude speedup over prior work. We also consider the effects of several potential countermeasures on DeepCoFFEA, finding that existing lightweight defenses are not sufficient to secure anonymity networks from this threat.

User-chosen passwords reflecting common strategies and patterns ease memorisation but offer uncertain and often weak security, while system-assigned passwords provide higher security guarantee but suffer from poor memorability. We thus examine the technique to enhance password memorability that incorporates a scientific understanding of long-term memory. In particular, we examine the efficacy of providing users with verbal cues—real-life facts corresponding to system-assigned keywords. We also explore the usability gain of including images related to the keywords along with verbal cues. In our multi-session lab study with 52 participants, textual recognition-based scheme offering verbal cues had a significantly higher login success rate (94.23%) compared to the control condition, i.e. textual recognition without verbal cues (61.54%). When users were provided with verbal cues, adding images contributed to faster recognition of the assigned keywords, and thus had an overall improvement in usability. So, we conducted a field study with 54 participants to further examine the usability of graphical recognition-based scheme offering verbal cues, which showed an average login success rate of 98% in a real-life setting and an overall improvement in login performance with more login sessions. These findings show a promising research direction to gain high memorability for system-assigned passwords.

In this work, we design and evaluate LociMotion, a training interface to learn a strong authentication secret in a single session. LociMotion automatically takes a random password with twelve lowercase letters (56-bit entropy) to generate the training interface. It first leverages users’ spatial and visual (declarative) memory by showing them a video clip based on the method of loci, and then consolidates the learning process by having them play a computer game that leverages their motor (procedural) memory. The results of a memorability study with 300 participants showed that LociMotion had a significantly higher recall success rate than a control condition. A second study with 200 participants demonstrated the effectiveness of LociMotion over a period of time (99%, 96%, and 81% recall success rates after 1, 4, and 18 days, respectively). LociMotion offers an alternative to the spaced repetition technique, as it does not require dozens of training sessions.

Purpose A huge amount of personal and sensitive data are shared on Facebook, which makes it a prime target for attackers. Adversaries can exploit third-party applications connected to a user’s Facebook profiles (i.e. Facebook apps) to gain access to this personal information. Users’ lack of knowledge and the varying privacy policies of these apps make them further vulnerable to information leakage. However, little has been done to identify mismatches between users’ perceptions and the privacy policies of Facebook apps. This paper aims to address this challenge in the work. Design/methodology/approach The authors conducted a lab study with 31 participants, where the authors received data on how they share information on Facebook, their Facebook-related security and privacy practices and their perceptions on the privacy aspects of 65 frequently-used Facebook apps in terms of data collection, sharing and deletion. The authors then compared participants’ perceptions with the privacy policy of each reported app. Participants also reported their expectations about the types of information that should not be collected or shared by any Facebook app. Findings The analysis reveals significant mismatches between users’ privacy perceptions and reality (i.e. privacy policies of Facebook apps), where the authors identified over-optimism not only in users’ perceptions of information collection but also in their self-efficacy in protecting their information in Facebook despite experiencing negative incidents in the past. Originality/value To the best of the knowledge, this is the first study on the gap between users’ privacy perceptions around Facebook apps and reality. The findings from this study offer direction for future research to address that gap through designing usable, effective and personalized privacy notices to help users to make informed decisions about using Facebook apps.

Abstract We introduce Generative Adversarial Networks for Data-Limited Fingerprinting (GANDaLF), a new deep-learning-based technique to perform Website Fingerprinting (WF) on Tor traffic. In contrast to most earlier work on deep-learning for WF, GANDaLF is intended to work with few training samples, and achieves this goal through the use of a Generative Adversarial Network to generate a large set of “fake” data that helps to train a deep neural network in distinguishing between classes of actual training data. We evaluate GANDaLF in low-data scenarios including as few as 10 training instances per site, and in multiple settings, including fingerprinting of website index pages and fingerprinting of non-index pages within a site. GANDaLF achieves closed-world accuracy of 87% with just 20 instances per site (and 100 sites) in standard WF settings. In particular, GANDaLF can outperform Var-CNN and Triplet Fingerprinting (TF) across all settings in subpage fingerprinting. For example, GANDaLF outperforms TF by a 29% margin and Var-CNN by 38% for training sets using 20 instances per site.

Website Fingerprinting (WF) is a type of traffic analysis attack that enables a local passive eavesdropper to infer the victim's activity, even when the traffic is protected by a VPN or an anonymity system like Tor. Leveraging a deep-learning classifier, a WF attacker can gain over 98% accuracy on Tor traffic. In this paper, we explore a novel defense, Mockingbird, based on the idea of adversarial examples that have been shown to undermine machine-learning classifiers in other domains. Since the attacker gets to design and train his attack classifier based on the defense, we first demonstrate that at a straightforward technique for generating adversarial-example based traces fails to protect against an attacker using adversarial training for robust classification. We then propose Mockingbird, a technique for generating traces that resists adversarial training by moving randomly in the space of viable traces and not following more predictable gradients. The technique drops the accuracy of the state-of-the-art attack hardened with adversarial training from 98% to 42-58% while incurring only 58% bandwidth overhead. The attack accuracy is generally lower than state-of-the-art defenses, and much lower when considering Top-2 accuracy, while incurring lower bandwidth overheads.

The creation and distribution of child sexual abuse materials (CSAM) involves a continuing violation of the victims? privacy beyond the original harms they document. A large volume of these materials is distributed via the Freenet anonymity network: in our observations, nearly one third of requests on Freenet were for known CSAM. In this paper, we propose and evaluate a novel approach for investigating these violations of exploited childrens' privacy. Our forensic method distinguishes whether or not a neighboring peer is the actual uploader or downloader of a file or merely a relayer. Our method requires analysis of the traffic sent to a single, passive node only. We evaluate our method extensively. Our in situ measurements of actual CSAM requests show an FPR of 0.002 ± 0.003 for identifying downloaders. And we show an FPR of 0.009 ± 0.018, a precision of 1.00 ± 0.01, and a TPR of 0.44 ± 0.01 for identifying uploaders based on in situ tests. Further, we derive expressions for the FPR and Power of our hypothesis test; perform simulations of single and concurrent downloaders; and characterize the Freenet network to inform parameter selection. We were participants in several United States Federal Court cases in which the use of our method was uniformly upheld.

Deepfakes can be used maliciously to sway public opinion, defame an individual, or commit fraud. Hence, it is vital for journalists and social media platforms, as well as the general public, to be able to detect deepfakes. Existing deepfake detection methods, while highly accurate on datasets they have been trained on, falter in open-world scenarios due to different deepfake generations algorithms, video formats, and compression levels. In this paper, we seek to address this by building on the XceptionNet-based deepfake detection technique that utilizes convolutional latent representations with recurrent structures. In particular, we explore how to leverage a combination of visual frames, edge maps, and dense optical flow maps together as inputs to this architecture. We evaluate these techniques using the FaceForensics++ and DFDC-mini datasets. We also perform extensive studies to evaluate the robustness of our network against adversarial post-processing as well as the generalization capabilities to out-of-domain datasets and manipulation strategies. Our methods, which we call XceptionNet*, achieve 100% accuracy on the popular Face-Forensics-s+ dataset and set new benchmark standards on the difficult DFDC-mini dataset. The XceptionNet* models are shown to exhibit superior performance on cross-domain testing and demonstrate surprising resilience to adversarial manipulations.

Visually similar characters, or homoglyphs, can be used to perform social engineering attacks or to evade spam and plagiarism detectors. It is thus important to understand the capabilities of an attacker to identify homoglyphs - particularly ones that have not been previously spotted - and leverage them in attacks. We investigate a deep-learning model using embedding learning, transfer learning, and augmentation to determine the visual similarity of characters and thereby identify potential homoglyphs. Our approach uniquely takes advantage of weak labels that arise from the fact that most characters are not homoglyphs. Our model drastically outperforms the Normal-ized Compression Distance approach on pairwise homoglyph identification, for which we achieve an average precision of 0.97. We also present the first attempt at clustering homoglyphs into sets of equivalence classes, which is more efficient than pairwise information for security practitioners to quickly lookup homoglyphs or to normalize confusable string encodings. To measure clustering performance, we propose a metric (mBIOU) building on the classic Intersection-Over-Union (IOU) metric. Our clustering method achieves 0.592 mBIOU, compared to 0.430 for the naive baseline. We also use our model to predict over 8,000 previously unknown homoglyphs, and find good early indications that many of these may be true positives. Source code and list of predicted homoglyphs are uploaded to Github: https://github.com/PerryXDeng/weaponizing_unicode.

Abstract A passive local eavesdropper can leverage Website Fingerprinting (WF) to deanonymize the web browsing activity of Tor users. The value of timing information to WF has often been discounted in recent works due to the volatility of low-level timing information. In this paper, we more carefully examine the extent to which packet timing can be used to facilitate WF attacks. We first propose a new set of timing-related features based on burst-level characteristics to further identify more ways that timing patterns could be used by classifiers to identify sites. Then we evaluate the effectiveness of both raw timing and directional timing which is a combination of raw timing and direction in a deep-learning-based WF attack. Our closed-world evaluation shows that directional timing performs best in most of the settings we explored, achieving: (i) 98.4% in undefended Tor traffic; (ii) 93.5% on WTF-PAD traffic, several points higher than when only directional information is used; and (iii) 64.7% against onion sites, 12% higher than using only direction. Further evaluations in the open-world setting show small increases in both precision (+2%) and recall (+6%) with directional-timing on WTF-PAD traffic. To further investigate the value of timing information, we perform an information leakage analysis on our proposed handcrafted features. Our results show that while timing features leak less information than directional features, the information contained in each feature is mutually exclusive to one another and can thus improve the robustness of a classifier.

The website fingerprinting attack allows a low-resource attacker to compromise the privacy guarantees provided by privacy enhancing tools such as Tor. In response, researchers have proposed defenses aimed at confusing the classification tools used by attackers. As new, more powerful attacks are frequently developed, raw attack accuracy has proven inadequate as the sole metric used to evaluate these defenses. In response, two security metrics have been proposed that allow for evaluating defenses based on hand-crafted features often used in attacks. Recent state-of-the-art attacks, however, use deep learning models capable of automatically learning abstract feature representations, and thus the proposed metrics fall short once again. In this study we examine two security metrics and (1) show how these methods can be extended to evaluate deep learning-based website fingerprinting attacks, and (2) compare the security metrics and identify their shortcomings.

Over 8 million users rely on the Tor network each day to protect their anonymity online. Unfortunately, Tor has been shown to be vulnerable to the website fingerprinting attack, which allows an attacker to deduce the website a user is visiting based on patterns in their traffic. The state-of-the-art attacks leverage deep learning to achieve high classification accuracy using raw packet information. Work thus far, however, has examined only one type of media delivered over the Tor network: web pages, and mostly just home pages of sites. In this work, we instead investigate the fingerprintability of video content served over Tor. We collected a large new dataset of network traces for 50 YouTube videos of similar length. Our preliminary experiments utilizing a convolutional neural network model proposed in prior works has yielded promising classification results, achieving up to 55% accuracy. This shows the potential to unmask the individual videos that users are viewing over Tor, creating further privacy challenges to consider when defending against website fingerprinting attacks.

Facebook remains the largest social media platform on the Internet with over one billion active monthly users. A variety of personal and sensitive data is shared on the platform, which makes it a prime target for attackers. Increasingly, we see phishing attacks that take advantage of users' lack of security knowledge, deceiving victims by using fake or compromised accounts to share malicious posts. These attacks may slip undetected by the Facebook defense system, exposing users to potentially be phished or have their devices infected with drive-by downloads and malware. Only a few studies have been conducted to date to understand how users interact with attacks like this in Facebook. In our prior work, we conducted a study to address this challenge using a simulated interface and think-aloud protocol. In this study, we aim to make further progress in understanding the impact of different factors on users' clicking decision in social media through a vignette study that encourages participants to think about realistic scenarios that they might face.

Website Fingerprinting (WF) attacks pose a serious threat to users' online privacy, including for users of the Tor anonymity system. By exploiting recent advances in deep learning, WF attacks like Deep Fingerprinting (DF) have reached up to 98% accuracy. The DF attack, however, requires large amounts of training data that needs to be updated regularly, making it less practical for the weaker attacker model typically assumed in WF. Moreover, research on WF attacks has been criticized for not demonstrating attack effectiveness under more realistic and more challenging scenarios. Most research on WF attacks assumes that the testing and training data have similar distributions and are collected from the same type of network at about the same time. In this paper, we examine how an attacker could leverage N-shot learning---a machine learning technique requiring just a few training samples to identify a given class---to reduce the effort of gathering and training with a large WF dataset as well as mitigate the adverse effects of dealing with different network conditions. In particular, we propose a new WF attack called Triplet Fingerprinting (TF) that uses triplet networks for N-shot learning. We evaluate this attack in challenging settings such as where the training and testing data are collected multiple years apart on different networks, and we find that the TF attack remains effective in such settings with 85% accuracy or better. We also show that the TF attack is also effective in the open world and outperforms traditional transfer learning. On top of that, the attack requires only five examples to recognize a website, making it dangerous in a wide variety of scenarios where gathering and training on a complete dataset would be impractical.

There is heightened concern over deliberately inaccurate news. Recently, so-called deepfake videos and images that are modified by or generated by artificial intelligence techniques have become more realistic and easier to create. These techniques could be used to create fake announcements from public figures or videos of events that did not happen, misleading mass audiences in dangerous ways. Although some recent research has examined accurate detection of deepfakes, those methodologies do not generalize well to real-world scenarios and are not available to the public in a usable form. In this project, we propose a system that will robustly and efficiently enable users to determine whether or not a video posted online is a deepfake. We approach the problem from the journalists' perspective and work towards developing a tool to fit seamlessly into their workflow. Results demonstrate accurate detection on both within and mismatched datasets.

Users of the Tor anonymity system suffer from less‐than‐ideal performance, in part because circuit building and selection processes are not tuned for speed. In this study, the authors examine both the process of selecting among pre‐built circuits and the process of selecting the path of relays for use in building new circuits to improve performance while maintaining anonymity. First, the authors show that having three pre‐built circuits available allows the Tor client to identify fast circuits and improves median time to first byte (TTFB) by 15% over congestion‐aware routing, the current state‐of‐the‐art method. Second, they propose a new path selection algorithm that includes broad geographic location information together with bandwidth to reduce delays. In shadow simulations, 20% faster median TTFB and 11% faster median total download times over congestion‐aware routing for accessing web page‐sized objects were found. The proposed security evaluations show that this approach leads to better or equal security against a generic relay‐level adversary compared to Tor, but increased vulnerability to targeted attacks. The authors explore this trade‐off and find settings of the proposed system that offers good performance, modestly better security against a generic adversary, and only slightly more vulnerability to a targeted adversary.

Website Fingerprinting (WF) is a traffic analysis attack that enables an eavesdropper to infer the victim's web activity even when encrypted and even when using the Tor anonymity system. Using deep learning classifiers, the attack can reach up to 98% accuracy. Existing WF defenses are either too expensive in terms of bandwidth and latency overheads (e.g. 2-3 times as large or slow) or ineffective against the latest attacks. In this work, we explore a novel defense based on the idea of adversarial examples that have been shown to undermine machine learning classifiers in other domains. Our Adversarial Traces defense adds padding to a Tor traffic trace in a manner that reliably fools the classifier into classifying it as coming from a different site. The technique drops the accuracy of the state-of-the-art attack from 98% to 60%, while incurring a reasonable 47% bandwidth overhead, showing its promise as a possible defense for Tor.

Website fingerprinting enables a local eavesdropper to determine which websites a user is visiting over an encrypted connection. State-of-the-art website fingerprinting attacks have been shown to be effective even against Tor. Recently, lightweight website fingerprinting defenses for Tor have been proposed that substantially degrade existing attacks: WTF-PAD and Walkie-Talkie. In this work, we present Deep Fingerprinting (DF), a new website fingerprinting attack against Tor that leverages a type of deep learning called Convolutional Neural Networks (CNN) with a sophisticated architecture design, and we evaluate this attack against WTF-PAD and Walkie-Talkie. The DF attack attains over 98% accuracy on Tor traffic without defenses, better than all prior attacks, and it is also the only attack that is effective against WTF-PAD with over 90% accuracy. Walkie-Talkie remains effective, holding the attack to just 49.7% accuracy. In the more realistic open-world setting, our attack remains effective, with 0.99 precision and 0.94 recall on undefended traffic. Against traffic defended with WTF-PAD in this setting, the attack still can get 0.96 precision and 0.68 recall. These findings highlight the need for effective defenses that protect against this new attack and that could be deployed in Tor.

Abstract The mechanism for picking guards in Tor suffers from security problems like guard fingerprinting and from performance issues. To address these issues, Hayes and Danezis proposed the use of guard sets, in which the Tor system groups all guards into sets, and each client picks one of these sets and uses its guards. Unfortunately, guard sets frequently need nodes added or they are broken up due to fluctuations in network bandwidth. In this paper, we first show that these breakups create opportunities for malicious guards to join many guard sets by merely tuning the bandwidth they make available to Tor, and this greatly increases the number of clients exposed to malicious guards. To address this problem, we propose a new method for forming guard sets based on Internet location. We construct a hierarchy that keeps clients and guards together more reliably and prevents guards from easily joining arbitrary guard sets. This approach also has the advantage of confining an attacker with access to limited locations on the Internet to a small number of guard sets. We simulate this guard set design using historical Tor data in the presence of both relay-level adversaries and networklevel adversaries, and we find that our approach is good at confining the adversary into few guard sets, thus limiting the impact of attacks.

Password schemes based on online map locations are an emerging topic in authentication research. GeoPass is a promising such scheme, as it provides satisfactory resilience against online guessing and showed high memorability (97%) in a single-password laboratory study. In this article, we investigate more deeply into the potential of GeoPass through four separate studies. First, in a 2-month-long field study, we found that users in a real-world setting remembered their location passwords 96.1% of the time and showed improvement with more login sessions. Then, in a study of interference effects in Geopass, in which each participant had to remember four separate location passwords, we found that memorability was <70%, with 41.5% of login failures due to interference. Based on these findings, we propose to address interference issues in GeoPass with mental stories, where users are asked to create a meaningful association between their location password and the corresponding account. We tested the efficacy of this approach through a second interference study, where the memorability rate for GeoPass was >97%, with only 3.4% of login attempts failing due to interference. We also conducted a shoulder-surfing study to examine the resilience of GeoPass against this attack. Based on our results, we identify the promising aspects of location passwords that should be further studied in future research.

Most phishing pages try to convince users that they are legitimate sites by imitating visual signals like logos from the websites they are targeting. Visual similarity detection methods look for these imitations between the screen-shots of the suspect pages and an image database of the most targeted websites. Existing approaches, however, are either too slow for real-time use or not robust to manipulation. In this work, we design a multi-phase framework for visual similarity detection. The first phase of the framework should rule out the bulk of websites quickly, but without introducing false negatives and with resistance to attacker manipulations. Later phases can use more heavyweight operations to decide whether or not to warn the user about possible phishing. In this abstract, we focus on the first phase. In experiments, our proposed method rules out more than half of the test cases with zero false negatives with less than 5 ms of processing time per page.

Purpose Two-factor authentication is being implemented more broadly to improve security against phishing, shoulder surfing, keyloggers and password guessing attacks. Although passwords serve as the first authentication factor, a common approach to implementing the second factor is sending a one-time code, either via e-mail or text message. The prevalence of smartphones, however, creates security risks in which a stolen phone leads to user’s accounts being accessed. Physical tokens such as RSA’s SecurID create extra burdens for users and cannot be used on many accounts at once. This study aims to improve the usability and security for two-factor online authentication. Design/methodology/approach The authors propose a novel second authentication factor that, similar to passwords, is also based on something the user knows but operates similarly to a one-time code for security purposes. The authors design this component to provide higher security guarantee with minimal memory burden and does not require any additional communication channels or hardware. Motivated by psychology research, the authors leverage users’ autobiographical memory in a novel way to create a secure and memorable component for two-factor authentication. Findings In a multi-session lab study, all of the participants were able to log in successfully on the first attempt after a one-week delay from registration and reported satisfaction on the usability of the scheme. Originality/value The results indicate that the proposed approach to leverage autobiographical memory is a promising direction for further research on second authentication factor based on something the user knows.

Abstract Prior approaches to AS-aware path selection in Tor do not consider node bandwidth or the other characteristics that Tor uses to ensure load balancing and quality of service. Further, since the AS path from the client’s exit to her destination can only be inferred once the destination is known, the prior approaches may have problems constructing circuits in advance, which is important for Tor performance. In this paper, we propose and evaluate DeNASA, a new approach to AS-aware path selection that is destination-naive, in that it does not need to know the client’s destination to pick paths, and that takes advantage of Tor’s circuit selection algorithm. To this end, we first identify the most probable ASes to be traversed by Tor streams. We call this set of ASes the Suspect AS list and find that it consists of eight highest ranking Tier 1 ASes. Then, we test the accuracy of Qiu and Gao AS-level path inference on identifying the presence of these ASes in the path, and we show that inference accuracy is 90%. We develop an AS-aware algorithm called DeNASA that uses Qiu and Gao inference to avoid Suspect ASes. DeNASA reduces Tor stream vulnerability by 74%. We also show that DeNASA has performance similar to Tor. Due to the destination-naive property, time to first byte (TTFB) is close to Tor’s, and due to leveraging Tor’s bandwidth-weighted relay selection, time to last byte (TTLB) is also similar to Tor’s.

Sybil attacks present a significant threat to many Internet systems and applications, in which a single adversary inserts multiple colluding identities in the system to compromise its security and privacy. Recent work has advocated the use of social-network-based trust relationships to defend against Sybil attacks. However, most of the prior security analyses of such systems examine only the case of social networks at a single instant in time. In practice, social network connections change over time, and attackers can also cause limited changes to the networks. In this work, we focus on the temporal dynamics of a variety of social-network-based Sybil defenses. We describe and examine the effect of novel attacks based on: (a) the attacker's ability to modify Sybil-controlled parts of the social-network graph, (b) his ability to change the connections that his Sybil identities maintain to honest users, and (c) taking advantage of the regular dynamics of connections forming and breaking in the honest part of the social network. We find that against some defenses meant to be fully distributed, such as SybilLimit and Persea, the attacker can make dramatic gains over time and greatly undermine the security guarantees of the system. Even against centrally controlled Sybil defenses, the attacker can eventually evade detection (e.g. against SybilInfer and SybilRank) or create denial-of-service conditions (e.g. against Ostra and SumUp). After analysis and simulation of these attacks using both synthetic and real-world social network topologies, we describe possible defense strategies and the trade-offs that should be explored. It is clear from our findings that temporal dynamics need to be accounted for in Sybil defense or else the attacker will be able to undermine the system in unexpected and possibly dangerous ways.

Given the choice, users produce passwords reflecting common strategies and patterns that ease recall but offer uncertain and often weak security. System-assigned passwords provide measurable security but suffer from poor memorability. To address this usability-security tension, we argue that systems should assign random passwords but also help with memorization and recall. We investigate the feasibility of this approach with CuedR, a novel cued-recognition authentication scheme that provides users with multiple cues (visual, verbal, and spatial) and lets them choose the cues that best fit their learning process for later recognition of system-assigned keywords. In our lab study, all 37 of our participants could log in within three attempts one week after registration (mean login time: 38.0 seconds). A pilot study on using multiple CuedR passwords also showed 100% recall within three attempts. Based on our results, we suggest appropriate applications for CuedR, such as financial and e-commerce accounts.

P2P systems are inherently vulnerable to Sybil attacks, in which an attacker creates a large number of identities and uses them to control a substantial fraction of the system. We propose Persea, a novel P2P system that derives its Sybil resistance by assigning IDs through a bootstrap tree, the graph of how nodes have joined the system through invitations. Unlike prior Sybil-resistant P2P systems based on social networks, Persea does not rely on two key assumptions: (1) that the social network is fast mixing and (2) that there is a small ratio of attack edges to honest nodes. Both assumptions have been shown to be unreliable in real social networks. A node joins Persea when it gets an invitation from an existing node in the system. The inviting node assigns a node ID to the joining node and gives it a chunk of node IDs for further distribution. For each chunk of ID space, the attacker needs to socially engineer a connection to another node already in the system. The hierarchical distribution of node IDs confines a large attacker botnet to a considerably smaller region of the ID space than in a normal P2P system. We then build upon this hierarchical ID space to make a distributed hash table (DHT) based on the Kad network. The Persea DHT uses a replication mechanism in which each (key, value) pair is stored in nodes that are evenly spaced over the network. Thus, even if a given region is occupied by attackers, the desired (key, value pair can be retrieved from other regions. We evaluate Persea in analysis and in simulations with social network datasets and show that it provides better lookup success rates than prior work with modest overheads.

Distributed hash tables (DHTs), such as Chord and Kademlia, offer an efficient means to locate resources in peer-to-peer networks. Unfortunately, malicious nodes on a lookup path can easily subvert such queries. Several systems, including Halo (based on Chord) and Kad (based on Kademlia), mitigate such attacks by using redundant lookup queries. Much greater assurance can be provided; we present Reputation for Directory Services (ReDS), a framework for enhancing lookups in redundant DHTs by tracking how well other nodes service lookup requests. We describe how the ReDS technique can be applied to virtually any redundant DHT including Halo and Kad. We also study the collaborative identification and removal of bad lookup paths in a way that does not rely on the sharing of reputation scores, and we show that such sharing is vulnerable to attacks that make it unsuitable for most applications of ReDS. Through extensive simulations, we demonstrate that ReDS improves lookup success rates for Halo and Kad by 80 percent or more over a wide range of conditions, even against strategic attackers attempting to game their reputation scores and in the presence of node churn.

P2P systems are inherently vulnerable to Sybil attacks, in which an attacker can have a large number of identities and use them to control a substantial fraction of the system. We propose Persea, a novel P2P system that is more robust against Sybil attacks than prior approaches. Persea derives its Sybil resistance by assigning IDs through a bootstrap tree, the graph of how nodes have joined the system through invitations. More specifically, a node joins Persea when it gets an invitation from an existing node in the system. The inviting node assigns a node ID to the joining node and gives it a chunk of node IDs for further distribution. For each chunk of ID space, the attacker needs to socially engineer a connection to another node already in the system. This hierarchical distribution of node IDs confines a large attacker botnet to a considerably smaller region of the ID space than in a normal P2P system. Persea uses a replication mechanism in which each (key,value) pair is stored in nodes that are evenly spaced over the network. Thus, even if a given region is occupied by attackers, the desired (key,value) pair can be retrieved from other regions. We compare our results with Kad, Whanau, and X-Vine and show that Persea is a better solution against Sybil attacks.

Despite advances in biometrics and other technologies, passwords remain the most commonly used means of authentication in computer systems. Users maintain different security levels for different passwords. In this study, we examine the degree of similarity among passwords of different security levels of a user. We conducted a laboratory experiment with 80 students from the University of Texas at Arlington (UTA). We asked the subjects to construct new passwords for websites of different security levels. We collected the lower-level passwords (e.g., passwords for online news sites) constructed by the subjects, combined them with a comprehensive wordlist, and performed dictionary attacks on their constructed passwords from the higher-level sites (e.g., banking websites). We could successfully crack almost one-third of their constructed passwords from the higher-level sites with this method. This suggests that, if a user's lower-level password is leaked, it can be used effectively by an attacker to crack some of the user's higher-level passwords.

Wireless Sensor Networks (WSN) are vulnerable to jamming attacks where an adversary injects strong noises to interfere with the normal transmission. It is crucial to detect such jamming attacks as fast as possible. Existing studies have shown that an effective indicator of jamming is the packet delivery ratio (PDR). However, current PDR-based schemes use the end-to-end packet delivery ratio, which requires one to observe communication for a long time before a good decision is made. In this paper, we propose collaborative detection, which evaluates the packet delivery ratio in an given area instead of a pair of nodes. The intuition is that the attacker often jams an area of his interest, not just two specific nodes. The benefit is that we can detect jamming attacks in a much faster way. We have evaluated the performance of our idea on TelosB motes. The results show that we can effectively and quickly detect jamming attacks.

In opportunistic environments, tasks such as content sharing and service execution among remote devices are facilitated by relays (devices with short-range wireless connectivity) that receive data, move around, and then forward the data. To achieve high throughput, it is important to secure forwarding and provide incentives for participation by relays. However, it is extremely challenging to monitor the behavior of relays in an opportunistic network due to sparse connectivity. Existing schemes do not work when selfish/malicious relays collude with each other to forge routing metrics, drop useful data, flood the network, or earn extra reward.

Due to the unattended nature of wireless sensor networks, an adversary can physically capture and compromise sensor nodes and then mount a variety of attacks with the compromised nodes. To minimize the damage incurred by the compromised nodes, the system should detect and revoke them as soon as possible. To meet this need, researchers have recently proposed a variety of node compromise detection schemes in wireless ad hoc and sensor networks. For example, reputation-based trust management schemes identify malicious nodes but do not revoke them due to the risk of false positives. Similarly, software-attestation schemes detect the subverted software modules of compromised nodes. However, they require each sensor node to be attested periodically, thus incurring substantial overhead. To mitigate the limitations of the existing schemes, we propose a zone-based node compromise detection and revocation scheme in wireless sensor networks. The main idea behind our scheme is to use sequential hypothesis testing to detect suspect regions in which compromised nodes are likely placed. In these suspect regions, the network operator performs software attestation against sensor nodes, leading to the detection and revocation of the compromised nodes. Through quantitative analysis and simulation experiments, we show that the proposed scheme detects the compromised nodes with a small number of samples while reducing false positive and negative rates, even if a substantial fraction of the nodes in the zone are compromised. Additionally, we model the detection problem using a game theoretic analysis, derive the optimal strategies for the attacker and the defender, and show that the attacker's gain from node compromise is greatly limited by the defender when both the attacker and the defender follow their optimal strategies.

While many protocols for sensor network security provide confidentiality for the content of messages, contextual information usually remains exposed. Such contextual information can be exploited by an adversary to derive sensitive information such as the locations of monitored objects and data sinks in the field. Attacks on these components can significantly undermine any network application. Existing techniques defend the leakage of location information from a limited adversary who can only observe network traffic in a small region. However, a stronger adversary, the global eavesdropper, is realistic and can defeat these existing techniques. This paper first formalizes the location privacy issues in sensor networks under this strong adversary model and computes a lower bound on the communication overhead needed for achieving a given level of location privacy. The paper then proposes two techniques to provide location privacy to monitored objects (source-location privacy)-periodic collection and source simulation-and two techniques to provide location privacy to data sinks (sink-location privacy)-sink simulation and backbone flooding. These techniques provide trade-offs between privacy, communication cost, and latency. Through analysis and simulation, we demonstrate that the proposed techniques are efficient and effective for source and sink-location privacy in sensor networks.

ABSTRACT Disruption‐tolerant networking holds a great deal of potential for making communications easier and more flexible in pervasive assistive environments. However, security and privacy must be addressed to make these communications acceptable with respect to protecting patient privacy. We propose EnPassant, a system for using disruption‐tolerant networking in privacy‐preserving way. EnPassant uses concepts from anonymous communications, re‐routing messages through groups of peer nodes to hide the relation between the sources and destinations. We describe a set of protocols that explore a practical range of tradeoffs between privacy and communication costs by modifying how closely the protocol adheres to the optimal predicted path. We also describe the cryptographic tools needed to facilitate changes in‐group membership. Finally, we present the results of extensive trace‐based simulation experiments that allow us to both compare between our proposed protocols and observe the costs of increasing the number of groups and intermediate nodes in a path. Copyright © 2010 John Wiley & Sons, Ltd.

Due to the unattended nature of wireless sensor networks, an adversary can capture and compromise sensor nodes, make replicas of them, and then mount a variety of attacks with these replicas. These replica node attacks are dangerous because they allow the attacker to leverage the compromise of a few nodes to exert control over much of the network. Several replica node detection schemes have been proposed in the literature to defend against such attacks in static sensor networks. However, these schemes rely on fixed sensor locations and hence do not work in mobile sensor networks, where sensors are expected to move. In this work, we propose a fast and effective mobile replica node detection scheme using the Sequential Probability Ratio Test. To the best of our knowledge, this is the first work to tackle the problem of replica node attacks in mobile sensor networks. We show analytically and through simulation experiments that our scheme detects mobile replicas in an efficient and robust manner at the cost of reasonable overheads.

A mix is a communication proxy that hides the relationship between incoming and outgoing messages. Routing traffic through a path of mixes is a powerful tool for providing privacy. When mixes are used for interactive communication, such as VoIP and web browsing, attackers can undermine user privacy by observing timing information along the path. Mixes can prevent these attacks by inserting dummy packets (cover traffic) to obfuscate timing information in each stream. A recently proposed defense called adaptive padding makes cover traffic more effective by ensuring that statistically unusual gaps between packets are partially filled in with dummy packets. In this work, we propose Selective Cross Correlation (SCC), an attack that an eavesdropper could employ to de-anonymize users despite the use of adaptive padding. The main insight of our approach is that, with the defense, the timings at one end of the stream are effectively a subset of the timings at the other end of the stream. By considering the network conditions, an appropriate correlation window can be found and used to effectively remove the cover traffic, thereby enabling us to correlate both ends of the stream. We have conducted real network experiments and have found that SCC greatly improves attacker effectiveness over prior techniques against the defense. With SCC, the attacker is nearly as successful as when no defense is applied. This attack demonstrates the need for more robust defenses against statistical timing attacks.

Due to the unattended nature of wireless sensor networks, an adversary can capture and compromise sensor nodes, generate their replicas, and thus mount a variety of attacks with these replicas. Such attacks are dangerous because they allow the attacker to leverage the compromise of a few nodes to exert control over much of the network. Several replica node detection schemes have been proposed in the literature to defend against such attacks in static sensor networks. However, these schemes rely on fixed sensor locations and hence do not work in mobile sensor networks, where sensors are expected to move. In this work, we propose a fast and effective mobile replica node detection scheme using the Sequential Probability Ratio Test. To the best of our knowledge, this is the first work to tackle the problem of replica node attacks in mobile sensor networks. We show analytically and through simulation experiments that our scheme provides effective and robust replica detection capability with reasonable overheads.

Effective security mechanisms are essential to the widespread deployment of pervasive systems. Much of the research focus on security in pervasive computing has revolved around distributed trust management. While such mechanisms are effective in specific environments, there is no generic framework for deploying and extending these mechanisms over a variety of pervasive systems. We present the design and implementation of a novel framework called distributed trust toolkit (DTT), for implementing and evaluating trust mechanisms in pervasive systems. The DTT facilitates the extension and adaptation of trust mechanisms by abstracting trust mechanisms into interchangeable components. Furthermore, the DTT provides a set of tools and interfaces to ease implementation of trust mechanisms and facilitate their execution on a variety of platforms and networks. In addition to the adaptability and extensibility provided by this design, we demonstrate through simulation that use of DTT improves utilization of resources and enhances performance of existing trust mechanisms in pervasive systems. We are currently developing an implementation of the DTT that can be easily deployed in pervasive environments.

Due to the unattended nature of wireless sensor networks, an adversary can physically capture and compromise sensor nodes and then mount a variety of attacks with these compromised nodes. To minimize the damage incurred by compromised nodes, the system should detect and revoke them as soon as possible. To meet this need, we propose a zone-based node compromise detection and revocation scheme in sensor networks. The main idea of the proposed scheme is to use the sequential hypothesis testing to detect suspect regions in which compromised nodes are likely placed. In these suspect regions, the network operator performs software attestation against sensor nodes, leading to the detection and revocation of the compromised nodes. Through analysis and simulation, we show that the proposed scheme provides effective and robust node compromise detection and revocation capability with little overhead.

In open environments, such as mobile peer-to-peer systems, participants may need to access resources from unknown users. A critical security concern in such systems is the access of faulty resources, thereby wasting the requester's time and energy and possibly causing damage to her system. A common approach to mitigating the problem involves reputation mechanisms; however, since reputation relies on cooperation, a reputation mechanism's effectiveness can be significantly diminished in hostile environments. Reputation systems also require substantial communication among peers leading to: i) vulnerability to errors caused by intermittent connectivity; ii) message delivery disruptions caused by malicious peers; and iii) energy sapping message overheads. In this paper, we present AREX, a low-cost, adaptive mechanism designed to provide security for peers in hostile and uncertain environments, which are common in mobile P2P systems. AREX features an adaptive exploration strategy that increases the system's utility for benign peers and decreases the systempsilas utility for malicious peers. AREX reduces vulnerabilities and energy costs by operating without communication between peers. Through simulation, we demonstrate AREX's ability to reduce energy costs, protect benign peers, and diminish malicious peers' motivation to attack in a variety of hostile environments.

Using analysis, simulation, and experimentation, we examine the threat against anonymous communications posed by passive-logging attacks. In previous work, we analyzed the success of such attacks under various assumptions. Here, we evaluate the effects of these assumptions more closely. First, we analyze the Onion Routing-based model used in prior work in which a fixed set of nodes remains in the system indefinitely. We show that for this model, by removing the assumption of uniformly random selection of nodes for placement in the path, initiators can greatly improve their anonymity. Second, we show by simulation that attack times are significantly lower in practice than bounds given by analytical results from prior work. Third, we analyze the effects of a dynamic membership model, in which nodes are allowed to join and leave the system; we show that all known defenses fail more quickly when the assumption of a static node set is relaxed. Fourth, intersection attacks against peer-to-peer systems are shown to be an additional danger, either on their own or in conjunction with the predecessor attack. Finally, we address the question of whether the regular communication patterns required by the attacks exist in real traffic. We collected and analyzed the Web requests of users to determine the extent to which basic patterns can be found. We show that, for our study, frequent and repeated communication to the same Web site is common.

While many protocols for sensor network security provide confidentiality for the content of messages, contextual information usually remains exposed. Such information can be critical to the mission of the sensor network, such as the location of a target object in a monitoring application, and it is often important to protect this information as well as message content. There have been several recent studies on providing location privacy in sensor networks. However, these existing approaches assume a weak adversary model where the adversary sees only local network traffic. We first argue that a strong adversary model, the global eavesdropper, is often realistic in practice and can defeat existing techniques. We then formalize the location privacy issues under this strong adversary model and show how much communication overhead is needed for achieving a given level of privacy. We also propose two techniques that prevent the leakage of location information: periodic collection and source simulation. Periodic collection provides a high level of location privacy, while source simulation provides trade-offs between privacy, communication cost, and latency. Through analysis and simulation, we demonstrate that the proposed techniques are efficient and effective in protecting location information from the attacker.

Highly distributed anonymous communications systems have the promise to reduce the effectiveness of certain attacks and improve scalability over more centralized approaches. Existing approaches, however, face security and scalability issues. Requiring nodes to have full knowledge of the other nodes in the system, as in Tor and Tarzan, limits scalability and can lead to intersection attacks in peer-to-peer configurations. MorphMix avoids this requirement for complete system knowledge, but users must rely on untrusted peers to select the path. This can lead to the attacker controlling the entire path more often than is acceptable.To overcome these problems, we propose Salsa, a structured approach to organizing highly distributed anonymous communications systems for scalability and security. Salsa is designed to select nodes to be used in anonymous circuits randomly from the full set of nodes, even though each node has knowledge of only a subset of the network. It uses a distributed hash table based on hashes of the nodes' IP addresses to organize the system. With a virtual tree structure, limited knowledge of other nodes is enough to route node lookups throughout the system. We use redundancy and bounds checking when performing lookups to prevent malicious nodes from returning false information without detection. We show that our scheme prevents attackers from biasing path selection, while incurring moderate overheads, as long as the fraction of malicious nodes is less than 20%. Additionally, the system prevents attackers from obtaining a snapshot of the entire system until the number of attackers grows too large (e.g. 15% for 10000 peers and 256 groups). The number of groups can be used as a tunable parameter in the system, depending on the number of peers, that can be used to balance performance and security.

Once electronic content has been released it is very difficult to prevent copies of the content from being widely distributed. Such distribution can cause economic harm to the content's copyright owner and others. Our protocol, SPIES, allows one party to sell a secret to second party and provides an economic incentive for two parties to limit sharing of a secret between themselves. We do not use watermarking or traditional DRM mechanisms. We focus on content which is to be shared between two parties only, which is valuable, and which only needs to be protected for a limited amount of time. Examples include passwords to a subscription service, pre-release of media for review, or content shared but bound by a non disclosure agreement. With SPIES, any possesor of the content can receive a portion of the funds placed in escrow by the two legitimate possesors. We analyze this system and show that the best strategy of the content provider and content consumer to maximize their utility is to use SPIES and not share the content further. We deal successfully with a "dummy registration" attack in which multiple false identities are used in an attempt to get a higher payment. We also discuss how to determine the correct escrow amount.

There have been a number of protocols proposed for anonymous network communication. In this paper, we investigate attacks by corrupt group members that degrade the anonymity of each protocol over time. We prove that when a particular initiator continues communication with a particular responder across path reformations, existing protocols are subject to the attack. We use this result to place an upper bound on how long existing protocols, including Crowds, Onion Routing, Hordes, Web Mixes, and DC-Net, can maintain anonymity in the face of the attacks described. This provides a basis for comparing these protocols against each other. Our results show that fully connected DC-Net is the most resilient to these attacks, but it suffers from scalability issues that keep anonymity group sizes small. We also show through simulation that the underlying topography of the DC-Net affects the resilience of the protocol: as the number of neighbors a node has increases the strength of the protocol increases, at the cost of higher communication overhead.

We study the threat that passive logging attacks pose to anonymous communications. Previous work analyzed these attacks under limiting assumptions. We first describe a possible defense that comes from breaking the assumption of uniformly random path selection. Our analysis shows that the defense improves anonymity in the static model, where nodes stay in the system, but fails in a dynamic model, in which nodes leave and join. Additionally, we use the dynamic model to show that the intersection attack creates a vulnerability in certain peer-to-peer systems for anonymous communications. We present simulation results that show that attack times are significantly lower in practice than the upper bounds given by previous work. To determine whether users' Web traffic has communication patterns required by the attacks, we collected and analyzed the Web requests of users. We found that, for our study frequent and repeated communication to the same Web site is common.