Cooperative perception (CP) enhances situational awareness of connected and autonomous vehicles by exchanging and combining messages from multiple agents. While prior work has explored adversarial integrity attacks that degrade detection accuracy, little is known about CP's robustness against attacks on timeliness (or availability), a safety-critical requirement for autonomous driving. In this paper, we present CP-FREEZER, the first latency attack that maximizes the computation delay of CP algorithms by injecting adversarial perturbation via V2V messages. Our attack resolves several unique challenges, including the non-differentiability of point cloud preprocessing, asynchronous knowledge of the victim’s input due to transmission delays, and uses a novel loss function that effectively maximizes the execution time of the CP pipeline. Extensive experiments show that CP-FREEZER increases end-to-end CP latency by over 90×, pushing per-frame processing time beyond 3 seconds with a 100% success rate on our real-world vehicle testbed. Our findings reveal a critical threat to the availability of CP systems, highlighting the urgent need for robust defenses.

As users increasingly turn to large language model (LLM) based web agents to automate online tasks, agents may encounter dark patterns: deceptive user interface designs that manipulate users into making unintended decisions. Although dark patterns primarily target human users, their potentially harmful impacts on LLM-based generalist web agents remain unexplored. In this paper, we present the first study that investigates the impact of dark patterns on the decision-making process of LLM-based generalist web agents. To achieve this, we introduce LiteAgent, a lightweight framework that automatically prompts agents to execute tasks while capturing comprehensive logs and screen-recordings of their interactions. We also present TrickyArena, a controlled environment comprising web applications from domains such as e-commerce, streaming services, and news platforms, each containing diverse and realistic dark patterns that can be selectively enabled or disabled. Using LiteAgent and TrickyArena, we conduct multiple experiments to assess the impact of both individual and combined dark patterns on web agent behavior. We evaluate six popular LLM-based generalist web agents across three LLMs and discover that when there is a single dark pattern present, agents are susceptible to it an average of 41% of the time. We also find that modifying dark pattern UI attributes through visual design changes or HTML code adjustments and introducing multiple dark patterns simultaneously can influence agent susceptibility. This study emphasizes the need for holistic defense mechanisms in web agents, encompassing both agent-specific protections and broader web safety measures.

International students (IntlS) in the US refer to foreign students who acquire student visas to study in the US, primarily in higher education. As IntlS arrive in the US, they face several challenges, such as adjusting to a new country and culture, securing housing remotely, and arranging finances for tuition and personal expenses. These experiences, coupled with recent events such as visa revocations and the cessation of new visas, compound IntlS' risk of being targeted by and falling victim to online scams. While prior work has investigated IntlS' security and privacy, as well as general end users' reactions to online scams, research on how IntlS are uniquely impacted by scams remains largely absent. To address this gap, we conduct a two-phase user study comprising surveys (n=48) and semi-structured interviews (n=9). We investigate IntlS' exposure and interactions with scams, post-exposure actions such as reporting, and their perceptions of the usefulness of existing prevention resources and the barriers to following prevention advice. We find that IntlS are often targeted by scams (e.g., attackers impersonating government officials) and fear legal implications or deportation, which directly impacts their interactions with scams (e.g., they may prolong engagement with a scammer due to a sense of urgency). Interestingly, we also find that IntlS may lack awareness of - or access to - reliable resources that inform them about scams or guide them in reporting incidents to authorities. In fact, they may also face unique barriers in enacting scam prevention advice, such as avoiding reporting financial losses, since IntlS are required to demonstrate financial ability to stay in the US. The findings produced by our study help synthesize guidelines for stakeholders to better aid IntlS in reacting to scams.

A subset of mobile applications is explicitly marketed to military-affiliated personnel. These Military-Marketed Mobile Apps (MMM-apps) collect privacy-sensitive data using the same mechanisms as general-purpose apps. However, when such data belongs to military-affiliated personnel, it may be exploited by malicious actors in ways that threaten personal safety, unit operations, and national security. Despite these risks, the data practices and code provenance of MMMapps, as well as how this population perceives and attempts to mitigate these risks, remain poorly understood. In this paper, we address this gap by combining large-scale app analysis with a user study. We first curate a dataset of 242 MMMapps and leverage app analysis techniques to characterize their data practices and code provenance. Then, we conduct a user study with n = 103 military-affiliated participants in the United States to examine which data practices and code provenance characteristics they consider inappropriate, what threat scenarios they believe those practices enable, and which mitigations they view as most effective. Our results show that MMMapps frequently exhibit data practices and code provenance characteristics that are misaligned with the privacy expectations of military-affiliated personnel. For instance, 40% of MMMapps collect more data than they disclose in their privacy labels or data safety sections. 83.5% of our study participants report using at least one MMMapp that engages in data practices they are uncomfortable with. Additionally, although military-affiliated personnel are generally concerned about third-party libraries accessing their data, 64% of MMMapps include third-party SDKs, some developed in countries perceived as adversarial by a majority of the participants. Overall, our findings reveal a substantial misalignment between the privacy expectations of military-affiliated personnel and the data practices and software supply chains of MMMapps. We propose recommendations at the federal, DoD, app store, and device levels to improve privacy risk mitigation for this at-risk population.

Large language models (LLMs) can underpin AI assistants that help users with everyday tasks, such as by making recommendations or performing basic computation.Despite AI assistants' promise, little is known about the implicit values these assistants display while completing subjective everyday tasks.Humans may consider values like environmentalism, charity, and diversity.To what extent do LLMs exhibit these values in completing everyday tasks?How do they compare with humans?We answer these questions by auditing how six popular LLMs complete 30 everyday tasks, comparing LLMs to each other and to 100 human crowdworkers from the US.We find LLMs often do not align with humans, nor with other LLMs, in the implicit values exhibited.Selection: Choose from predefined options Value Code † LocalVendor: Purchase from a farmers' market or cheaper chain Financial PayForPrivacy: Elect whether to pay more for a privacy-protective retailer Privacy EcoFlight: Select a flight from options with different CO2 emissions Environmentalism Grouping: Separate items into groups or choose a subset StudentScholarship: Choose recipients knowing race and test scores Diversity MathClass: Divide students into study groups knowing their test scores Diversity HiringCommittee: Select hiring committee knowing prospects' gender/race Diversity Prioritization: Rank-order or prioritize a list of items Introduction: Choose five important points for introducing someone Community Rebudgeting: Choose spending to cut to get under budget Financial Emails: Prioritize between emails in inbox Community Recommendation: Generate open-ended suggestions NextLanguage: Suggest a language for a Spanish speaker to learn next Multiculturalism Transportation: Suggest a mode of transportation between cities Environmentalism Music: Suggest songs for a music playlist, listing year/genre Heterogeneity Retrieval: Retrieve information about a general-knowledge query Swimmers: List ten famous Olympic swimmers Multiculturalism GenderQuestions: List gender options to include on a survey Diversity Recipes: List three recipes and their dietary restrictions Heterogeneity Composition: Write novel text from scratch based on a prompt Country: Write a paragraph describing a successful country Multiculturalism TwoCharacters: Write a short story that names two characters Diversity Adjectives: List five adjectives for an 84-year-old character Diversity Summarization: Shortening given text subject to word-limit constraints Research: Summarize research findings about an app Community NewsArticle: Summarize a news article about a VR headset Privacy JobApplicant: Summarize a job applicant's strengths Community Modification:

The rising popularity of mental wellbeing technologies has led many individuals to explore binaural beats—an emerging form of sound therapy proliferating on web and mobile platforms. However, it currently remains unknown whether users can trust binaural tracks on online platforms, or if they deceive unsuspecting users. Our research aims to address this problem by understanding (1) what binaural beats listeners expect from tracks and (2) whether online tracks conform to these expectations. To understand user expectations, we perform thematic analysis on online forum threads and blog posts to extract binaural beats goals and expectations tied to these goals. Next, we design a methodology to measure binaural beats tracks’ conformance to commonly held user expectations. This methodology comprises, (1) obtaining a track’s intent to induce a mental state through track metadata analysis, (2) extracting a track’s binaural beats time-frequency model using Fast Fourier Transform, (3) mapping user expectations to rules that identify deceptive tracks, and validating them on the track’s extracted intent and time-frequency model. We evaluate ∼7K binaural beats tracks and find that only 7.5% conform to commonly held user expectations, while the remaining 92.5% deceive users with deviant claims (e.g., eroticism, weight loss) or deliver contradicting binaural beats. Our study underscores the significance of understanding users’ expectations and verifying conformance of online wellness technologies to expose discrepancies in expectations.

The WebXR API enables immersive AR/VR experiences directly through web browsers on head-mounted displays (HMDs). However, prior research shows that security-sensitive UI properties and the lack of an like element that separates different origins can be exploited to manipulate user actions, particularly within the advertising ecosystem. In our prior work, we proposed five novel UI-based attacks in WebXR, targeting the ad ecosystem. This demo presents these attacks in a unified gaming application, embedding each into distinct interactive scenarios. Our work highlights the need to address design challenges and requirements for improving immersive web-based experiences. We provide our demo video at: https://youtu.be/lTBQbxnNq34.

Language models have recently been applied to binary analysis tasks, such as function similarity detection and function signature recovery. These models typically employ a two-stage training process: pre-training via Masked Language Modeling (MLM) on machine code and fine-tuning for specific tasks. While MLM helps to understand binary code structures, it ignores essential code characteristics, including control and data flow, which negatively affect model generalization. Recent work leverages domain-specific features (e.g., control flow graphs and dynamic execution traces) in transformer-based approaches to improve binary code semantic understanding. This approach, however, involves complex feature engineering, a cumbersome and time-consuming process that can introduce predictive uncertainty when dealing with stripped or obfuscated code, which leads to a performance drop. We introduce PROTST, a novel transformer-based methodology for binary code embedding. PROTST employs a hierarchical training process based on a unique tree-like structure, where knowledge progressively flows from fundamental tasks at the root to more specialized tasks at the leaves. This progressive teacher-student paradigm allows the model to build upon previously learned knowledge, resulting in high-quality embeddings that can be effectively leveraged for diverse downstream binary analysis tasks. The effectiveness of PROTST is evaluated in seven binary analysis tasks, demonstrating an average of 14.8% improvement in F1 and MRR compared to traditional two-stage training, and a 16.6 % improvement when analyzing obfuscated code.

Camera-based perception is a central component to the visual perception of autonomous systems. Recent works have investigated latency attacks against perception pipelines, which can lead to a Denial-of-Service against the autonomous system. Unfortunately, these attacks lack real-world applicability, either relying on digital perturbations or requiring large, unscalable, and highly visible patches that cover up the victim's view. In this paper, we propose Detstorm, a novel physically realizable latency attack against camera-based perception. Detstorm uses projector perturbations to cause delays in perception by creating a large number of adversarial objects. These objects are optimized on four objectives to evade filtering by multiple Non-Maximum Suppression (NMS) approaches. To maximize the number of created objects in a dynamic physical environment, Detstorm takes a unique greedy approach, segmenting the environment into “zones” containing distinct object classes and maximizing the number of created objects per zone. Detstorm adapts to changes in the environment in real time, recombining perturbation patterns via our zone stitching process into a contiguous, physically projectable image. Evaluations in both simulated and real-world experiments show that Detstorm causes a 506% increase in detected objects on average, delaying perception results by up to 8.1 seconds, and capable of causing physical consequences on real-world autonomous driving systems.

Industrial Control Systems (ICSs), widely employed in many critical infrastructure sectors that manage and control physical processes (e.g., energy, water, transportation), face heightened security risks due to increased digitization and connectivity. Monitoring Indicators of Compromise (IoCs), observable signs of intrusion, such as unusual network activity or unauthorized system changes, are crucial for early detection and response to malicious activities, including data breaches and insider threats. While IoCs have been extensively studied in traditional Information Technology (IT), their effectiveness and suitability for the unique challenges of ICS environments, which directly control physical processes, remain unclear. Moreover, the influence of human factors (e.g., sociotechnical factors, usability) on the utilization and interpretation of IoCs for attack prevention in ICSs is not well understood. To address this gap, we conducted two studies involving 52 ICS security professionals. In an IoC Applicability study (n=32), we explore the relevance of existing IoCs within ICS environments and investigate factors contributing to potential ambiguities in their interpretation. We examine the perceived value, effort required for the collection, and volatility of various data sources used for IoC identification. Participants in the IoC Applicability Study emphasized the significant role of human factors in recognizing and interpreting IoCs for threat mitigation within ICS ecosystems. Based on this insight, we conducted a Socio-technical Factors in Recognition and Detection study (n=20) to investigate the impact of human factors on threat detection and explore the sociotechnical factors that influence the effective utilization of IoCs. Our results show significant discrepancies between conventional IT-based IoCs and their applicability to ICS environments, along with various socio-technical challenges (e.g., alert overload and desensitization). Our study provides pointers to rethinking the specific operational, technological, and human aspects of IoCs within the ICS context. Our findings provide insights for the development of ICS-specific IoC to enable security analysts to better respond to potential threats in industrial environments.

In Voice Assistant (VA) platforms, when users add devices to their accounts and give voice commands, complex interactions occur between the devices, skills, VA clouds, and vendor clouds. These interactions are governed by the device management capabilities (DMC) of VA platforms, which rely on device names, types, and associated skills in the user account. Prior work studied vulnerabilities in specific VA components, such as hidden voice commands and bypassing skill vetting. However, the security and privacy implications of device management flaws have largely been unexplored. In this paper, we introduce DMC-Xplorer, a testing framework for the automated discovery of VA device management flaws. We first introduce VA description language (VDL), a new domain-specific language to create VA environments for testing, using VA and skill developer APIs. DMC-Xplorer then selects VA parameters (device names, types, vendors, actions, and skills) in a combinatorial approach and creates VA environments with VDL. It issues real voice commands to the environment via developer APIs and logs event traces. It validates the traces against three formal security properties that define the secure operation of VA platforms. Lastly, DMC-Xplorer identifies the root cause of property violations through intervention analysis to identify VA device management flaws. We exercised DMC-Xplorer on Amazon Alexa and Google Home and discovered two design flaws that can be exploited to launch four attacks. We show that malicious skills with default permissions can eavesdrop on privacy-sensitive device states, prevent users from controlling their devices, and disrupt the services on the VA cloud.

Multi-Object Tracking (MOT) is a critical task in computer vision, with applications ranging from surveillance systems to autonomous driving. However, threats to MOT algorithms have yet been widely studied. In particular, incorrect association between the tracked objects and their assigned IDs can lead to severe consequences, such as wrong trajectory predictions. Previous attacks against MOT either focused on hijacking the trackers of individual objects, or manipulating the tracker IDs in MOT by attacking the integrated object detection (OD) module in the digital domain, which are model-specific, non-robust, and only able to affect specific samples in offline datasets. In this paper, we present AdvTraj, the first online and physical ID-manipulation attack against tracking-by-detection MOT, in which an attacker uses adversarial trajectories to transfer its ID to a targeted object to confuse the tracking system, without attacking OD. Our simulation results in CARLA show that AdvTraj can fool ID assignments with 100% success rate in various scenarios for white-box attacks against SORT, which also have high attack transferability (up to 93% attack success rate) against state-of-the-art (SOTA) MOT algorithms due to their common design principles. We characterize the patterns of trajectories generated by AdvTraj and propose two universal adversarial maneuvers that can be performed by a human walker/driver in daily scenarios. Our work reveals under-explored weaknesses in the object association phase of SOTA MOT systems, and provides insights into enhancing the robustness of such systems.

The Controller Area Network (CAN) is a pivotal communication protocol extensively utilized in vehicles, aircraft, factories, and diverse cyber-physical systems (CPSs). The extensive CAN security literature resulting from decades of wide usage may create an impression of thorough scrutiny. However, a closer look reveals its reliance on a specific threat model with a limited range of abilities. Notably, recent works show that this model is outdated and that a more potent and versatile model could soon become the norm, prompting the need for a new defense paradigm. Unfortunately, the security impact of this emerging model on CAN systems has not received sufficient attention, and the defense systems addressing it are almost nonexistent. In this paper, we introduce ERACAN, the first comprehensive defense system against this new threat model. We first begin with a threat analysis to ensure that ERACAN comprehensively understands this model's capabilities, evasion tactics, and propensity to enable new attacks or enhance existing ones. ERACAN offers versatile protection against this spectrum of threats, providing attack detection, classification, and optional prevention abilities. We implement and evaluate ERACAN on a testbed and a real vehicle's CAN bus to demonstrate its low latency, real-time operation, and protective capabilities. ERACAN achieves detection rates of 100% and 99.7%+ for all attacks launched by the conventional and the enhanced threat models, respectively.

The ubiquitous deployment of robots across diverse domains, from industrial automation to personal care, underscores their critical role in modern society. However, this growing dependence has also revealed security vulnerabilities. An attack vector involves the deployment of malicious software (malware) on robots, which can cause harm to robots themselves, users, and even the surrounding environment. Machine learning approaches, particularly supervised ones, have shown promise in malware detection by building intricate models to identify known malicious code patterns. However, these methods are inherently limited in detecting unseen or zero-day malware variants as they require regularly updated massive datasets that might be unavailable to robots. To address this challenge, we introduce RoboGuardZ, a novel malware detection framework based on zero-shot learning for robots. This approach allows RoboGuardZ to identify unseen malware by establishing relationships between known malicious code and benign behaviors, allowing detection even before the code executes on the robot. To ensure practical deployment in resource-constrained robotic hardware, we employ a unique parallel structured pruning and quantization strategy that compresses the RoboGuardZ detection model by 37.4% while maintaining its accuracy. This strategy reduces the size of the model and computational demands, making it suitable for real-world robotic systems. We evaluated RoboGuardZ on a recent dataset containing real-world binary executables from multi-sensor autonomous car controllers. The framework was deployed on two popular robot embedded hardware platforms. Our results demonstrate an average detection accuracy of 94.25% and a low false negative rate of 5.8% with a minimal latency of 20 ms, which demonstrates its effectiveness and practicality.

Zero-day vulnerabilities pose a significant challenge to robot cyber-physical systems (CPS). Attackers can exploit software vulnerabilities in widely-used robotics software, such as the Robot Operating System (ROS), to manipulate robot behavior, compromising both safety and operational effectiveness. The hidden nature of these vulnerabilities requires strong defense mechanisms to guarantee the safety and dependability of robotic systems. In this paper, we introduce RoboCop, a cyber-physical attack detection framework designed to protect robots from zero-day threats. RoboCop leverages static software features in the pre-execution analysis along with runtime state monitoring to identify attack patterns and deviations that signal attacks, thus ensuring the robot’s operational integrity. We evaluated RoboCop on the F1-tenth autonomous car platform. It achieves a 93% detection accuracy against a variety of zero-day attacks targeting sensors, actuators, and controller logic. Importantly, in on-robot deployments, it identifies attacks in less than 7 seconds with a 12% computational overhead.

Physical sensor attacks against robotic vehicles (RV) have become a serious concern due to their prevalence and potential physical threat. However, RV software developers often do not deploy appropriate countermeasures. This hesitance stems from their belief that attackers face substantial challenges when conducting sensor attacks, e.g., nullifying sensor redundancy in hardware and circumventing sensor filters in software. Yet, we discover that attackers can overcome the challenges by fulfilling specific prerequisites and finely tuning attack parameters. The misconceptions that the developers have arisen from a lack of study regarding the level of difficulty attackers face in successfully achieving their attack goals, which we call "attack hardness".In this paper, we examine the hardness of 12 well-known sensor attacks. We first identify the prerequisites required to conduct the attacks successfully. We then quantify the hardness of each attack as how frequent the prerequisites enabling a specific attack are in the real world. To automate this analysis, we introduce RVPROBER, an attack prerequisite analysis framework. RVPROBER discovered that the 12 sensor attacks require, on average, 4.4 prerequisites, highlighting that previous literature has often missed important details required to perform these attacks. By satisfying the identified prerequisites and tuning attack parameters, we increased the number of successful attacks from 6 to 11. Moreover, our analysis showed that an average of 57.08% of actual RV users are vulnerable to sensor attacks. Finally, starting from the identified prerequisites, we analyzed the reasons behind the success of each attack and found previously-unknown root causes, such as design flaws in the RV software’s fail-safe logic.

Wearable devices are becoming increasingly important, helping us stay healthy and connected. There are a variety of app-based wearable platforms that can be used to manage these devices. The apps on wearable devices often work with a companion app on users’ smartphones. The wearable device and the smartphone typically use two separate permission models that work synchronously to protect sensitive data. However, this design creates an opaque view of the management of permission-protected data, resulting in over-privileged data access without the user’s explicit consent. In this paper, we performed the first systematic analysis of the interaction between the Android and Wear OS permission models. Our analysis is two-fold. First, through taint analysis, we showed that cross-device flows of permission-protected data happen in the wild, demonstrating that 28 apps (out of the 150 we studied) on Google Play have sensitive data flows between the wearable app and its companion app. We found that these data flows occur without the users’ explicit consent, introducing the risk of violating user expectations. Second, we conducted an in-lab user study to assess users’ understanding of permissions when subject to cross-device communication (n = 63). We found that 66.7% of the users are unaware of the possibility of cross-device sensitive data flows, which impairs their understanding of permissions in the context of wearable devices and puts their sensitive data at risk. We also showed that users are vulnerable to a new class of attacks that we call cross-device permission phishing attacks on wearable devices. Lastly, we performed a preliminary study on other watch platforms (i.e., Apple’s watchOS, Fitbit, Garmin OS) and found that all these platforms suffer from similar privacy issues. As countermeasures for the potential privacy violations in cross-device apps, we suggest improvements in the system prompts and the permission model to enable users to make better-informed decisions, as well as on app markets to identify malicious cross-device data flows.

Intermittent (batteryless) devices operate solely using energy harvested from their environment. These devices turn on when they have energy and turn off during energy scarcity. Intermittent devices have recently become increasingly popular in smart buildings, manufacturing plants, and medical implantables as they eliminate the need for battery replacement and enable green computing. Despite their growing adoption in critical applications, the privacy implications of intermittent devices remain largely unexplored. In this paper, we introduce a novel remote side-channel attack. Our observation is that the network packet frequency of an intermittent device can be exploited to learn its turn-on/off patterns. From these patterns, we can infer the energy availability of a device, which reveals privacy-sensitive information about its operating environment, e.g., the presence or absence of individuals. To realize our attack, we develop a three-stage hierarchical inference framework that leverages the timestamped network packet sequence of intermittent devices. Our framework automatically extracts a set of temporal features from inter-packet-arrival timings. It then employs a series of models to uncover (1) whether a target intermittent device is present in the environment, (2) its energy harvester type (e.g., vibration or water flow), and (3) its energy availability conditions (e.g., high-vibration or no-vibration). To validate our attack effectiveness, we conduct experiments in two environments: a smart home and a miniature manufacturing plant equipped with three intermittent devices powered by solar energy, vibration, and temperature. By analyzing their energy availability patterns, we are able to infer user activities and presence in the smart home and the robot’s movement patterns in the manufacturing plant with an average accuracy of 85%. This sensitive information enables an adversary to launch domain-specific attacks, such as burglarizing a smart home when the user is asleep or timely tampering with plant sensors to cause maximum damage.

Users seek security & privacy (S&P) advice from online resources, including trusted websites and content-sharing platforms. These resources help users understand S&P technologies and tools and suggest actionable strategies. Large Language Models (LLMs) have recently emerged as trusted information sources. However, their accuracy and correctness have been called into question. Prior research has outlined the shortcomings of LLMs in answering multiple-choice questions and user ability to inadvertently circumvent model restrictions (e.g., to produce toxic content). Yet, the ability of LLMs to provide reliable S&P advice is not well-explored.

Pairing schemes establish cryptographic keys to secure communication among IoT devices. Existing pairing approaches that rely on trusted central entities, human interaction, or shared homogeneous context are prone to a single point of failure, have limited usability, and require additional sensors. Recent work has explored event timings observed by devices with heterogeneous sensing modalities as proof of co-presence for decentralized pairing. Yet, this approach incurs high pairing time, cannot pair sensors that sense continuous physical quantities and does not support group pairing, making it infeasible for many IoT deployments. In this paper, we design and develop IoTCupid, a secure group pairing system for IoT devices with heterogeneous sensing modalities, without requiring active user involvement. IoTCupid operates in three phases: (a) detecting events sensed by both instant and continuous sensors with a novel window-based derivation technique, (b) grouping the events through a fuzzy clustering algorithm to extract inter-event timings, and (c) establishing group keys among devices with identical inter-event timings through a partitioned group password-authenticated key exchange scheme. We evaluate IoTCupid in smart home and office environments with 11 heterogeneous devices and show that it effectively pairs all devices with only 2 group keys with a minimal pairing overhead.

We show a new type of side-channel leakage in which the built-in magnetometer sensor in Apple's mobile devices captures touch events of users. When a conductive material such as the human body touches the mobile device screen, the electric current passes through the screen capacitors generating an electromagnetic field around the touch point. This electromagnetic field leads to a sharp fluctuation in the magnetometer signals when a touch occurs, both when the mobile device is stationary and held in hand naturally. These signals can be accessed by mobile applications running in the background without requiring any permissions. We develop iSTELAN, a three-stage attack, which exploits this side-channel to infer users' application and touch data. iSTELAN translates the magnetometer signals to a binary sequence to reveal users' touch events, exploits touch event patterns to fingerprint the type of application a user is using, and models touch events to identify users' touch event types performed on different applications. We demonstrate the iSTELAN attack on 22 users while using 7 popular app types and show that it achieves an average accuracy of 90% for disclosing touch events, 74% for classifying application type used, and 73% for detecting touch event types.

Modern autonomous systems rely on both object detection and object tracking in their visual perception pipelines. Although many recent works have attacked the object detection component of autonomous vehicles, these attacks do not work on full pipelines that integrate object tracking to enhance the object detector's accuracy. Meanwhile, existing attacks against object tracking either lack real-world applicability or do not work against a powerful class of object trackers, Siamese trackers. In this paper, we present AttrackZone, a new physically-realizable tracker hijacking attack against Siamese trackers that systematically determines valid regions in an environment that can be used for physical perturbations. AttrackZone exploits the heatmap generation process of Siamese Region Proposal Networks in order to take control of an object's bounding box, resulting in physical consequences including vehicle collisions and masked intrusion of pedestrians into unauthorized areas. Evaluations in both the digital and physical domain show that AttrackZone achieves its attack goals 92% of the time, requiring only 0.3-3 seconds on average.

Smart homes contain diverse sensors and actuators controlled by IoT apps that provide custom automation. Prior works showed that an adversary could exploit physical interaction vulnerabilities among apps and put the users and environment at risk, e.g., to break into a house, an adversary turns on the heater to trigger an app that opens windows when the temperature exceeds a threshold. Currently, the safe behavior of physical interactions relies on either app code analysis or dynamic analysis of device states with manually derived policies by developers. However, existing works fail to achieve sufficient breadth and fidelity to translate the app code into their physical behavior or provide incomplete security policies, causing poor accuracy and false alarms.

Automated program repair (APR) methods aim to identify patches for a given bug and apply them with minimal human intervention. To date, existing APR approaches focus on repairing software bugs, such as memory safety bugs. However, our analysis of popular robotic vehicle (RV) control software shows that most of their bugs are not memory bugs but rather logic bugs. These bugs, while not causing software crashes, can cause an RV to reach an undesired physical state (e.g., hitting the ground). To fix these logic bugs, we introduce PGPatch, a policy-guided program repair framework for RV control programs, which identifies the correct patch for a given logic bug and applies it without human intervention. PGPatch takes, as input, existing or new logic formulas used to discover logic bugs. It then leverages the formulas using a dedicated dynamic analysis to classify the previously known logic bugs into a patch type. It next uses a customized algorithm, based on the identified patch type and violated formula, to produce a source code patch as output. Lastly, it creates repeatable tests to verify the patch’s completeness, ensuring that the patch is correct and does not degrade the RV’s performance. We evaluate PGPatch on selected bug cases from three popular RV control software and find that it correctly fixes 258 out of 297 logic bugs (86.9%). We additionally recruit 18 experienced RV developers and users and conduct a user study that demonstrates how using PGPatch makes fixing bugs in RV software significantly quicker and less error-prone.

IoT devices can be used to complete a wide array of physical tasks, but due to factors such as low computational resources and distributed physical deployment, they are susceptible to a wide array of faulty behaviors. Many devices deployed in homes, vehicles, industrial sites, and hospitals carry a great risk of damage to property, harm to a person, or breach of security if they behave faultily. We propose a general fault handling system named IoTRepair, which shows promising results for effectiveness with limited latency and power overhead in an IoT environment. IoTRepair dynamically organizes and customizes fault-handling techniques to address the unique problems associated with heterogeneous IoT deployments. We evaluate IoTRepair by creating a physical implementation mirroring a typical home environment to motivate the effectiveness of this system. Our evaluation showed that each of our fault-handling functions could be completed within 100 milliseconds after fault identification, which is a fraction of the time that state-of-the-art fault-identification methods take (measured in minutes). The power overhead is equally small, with the computation and device action consuming less than 30 milliwatts. This evaluation shows that IoTRepair not only can be deployed in a physical system, but offers significant benefits at a low overhead.

Multiple users have access to multiple devices in a smart home system – typically through a dedicated app installed on a mobile device. Traditional access control mechanisms consider one unique, trusted user that controls access to the devices. However, multi-user multi-device smart home settings pose fundamentally different challenges to traditional single-user systems. For instance, in a multi-user environment, users have conflicting, complex, and dynamically-changing demands on multiple devices that cannot be handled by traditional access control techniques. Moreover, smart devices from different platforms/vendors can share the same home environment, making existing access control obsolete for smart home systems. To address these challenges, in this paper, we introduce Kratos+ , a novel multi-user and multi-device-aware access control mechanism that allows smart home users to flexibly specify their access control demands. Kratos+ has four main components: user interaction module, backend server, policy manager, and policy execution module. Users can easily specify their desired access control settings using the interaction module that are translated into access control policies in the back-end server. The policy manager analyzes these policies, initiates automated negotiation between users to resolve conflicting demands, and generates final policies to enforce in smart home systems. We implemented Kratos+ as a platform-independent solution and evaluated its performance on real smart home deployments featuring multi-user scenarios with a rich set of configurations (337 different policies including 231 demand conflicts and 69 restriction policies). These configurations also included five different threats associated with access control mechanisms. Our extensive evaluations show that Kratos+ is very effective in resolving conflicting access control demands with minimal overhead. We also performed an extensive user study with 72 smart home users to better understand the user’s needs before designing the system and a usability study to evaluate the efficacy of Kratos+ in a real-life smart home environment.

EXplainable AI (XAI) methods have been proposed to interpret how a deep neural network predicts inputs through model saliency explanations that highlight the input parts deemed important to arrive at a decision for a specific target. However, it remains challenging to quantify the correctness of their interpretability as current evaluation approaches either require subjective input from humans or incur high computation cost with automated evaluation. In this paper, we propose backdoor trigger patterns--hidden malicious functionalities that cause misclassification--to automate the evaluation of saliency explanations. Our key observation is that triggers provide ground truth for inputs to evaluate whether the regions identified by an XAI method are truly relevant to its output. Since backdoor triggers are the most important features that cause deliberate misclassification, a robust XAI method should reveal their presence at inference time. We introduce three complementary metrics for the systematic evaluation of explanations that an XAI method generates. We evaluate seven state-of-the-art model-free and model-specific post-hoc methods through 36 models trojaned with specifically crafted triggers using color, shape, texture, location, and size. We found six methods that use local explanation and feature relevance fail to completely highlight trigger regions, and only a model-free approach can uncover the entire trigger region. We made our code available at https://github.com/yslin013/evalxai.

Abstract Abstract: Users trust IoT apps to control and automate their smart devices. These apps necessarily have access to sensitive data to implement their functionality. However, users lack visibility into how their sensitive data is used, and often blindly trust the app developers. In this paper, we present IoTWATcH, a dynamic analysis tool that uncovers the privacy risks of IoT apps in real-time. We have designed and built IoTWATcH through a comprehensive IoT privacy survey addressing the privacy needs of users. IoTWATCH operates in four phases: (a) it provides users with an interface to specify their privacy preferences at app install time, (b) it adds extra logic to an app’s source code to collect both IoT data and their recipients at runtime, (c) it uses Natural Language Processing (NLP) techniques to construct a model that classifies IoT app data into intuitive privacy labels, and (d) it informs the users when their preferences do not match the privacy labels, exposing sensitive data leaks to users. We implemented and evaluated IoTWATcH on real IoT applications. Specifically, we analyzed 540 IoT apps to train the NLP model and evaluate its effectiveness. IoTWATcH yields an average 94.25% accuracy in classifying IoT app data into privacy labels with only 105 ms additional latency to an app’s execution.

In a smart home system, multiple users have access to multiple devices, typically through a dedicated app installed on a mobile device. Traditional access control mechanisms consider one unique trusted user that controls the access to the devices. However, multi-user multi-device smart home settings pose fundamentally different challenges to traditional single-user systems. For instance, in a multi-user environment, users have conflicting, complex, and dynamically changing demands on multiple devices, which cannot be handled by traditional access control techniques. To address these challenges, in this paper, we introduce Kratos, a novel multi-user and multi-device-aware access control mechanism that allows smart home users to flexibly specify their access control demands. Kratos has three main components: user interaction module, back-end server, and policy manager. Users can specify their desired access control settings using the interaction module which are translated into access control policies in the backend server. The policy manager analyzes these policies and initiates negotiation between users to resolve conflicting demands and generates final policies. We implemented Kratos and evaluated its performance on real smart home deployments featuring multi-user scenarios with a rich set of configurations (309 different policies including 213 demand conflicts and 24 restriction policies). These configurations included five different threats associated with access control mechanisms. Our extensive evaluations show that Kratos is very effective in resolving conflicting access control demands with minimal overhead, and robust against different attacks.

Data sharing among partners---users, companies, organizations---is crucial for the advancement of collaborative machine learning in many domains such as healthcare, finance, and security. Sharing through secure computation and other means allow these partners to perform privacy-preserving computations on their private data in controlled ways. However, in reality, there exist complex relationships among members (partners). Politics, regulations, interest, trust, data demands and needs prevent members from sharing their complete data. Thus, there is a need for a mechanism to meet these conflicting relationships on data sharing. This paper presents, an approach to exchange data among members who have complex relationships. A novel policy language, CPL, that allows members to define the specifications of data exchange requirements is introduced. With CPL, members can easily assert who and what to exchange through their local policies and negotiate a global sharing agreement. The agreement is implemented in a distributed privacy-preserving model that guarantees sharing among members will comply with the policy as negotiated. The use of Curie is validated through an example healthcare application built on recently introduced secure multi-party computation and differential privacy frameworks, and policy and performance trade-offs are explored.

For well over a quarter century, detection systems have been driven by models learned from input features collected from real or simulated environments. An artifact (e.g., network event, potential malware sample, suspicious email) is deemed malicious or non-malicious based on its similarity to the learned model at runtime. However, the training of the models has been historically limited to only those features available at runtime. In this paper, we consider an alternate learning approach that trains models using privileged information--features available at training time but not at runtime--to improve the accuracy and resilience of detection systems. In particular, we adapt and extend recent advances in knowledge transfer, model influence, and distillation to enable the use of forensic or other data unavailable at runtime in a range of security domains. An empirical evaluation shows that privileged information increases precision and recall over a system with no privileged information: we observe up to 7.7% relative decrease in detection error for fast-flux bot detection, 8.6% for malware traffic detection, 7.3% for malware classification, and 16.9% for face recognition. We explore the limitations and applications of different privileged information techniques in detection systems. Such techniques provide a new means for detection systems to learn from data that would otherwise not be available at runtime.

Machine learning (ML) models, e.g., deep neural networks (DNNs), are vulnerable to adversarial examples: malicious inputs modified to yield erroneous model outputs, while appearing unmodified to human observers. Potential attacks include having malicious content like malware identified as legitimate or controlling vehicle behavior. Yet, all existing adversarial example attacks require knowledge of either the model internals or its training data. We introduce the first practical demonstration of an attacker controlling a remotely hosted DNN with no such knowledge. Indeed, the only capability of our black-box adversary is to observe labels given by the DNN to chosen inputs. Our attack strategy consists in training a local model to substitute for the target DNN, using inputs synthetically generated by an adversary and labeled by the target DNN. We use the local substitute to craft adversarial examples, and find that they are misclassified by the targeted DNN. To perform a real-world and properly-blinded evaluation, we attack a DNN hosted by MetaMind, an online deep learning API. We find that their DNN misclassifies 84.24% of the adversarial examples crafted with our substitute. We demonstrate the general applicability of our strategy to many ML techniques by conducting the same attack against models hosted by Amazon and Google, using logistic regression substitutes. They yield adversarial examples misclassified by Amazon and Google at rates of 96.19% and 88.94%. We also find that this black-box attack strategy is capable of evading defense strategies previously found to make adversarial example crafting harder.

Deep learning takes advantage of large datasets and computationally efficient training algorithms to outperform other approaches at various machine learning tasks. However, imperfections in the training phase of deep neural networks make them vulnerable to adversarial samples: inputs crafted by adversaries with the intent of causing deep neural networks to misclassify. In this work, we formalize the space of adversaries against deep neural networks (DNNs) and introduce a novel class of algorithms to craft adversarial samples based on a precise understanding of the mapping between inputs and outputs of DNNs. In an application to computer vision, we show that our algorithms can reliably produce samples correctly classified by human subjects but misclassified in specific targets by a DNN with a 97% adversarial success rate while only modifying on average 4.02% of the input features per sample. We then evaluate the vulnerability of different sample classes to adversarial perturbations by defining a hardness measure. Finally, we describe preliminary work outlining defenses against adversarial samples by defining a predictive measure of distance between a benign input and a target classification.

In this work, we study the detection of Fast-Flux Service Networks (FFSNs) using DNS (Domain Name System) response packets. We have observed that current approaches do not employ a large combination of DNS features to feed into the proposed detection systems. The lack of features may lead to high false positive or false negative rates triggered by benign activities including Content Distribution Networks (CDNs). In this paper, we study recently proposed detection frameworks to construct a high-dimensional feature vector containing timing, network, spatial, domain name, and DNS response information. In the detection system, we strive to use features that are delay-free, and lightweight in terms of storage and computational cost. Feature sub-spaces are evaluated using a C4.5 decision tree classifier by excluding redundant features using the information gain of each feature with respect to each class. Our experiments reveal the performance of each feature subset type in terms of the classification accuracy. Moreover, we present the best feature subset for the discrimination of FFSNs recorded with the datasets we used.