1,721,077 research outputs found
Preventing Content-Mismatch Attacks on Video-Based Protocol Tunneling
No full textMultimedia protocol tunneling has emerged as a promising approach to allowing users to circumvent online censorship. By encapsulating data within an audio/video channel like that provided by Skype, a user's traffic can be hidden from censors. However, the
unconventional use of the channel required by such approaches can produce traffic with abnormal characteristics that allow it to be identified and blocked by censors, in what is known as a content-mismatch attack. In this thesis we focus on protocol tunneling over video channels, and investigate whether using steganography to carefully hide data within a video can minimize differences in channel content, preventing content-mismatch attacks. We designed and implemented a prototype which uses this approach to send data over the Skype video channel, and evaluated whether this prototype is more resilient to content-mismatch attacks. We found that compared to other video-based approaches, the traffic produced by our prototype is harder to distinguish from normal traffic. However, because it can still be identified with a low false positive rate, censors can block a majority of the traffic produced by our prototype while disrupting normal traffic very little, showing that further work is needed
Guard Placement Attacks on Path Selection Algorithms for Tor
No full textThe popularity of Tor has made it an attractive target for a variety of deanonymization and fingerprinting attacks. Location-based path selection algorithms have been proposed as a countermeasure to defend against such attacks. However, adversaries can exploit the location-awareness of these algorithms by strategically placing relays in locations that increase their chances of being selected as a client's guard. Being chosen as a guard facilitates website fingerprinting and traffic correlation attacks over extended time periods. In this thesis, we rigorously define and analyze the guard placement attack. We present novel guard placement attacks and show that three state-of-the-art Tor path selection algorithms---Counter-RAPTOR, DeNASA, and LASTor---are vulnerable to these attacks. We overcome defenses considered by all three systems. Our findings indicate that existing location-based path-selection algorithms allow guards to achieve disproportionately high selection probabilities relative to the cost required to run the guard. Lastly, we propose and evaluate a generic defense mechanism that provably defends any guard selection algorithm against guard placement attacks. We run the defense mechanism on each of the three algorithms we attacked, and find that our defense significantly enhances the security of these algorithms against guard placement attacks with only minimal impact to their original security or performance goals
An Evaluation of Snowflake as an Indistinguishable Censorship Circumvention Tool
No full textWhile absolute control of user activity has become infeasible, many forms of internet censorship are prevalent in authoritarian countries. The Tor network is a powerful tool for circumvention but is often detectable using deep-packet inspection. Pluggable transports address this attack by transforming the traffic between the client and the bridge. In this thesis, we evaluate Snowflake, a novel pluggable transport, as an indistinguishable censorship circumvention tool. Snowflake employs WebRTC, a popular suite of web frameworks and protocols, to establish a connection to the Tor network. We collect 865 instances of WebRTC from Snowflake, Facebook Messenger, Google Hangouts, and Discord and observe that Snowflake is identifiable among these applications with 100% accuracy. We show that several features of Snowflake’s WebRTC implementation, among them the extensions and cipher suites offered, are unique to Snowflake. Finally, we suggest recommendations for improving fingerprint resistance in Snowflake and future work to continue strengthening its implementation
Towards Live Monitoring of CAA Compliance: A Modern Look at the CAA Landscape
No full textCertificate Authorities (CAs) must abide by Certificate Authority Authorization (CAA) records which specify which CAs can issue certificates for particular domains. I am developing a machine learning model to aid in the process of live monitoring for potential certificate mis-issuances which differentiates between benign and malicious mismatches. However, since malicious CAs are so rare and difficult to come across, we are unlikely to observe this occurrence in training data. Thus, the goal of this project is to build a framework for mismatch detection that can use information from the past, specifically past observations about what triggered false positives, to appropriately flag mismatches that display real cause for concern.
This report features the data collection and processing pipelines for certificate data, as well as initial insights into what causes mismatches between the certificate issuer and those listed on the CAA record. Utilizing features attained from manual inspection, I aim to group together similar false positives through unsupervised learning to allow for more streamlined investigation. Future work includes further manual inspection, clustering, more robust data collection, analysis of clustering using chosen features on unseen data, creation of a classifier, and creation of a database of CA relationships. This work will be integrated with the parallel work of Kenny Poor to create a live monitoring system for CA compliance
Implementing Multi-Perspective Issuance Corroboration (MPIC) on AWS
No full textThis project aims to implement Multi-Perspective Issuance Corroboration
(MPIC) using Amazon’s cloud compute service AWS to run remote vantage points.
The goal is to implement MPIC in a way that is quickly useful to smaller Certificate
Authorities (CA). The open-source MPIC implementation uses an open API Gate-
way to handle CA requests. The API allows CA’s to specify DNS or HTTP based
validation as well as the number of perspectives, their locations, and how many
perspective constitute quorum. Using three perspectives both validation methods
have a total latency < 3s. This implementation will assist smaller CAs in quickly
implementing MPIC. Additionally, making MPIC a minimum CA requirement is
in CA Browser Forum ballot SC-067. This implementation will make passing this
ballot easier as it reduced the burden on smaller CAs
The Impact of the Online Certificate Status Protocol on User Privacy
No full textIn the current Internet Public Key Infrastructure (PKI), trusted third parties called Certificate Authorities (CAs) issue digitally-signed certificates affirming ownership of a domain. The Online Certificate Status Protocol (OCSP) was introduced to allow applications to verify that a given certificate had not been revoked, by querying an OCSP responder (server). By requiring applications to send unencrypted requests to third-party responders each time they wish to validate a certificate’s status, OCSP is generally understood to present a threat to user privacy, leaking information about user web behavior. Ironically, the publicity of this issue has not led to significant study of the nature and scope of this threat.
In this work, we outline and conduct a large-scale measurement of OCSP traffic associated with visits to popular websites. We then use this data to assess the extent to which entities at three levels—CAs, Content Delivery Networks, and ASes—are capable of inferring the destination sites a user visits using passively-observed OCSP traffic. We also consider the ramifications of OCSP privacy leakage in the context of anonymous networks such as Tor, and propose a novel correlation attack that incorporates OCSP traffic
Quantifying Attributes of Privacy Policies Using Contextual Integrity
No full textIdentifying vulnerabilities in systems is a critical step in safeguarding users’ privacy but is often only accomplished after an adversary exploiting an existing flaw in a system. Contextual Integrity provides a framework for understanding information flows which are appropriate to the privacy norms associated with a given context. By using the framework which Contextual Integrity affords, I discovered a way to potentially model the privacy characteristics of a service’s privacy policy in a systematic and quantitative manner. This characterization may render privacy policies more transparent for users and assist developers in making their services more secure
An Evaluation of Snowflake as an Indistinguishable Censorship Circumvention Tool
No full textWhile absolute control of user activity has become infeasible, many forms of internet censorship are prevalent in authoritarian countries. The Tor network is a powerful tool for circumvention but is often detectable using deep-packet inspection. Pluggable transports address this attack by transforming the traffic between the client and the bridge. In this thesis, we evaluate Snowflake, a novel pluggable transport, as an indistinguishable censorship circumvention tool. Snowflake employs WebRTC, a popular suite of web frameworks and protocols, to establish a connection to the Tor network. We collect 865 instances of WebRTC from Snowflake, Facebook Messenger, Google Hangouts, and Discord and observe that Snowflake is identifiable among these applications with 100% accuracy. We show that several features of Snowflake’s WebRTC implementation, among them the extensions and cipher suites offered, are unique to Snowflake. Finally, we suggest recommendations for improving fingerprint resistance in Snowflake and future work to continue strengthening its implementation
Censorship Circumvention Using Generative Adversarial Networks
No full textCensorship circumventing technologies have been developed in response to attempts to censor Internet communication, but the technological capabilities of censors have continued to advance. Recent approaches to censorship circumvention have focused on multimedia protocol tunneling as a means to transmit covert information while evading detection by censors. One such approach, Voiceover, is an audio-based protocol tunnel that encodes covert data in audio signals and shapes the audio signals to match the timing properties of human speech in order to mitigate a censor's ability to identify Voiceover traffic. However, any censorship circumvention regime needs to also provide reliable communication. Voiceover as currently proposed does not possess any guarantees of data integrity or the reliability of the protocol amid application-layer transformations and disruptions. This thesis aims to be a continuation and evaluation of the work done in Voiceover. Our first contribution is to implement a rudimentary reliability layer within Voiceover that provides for message integrity and increased message recoverability through notions of data framing, checksums and redundancy. Our second contribution is to improve the usability of Voiceover through automation, maximizing throughput, improving demodulation time, and increasing the robustness of bidirectional communication. Our third contribution is to demonstrate the value of choice of protocol tunnel and the flexibility of the reliability layer by showing that Skype for Web provides a transmission channel unobservable to packet size analysis. Our fourth contribution is to demonstrate the value of the novel audio shaping approach by showing that audio shaping decreases the ability of a classifier to identify Voiceover transmissions based on inter-packet timing statistics. These results demonstrate that the design choices of Voiceover go a long way to achieving unobservable communication
Towards a provably-certifiable defense for multi-label classifiers against adversarial patches
No full textThe advent of deep learning has brought about vast improvements to computer vision systems and enabled technologies such as self-driving cars, facial recognition, etc. Nevertheless, these models have been found to be susceptible to adversarial attacks. Of particular importance to the research community are patch attacks, which have been found to be realizable in the physical world. As a result, researchers have proposed a variety of defense mechanisms in order to circumvent patch attacks. A security "arms race" between attackers and defenders has made certifiable defenses for patch attacks, which feature provable guarantees on robustness, especially valuable in the ML security community. While certifiable defenses like PatchCleanser and ObjectSeeker have been successful at providing guarantees on robustness in the single-label classification and object detection domains respectively, less work has been done on proposing a certifiable defense for patch attacks in the multi-label classification domain. To this end, we propose an extension to PatchCleanser for the multi-label classification domain called Multi-Label PatchCleanser. By constructing an inference algorithm and certification procedure in tandem, we are able to define notions of robustness for the multi-label classification setting based on precision and recall. We find that Multi-Label PatchCleanser can achieve non-trivial robustness on the MSCOCO 2014 validation dataset while maintaining high clean performance; this can be augmented by techniques
such as cutout pre-training and alternative computer vision backbones. Additionally, we discover a promising insight about mask augmentations in the multi-label classification domain which provides improvements to our current baseline robustness
- …
