Robotic aerial vehicles (RAVs) have been increasingly deployed in various areas (e.g., commercial, military, scientific, and entertainment). However, RAVs’ security and safety issues could not only arise from either of the “cyber” domain (e.g., control software) and “physical” domain (e.g., vehicle control model) but also stem in their interplay. Unfortunately, existing work had focused mainly on either the “cyber-centric” or “control-centric” approaches. However, such a single-domain focus could overlook the security threats caused by the interplay between the cyber and physical domains. <br>In this thesis, we present cyber-physical analysis and hardening to secure RAV controllers. Through a combination of program analysis and vehicle control modeling, we first developed novel techniques to (1) connect both cyber and physical domains and then (2) analyze individual domains and their interplay. Specifically, we describe how to detect bugs after RAV accidents using provenance (Mayday), how to proactively find bugs using fuzzing (RVFuzzer), and how to patch vulnerable firmware using binary patching (DisPatch). As a result, we have found 91 new bugs in modern RAV control programs, and their developers confirmed 32 cases and patch 11 cases.

<div>Driven by breakthroughs in mobile and IoT devices, on-device computation becomes promising. Meanwhile, there is a growing concern over its security: it faces many threats</div><div>in the wild, while not supervised by security experts; the computation is highly likely to touch users’ privacy-sensitive information. Towards trustworthy on-device computation, we present novel system designs focusing on two key applications: stream analytics, and machine learning training and inference.</div><div><br></div><div>First, we introduce Streambox-TZ (SBT), a secure stream analytics engine for ARM-based edge platforms. SBT contributes a data plane that isolates only analytics’ data and</div><div>computation in a trusted execution environment (TEE). By design, SBT achieves a minimal trusted computing base (TCB) inside TEE, incurring modest security overhead.</div><div><br></div><div>Second, we design a minimal GPU software stack (50KB), called GPURip. GPURip allows developers to record GPU computation ahead of time, which will be replayed later</div><div>on client devices. In doing so, GPURip excludes the original GPU stack from run time eliminating its wide attack surface and exploitable vulnerabilities.</div><div><br></div><div>Finally, we propose CoDry, a novel approach for TEE to record GPU computation remotely. CoDry provides an online GPU recording in a safe and practical way; it hosts GPU stacks in the cloud that collaboratively perform a dryrun with client GPU models. To overcome frequent interactions over a wireless connection, CoDry implements a suite of key optimizations.</div>

<div>As wearable devices, like smartwatches and fitness monitors, gain popularity and are being touted for clinical purposes, evaluating the resilience and security of wearable operating systems (OSes) and their corresponding ecosystems becomes essential. One of the most dominant OSes for wearable devices is Wear OS, created by Google. Wear OS and Android (its counterpart OS for mobile devices) share similar features, but the unique characteristics and uses of wearable devices posses new challenges. For example, wearable applications are generally more dependent on device sensors, have complex communication patterns (both intra-device and inter-device), and are context-aware. Current research efforts on the Wear OS are more focused on the efficiency and performance of the OS itself, overlooking the resilience or security of the OS or its ecosystem.</div><div> </div><div>This dissertation introduces a systematic analysis to evaluate the Wear OS's resilience and security. The work is divided into two main parts. First, we focus our efforts on developing novel tools to evaluate the robustness of the wearable OS and uncover vulnerabilities and failures in the wearable ecosystem. We provide an assessment and propose techniques to improve the system's overall reliability. Second, we turn our attention to the security and privacy of smart devices. We assess the privacy and security of highly interconnected devices. We demonstrate the feasibility of privacy attacks under these scenarios and propose a defense mechanism to mitigate these attacks.</div><div> </div><div>For the resilience part, we evaluate the overall robustness of the Wear OS ecosystem using a fuzz testing-based tool [DSN2018]. We perform an extensive fault injection study by mutating inter-process communication messages and UI events on a set of popular wearable and mobile applications. The results of our study show similarities in the root cause of failures between Wear OS and Android; however, the distribution of exception differ in both OSes. Further, our study evidence that input validation has improved in the Android ecosystem with respect to prior studies. Then, we study the impact of the state of a wearable device on the overall reliability of the applications running in Wear OS [MobiSys2020]. We use distinguishable characteristics of wearable apps, such as sensor activation and mobile-wearable communication patterns, to derive a state model and use this model to target specific fuzz injection campaigns against a set of popular wearable apps. Our experiments revealed an abundance of improper exception handling on wearable applications and error propagation across mobile and wearable devices. Furthermore, our results unveiled a flawed design of the wearable OS, which caused the device to reboot due to excessive sensor use.</div><div><br></div><div>For the security and privacy part, we assess user awareness toward privacy risks under scenarios with multiple interconnected devices. Our results show that a significant majority of the users have no reservation while granting permission to their devices. Furthermore, users tend to be more conservative while granting permission on their wearables. Based on the results of our study, we demonstrate the practicability of leaking sensitive information inferred from the user by orchestrating an attack using multiple devices. Finally, we introduce a tool based on NLP (Natural Language Processing) techniques that can aid the user in detecting this type of attack.</div>

<div>In a potentially hostile networked environment, a large diversity of security mechanisms with varying degree of sophistication are being deployed to protect valuable computer systems and digital assets. </div><div><br></div><div>While many competing implementations of similar security mechanisms are available in the current software development landscape, the robustness and reliability of such implementations are often overlooked, resulting in exploitable flaws in system deployments. In this dissertation, we systematically evaluate implementations of security mechanisms that are deployed in the wild. First, we examine how content distribution applications on the Android platform control access to their multimedia contents. With respect to a well-defined hierarchy of adversarial capabilities and attack surfaces, we find that many content distribution applications, including that of some world-renowned publications and streaming services, are vulnerable to content extraction due to the use of unjustified assumptions in their security mechanism designs and implementations. Second, we investigate the validation logic of X.509 certificate chains as implemented in various open-source TLS libraries. X.509 certificates are widely used in TLS as a means to achieve authentication. A validation logic that is overly restrictive could lead to the loss of legitimate services, while an overly permissive implementation could open door to impersonation attacks. Instead of manual analysis and unguided fuzzing, we propose a principled approach that leverages symbolic execution to achieve better coverage and uncover logical flaws that are buried deep in the code. We find that many TLS libraries deviate from the specification. Finally, we study the verification of RSA signatures, as specified in the PKCS#1 v1.5 standard, which is widely used in many security-critical network protocols. We propose an approach to automatically generate meaningful concolic test cases for this particular problem, and design and implement a provenance tracking mechanism to assist root-cause analysis in general. Our investigation revealed that several crypto and IPSec implementations are susceptible to new variants of the Bleichenbacher low-exponent signature forgery.</div>

WhatsApp is the most popular instant messaging application worldwide. Since 2016, users can send and receive messages through desktop clients, either through the WhatsApp desktop application or the web client accessible from supported web browsers. The author identified a gap in the literature in terms of WhatsApp forensics for desktop and web clients. The aim of the study was to locate forensic artifacts on WhatsApp clients. These clients included the desktop application on both Windows and Mac operating systems. Chrome and Firefox web clients were also analyzed for the Windows operating system, as well as Chrome and Safari web clients on the Mac operating system. A WhatsApp log file was identified as the main artifact providing information throughout all clients analyzed. Cached profile pictures were also found, as well as history information about visited websites and ran applications.

The Internet of Things is deploying large numbers of bare-metal systems that have no protection against memory corruption and control-flow hijacking attacks. These attacks have enabled unauthorized entry to hotel rooms, malicious control of unmanned aerial vehicles, and invasions of privacy. Using static and dynamic analysis these systems can utilize state-of-the-art testing techniques to identify and<br>prevent memory-corruption errors and employ defenses against memory corruption and control-flow hijacking attacks in bare-metal systems that match or exceed those currently employed on desktop systems. This is shown using three case studies.<br><br>(1) EPOXY which, automatically applies data execution prevention, diversity, stack defenses, and separating privileged code from unprivileged code using a novel<br>technique called privileged overlaying. These protections prevent code injection attacks, and reduce the number of privileged instruction to 0.06% verses an unprotected<br>application.<br><br>(2) Automatic Compartments for Embedded Systems (ACES), which automatically creates compartments that enforce data integrity and code isolation within bare-metal applications. ACES enables exploring policies to best meet security and performance requirements for individual applications. Results show ACES' can form 10s of compartments within a single thread and has a 15% runtime overhead on average.<br><br><div>(3) HALucinator breaks the requirement for specialized hardware to perform bare-metal system testing. This enables state-of-the-art testing techniques –e.g., coverage based fuzzing – to scale with the availability of commodity computers, leading to the discovery of exploitable vulnerabilities in bare-metal systems. <br></div><div><br></div><div>Combined, these case studies advance the security of embedded system several decades and provide essential protections for today’s connected devices.</div>

Many e-voting techniques have been proposed but not widely used in reality. One of the problems associated with most of existing e-voting techniques is the lack of transparency, leading to a failure to deliver voter assurance. In this work, we propose a transparent, auditable, end-to-end verifiable, and mutual restraining e-voting protocol that exploits the existing multi-party political dynamics such as in the US. The new e-voting protocol consists of three original technical contributions -- universal verifiable voting vector, forward and backward mutual lock voting, and in-process check and enforcement -- that, along with a public real time bulletin board, resolves the apparent conflicts in voting such as anonymity vs. accountability and privacy vs. verifiability. Especially, the trust is split equally among tallying authorities who have conflicting interests and will technically restrain each other. The voting and tallying processes are transparent to voters and any third party, which allow any voter to verify that his vote is indeed counted and also allow any third party to audit the tally. For the environment requiring receipt-freeness and coercion-resistance, we introduce additional approaches to counter vote-selling and voter-coercion issues. Our interactive voting protocol is suitable for small number of voters like boardroom voting where interaction between voters is encouraged and self-tallying is necessary; while our non-interactive protocol is for the scenario of large number of voters where interaction is prohibitively expensive. Equipped with a hierarchical voting structure, our protocols can enable open and fair elections at any scale.

<div> <div> <div> <p>Cloud computing offers a cost-efficient data analytics platform. This is enabled by constant innovations in tools and technologies for analyzing large volumes of data through distributed batch processing systems and real-time data through distributed stream processing systems. However, due to the sensitive nature of data, many organizations are reluctant to analyze their data in public clouds. To address this stalemate, both software-based and hardware-based solutions have been proposed yet all have substantial limitations in terms of efficiency, expressiveness, and security. In this thesis, we present solutions that enable practical and expressive confidentiality- preserving batch and stream-based analytics. We achieve this by performing computations over encrypted data using Partially Homomorphic Encryption (PHE) and Property-Preserving Encryption (PPE) in novel ways, and by utilizing remote or Trusted Execution Environment (TEE) based trusted services where needed.</p><p><br></p><p>We introduce a set of extensions and optimizations to PHE and PPE schemes and propose the novel abstraction of Secure Data Types (SDTs) which enables the application of PHE and PPE schemes in ways that improve performance and security. These abstractions are leveraged to enable a set of compilation techniques making data analytics over encrypted data more practical. When PHE alone is not expressive enough to perform analytics over encrypted data, we use a novel planner engine to decide the most efficient way of utilizing client-side completion, remote re-encryption, or trusted hardware re-encryption based on Intel Software Guard eXtensions (SGX) to overcome the limitations of PHE. We also introduce two novel symmetric PHE schemes that allow arithmetic operations over encrypted data. Being symmetric, our schemes are more efficient than the state-of-the-art asymmetric PHE schemes without compromising the level of security or the range of homomorphic operations they support. We apply the aforementioned techniques in the context of batch data analytics and demonstrate the improvements over previous systems. Finally, we present techniques designed to enable the use of PHE and PPE in resource-constrained Internet of Things (IoT) devices and demonstrate the practicality of stream processing over encrypted data.</p></div></div></div><div><div><div> </div> </div> </div>

<div>Comprehensive networking security is a goal to achieve for enterprise networks. In forensics, the traffic analysis, causality dependence in intricate program network flows is needed in flow-based attribution techniques. The provenance, the connection between stealthy advanced persistent threats (APTs) and the execution of loadable modules is stripped because loading a module does not guarantee an execution. The reports of common vulnerabilities and exposures (CVE) demonstrate that lots of vulnerabilities have been introduced in protocol engineering process, especially for the emerging Internet-of-Things (IoT) applications. A code generation framework targeting secure protocol implementations can substantially enhance security.</div><div>A novel automaton-based technique, NetCrop, to infer fine-grained program behavior by analyzing network traffic is proposed in this thesis. Based on network flow causality, it constructs automata that describe both the network behavior and the end-host behavior of a whole program to attribute individual packets to their belonging programs and fingerprint the high-level program behavior. A novel provenance-oriented library tracing system, Lprov, which enforces library tracing on top of existing syscall logging based provenance tracking approaches is investigated. With the dynamic library call stack, the provenance of implicit library function execution is revealed and correlated to system events, facilitating the locating and defense of malicious libraries. The thesis presents ProFactory, in which a protocol is modeled, checked and securely generated, averting common vulnerabilities residing in protocol implementations.</div>

Malicious URLs pose serious cyber-security threats to the Internet users. It is critical to detect malicious URLs so that they could be blocked from user access. In the past few years, several techniques have been proposed to differentiate malicious URLs from benign ones with the help of machine learning. Machine learning algorithms learn trends and patterns in a data-set and use them to identify any anomalies. In this work, we attempt to find generic features for detecting malicious URLs by analyzing two publicly available malicious URL data-sets. In order to achieve this task, we identify a list of substantial features that can be used to classify all types of malicious URLs. Then, we select the most significant lexical features by using Chi-Square and ANOVA based statistical tests. The effectiveness of these feature sets is then tested by using a combination of single and ensemble machine learning algorithms. We build a machine learning based real-time malicious URL detection system as a web service to detect malicious URLs in a browser. We implement a chrome extension that intercepts a browser’s URL requests and sends them to web service for analysis. We implement the web service as well that classifies a URL as benign or malicious using the saved ML model. We also evaluate the performance of our web service to test whether the service is scalable.