Target For Adversary Research Articles

Edge server enhanced secure and privacy preserving federated learning

Federated learning (FL) has been smoothly embedded into current IoT edge computing architecture, and hatching advanced IoT-FL applications. While so, sensitive messages generated on terminal sides brings privacy issues; and vulnerable terminal devices are easy targets for Byzantine adversaries, they inject malicious data to manipulate the IoT-FL system. Now an inherent dilemma is, the privacy preserving requires anonymous indistinguishable individual characteristics, while the malicious adversary detection requires transparent distinguishable results. Existing fragmented schemes are not able to handle both problems coherently. This time, inherit our previous study, based on the IoT edge computing architecture, we utilize Homomorphic Encryption (HE) and Threshold Secret Sharing (SS) methods, propose a joint scheme (namely, Sec-IoTFL) for anonymous adversary detection. Specifically, we batch encode the clients’ training result vectors (message) as polynomials, and split them as secret pieces with a unified SS key set; then edge server and cloud server joint deal with these secret pieces in a specialized flow, where Byzantine adversaries will be filtered out. The theoretical analysis and solid experimental results suggest, in our scheme, local sensitive data is well preserved, and malicious behavior is precisely screened. Comparing with traditional methods, our Sec-IoTFL scheme shows the superiority in accuracy and efficiency.

Securing Verified IO Programs Against Unverified Code in F*

We introduce SCIO*, a formally secure compilation framework for statically verified programs performing input-output (IO). The source language is an F* subset in which a verified program interacts with its IO-performing context via a higher-order interface that includes refinement types as well as pre- and post-conditions about past IO events. The target language is a smaller F* subset in which the compiled program is linked with an adversarial context that has an interface without refinement types, pre-conditions, or concrete post-conditions. To bridge this interface gap and make compilation and linking secure we propose a formally verified combination of higher-order contracts and reference monitoring for recording and controlling IO operations. Compilation uses contracts to convert the logical assumptions the program makes about the context into dynamic checks on each context-program boundary crossing. These boundary checks can depend on information about past IO events stored in the state of the monitor. But these checks cannot stop the adversarial target context before it performs dangerous IO operations. Therefore linking in SCIO* additionally forces the context to perform all IO actions via a secure IO library, which uses reference monitoring to dynamically enforce an access control policy before each IO operation. We prove in F* that SCIO* soundly enforces a global trace property for the compiled verified program linked with the untrusted context. Moreover, we prove in F* that SCIO* satisfies by construction Robust Relational Hyperproperty Preservation, a very strong secure compilation criterion. Finally, we illustrate SCIO* at work on a simple web server example.

Novel SMO-Based Detection and Isolation of False Data Injection Attacks Against Frequency Control Systems

The integration of modern power systems with information and communication technologies exposes them to various cyber threats. Load frequency control (LFC) is a communication-based automation in power systems that regulates the frequency of the grid. Its critical role makes it a highly attractive target for adversaries. This paper proposes a novel detection and isolation method of False Data Injection Attacks (FDIAs) against LFC. The defense method employs sliding mode observation techniques to detect FDIAs against LFC in real-time and discover which parts of the control loop have been compromised. Attacks are identified by comparing the generated residuals with a specific threshold that is designed in an adaptive manner. The proposed method is able to successfully distinguish the FDIAs from other system disturbances and is robust against uncertainties in power system parameters and noisy measurements. The effectiveness and scalability of the proposed defense method are confirmed on realistic power system models, considering nonlinearities, different topologies and diverse types of transmission links.

Hybrid Warfare on Pakistan: Examining the Determinants and Impact

Hybrid warfare, an ambiguous term in academia, has recently gained traction in scholarly work. However, its manifestation appears too vague in terms of observable phenomena. In the case of Pakistan, hybrid warfare has become more of a cliché or jargon. The nation contends with hybrid warfare on all fronts. Due to political instability and fragile governance, it has become an easy target for adversaries to employ hybrid warfare against, especially since its nuclearization. While this constant use of hybrid warfare may have made Pakistan resilient on the military front, it has had a far-reaching impact on Pakistani society. From the human development index to economic development, Pakistan has undergone a phase marked by a significant loss of peace and harmony. The research paper explores these implications in detail to foster an understanding of its impacts on Pakistan.

Magnets to Adversaries—An Analysis of the Attacks on Public Cloud Servers

Security adversaries are always constantly looking for targets to exploit. The mechanism of exploitation used by security adversaries varies significantly. Many focus on easy compromises as mere pivots to extend their attacks from these exploited systems to continue accomplishing their original goals. The cloud environment is a highly susceptible target for adversaries and provides a solid mechanism for observing adversary behavior. The sheer volume of attacks on the cloud provides insights into the attacker’s objectives and attack patterns, which can be leveraged for protecting infrastructure. This work deep dives into the practices used by adversaries on the commonly exposed protocols in the Amazon Web Services (AWS), Microsoft Azure (Azure), Google Cloud Platform (GCP), and Oracle Cloud Infrastructure (OCI) platforms. A robust honeypot model is documented that compares attacker behavior across various ports and protocols running in multiple cloud environments. This work illustrates that adversary activity is highly versatile in the public cloud environment, with an average of 700 new and unique IP addresses found attacking honeypot infrastructure daily. Further, this article illustrates the security safeguards a typical organization can leverage to mitigate the threats from these adversaries constantly probing insecure targets on the cloud platform.

Challenging Nuclear Bromides

Despite the universal implications of nuclear weapons for humanity, broad knowledge of nuclear policy is curiously absent from the public sphere and even among many senior figures in government. Nuclear strategists have nourished the belief that the principles of nuclear deterrence are so esoteric as to exempt them from non-expert scrutiny, creating an intellectual vacuum within which a number of fallacies about nuclear strategy have taken root. These fallacies in turn find expression in the configuration of several states’ nuclear postures, from the number of warheads posited as necessary for deterrence to the selection of adversary targets. Puncturing the myth of complexity that surrounds nuclear policy is a necessary starting point to assess its more questionable features. This essay evaluates some of the prevailing tenets of nuclear strategy to expose for non-experts the many misconceptions that shape the major powers’ strategic postures.

SRoUDA: Meta Self-Training for Robust Unsupervised Domain Adaptation

As acquiring manual labels on data could be costly, unsupervised domain adaptation (UDA), which transfers knowledge learned from a rich-label dataset to the unlabeled target dataset, is gaining increasingly more popularity. While extensive studies have been devoted to improving the model accuracy on target domain, an important issue of model robustness is neglected. To make things worse, conventional adversarial training (AT) methods for improving model robustness are inapplicable under UDA scenario since they train models on adversarial examples that are generated by supervised loss function. In this paper, we present a new meta self-training pipeline, named SRoUDA, for improving adversarial robustness of UDA models. Based on self-training paradigm, SRoUDA starts with pre-training a source model by applying UDA baseline on source labeled data and taraget unlabeled data with a developed random masked augmentation (RMA), and then alternates between adversarial target model training on pseudo-labeled target data and fine-tuning source model by a meta step. While self-training allows the direct incorporation of AT in UDA, the meta step in SRoUDA further helps in mitigating error propagation from noisy pseudo labels. Extensive experiments on various benchmark datasets demonstrate the state-of-the-art performance of SRoUDA where it achieves significant model robustness improvement without harming clean accuracy.

CANova: A hybrid intrusion detection framework based on automatic signal classification for CAN

Over the years, vehicles have become increasingly complex and an attractive target for malicious adversaries. This raised the need for effective and efficient Intrusion Detection Systemss (IDSs) for onboard networks able to work with the stringent requirements and the heterogeneity of information transmitted on the Controller Area Network. While state-of-the-art solutions are effective in detecting specific types of anomalies and work on a subset of the CAN signals, no single method can perform better than the others on all types of attacks, particularly if they need to provide predictions to comply with the domain’s real-time constraints. In this paper, we present CANova, a modular framework that exploits the characteristics of the different Controller Area Network (CAN) packets to select the Intrusion Detection Systemss (IDSs) that better fits them. In particular, it uses flow- and payload-based IDSs to analyze the packets’ content and arrival time. We evaluate CANova by comparing its performance against state-of-the-art Intrusion Detection Systemss (IDSs) for in-vehicle network and a comprehensive set of synthetic and real attacks in real-world CAN datasets. We demonstrate that our approach can achieve good performances in terms of detection, false positive rates, and temporal performances.

Membership Inference Attacks Against Temporally Correlated Data in Deep Reinforcement Learning

While significant research advances have been made in the field of deep reinforcement learning, there have been no concrete adversarial attack strategies in literature tailored for studying the vulnerability of deep reinforcement learning algorithms to membership inference attacks. In such attacking systems, the adversary targets the set of collected input data on which the deep reinforcement learning algorithm has been trained. To address this gap, we propose an adversarial attack framework designed for testing the vulnerability of a state-of-the-art deep reinforcement learning algorithm to a membership inference attack. In particular, we design a series of experiments to investigate the impact of temporal correlation, which naturally exists in reinforcement learning training data, on the probability of information leakage. Moreover, we compare the performance of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">collective</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">individual</i> membership attacks against the deep reinforcement learning algorithm. Experimental results show that the proposed adversarial attack framework is surprisingly effective at inferring data with an accuracy exceeding 84% in individual and 97% in collective modes in three different continuous control Mujoco tasks, which raises serious privacy concerns in this regard. Finally, we show that the learning state of the reinforcement learning algorithm influences the level of privacy breaches significantly.

A Novel Approach to Reduce Breaches of Aircraft Communication Data

Aircraft are complex systems that rely heavily on monitoring and real-time communications with the base station. During aviation and flight operations, diverse data are gathered from different sources, including the Cockpit Voice Recorder (CVR), Flight Data Recorder (FDR), logbook, passenger data, passenger manifest etc. Given the high sensitivity of flight data, it is an attractive target for adversaries which could result in operational, financial and safety related incidents. Communications between aircraft pilots and air traffic controllers are all unencrypted. The data, mainly audio communication files, are placed openly within data centers on the ground stations which could lead to a serious compromise in security and privacy. One may rely on the cloud owing to its on-demand features but to thwart possible attacks, the data need to be encrypted first, giving rise to the issue of conducting search over encrypted data. This research presents a novel approach for data security in aviation industry by introducing a semantic-based searchable encryption scheme over the cloud. The designed system has proven to be extraordinarily effective for semantic-based searchable encryption at the word and the text level. The rigorous security and complexity analysis shows that the proposed solution provides a high level of security and efficiency and can be effectively deployed in the aviation sector. The designed scheme is tested through a real-world aviation dataset collected to demonstrate the significance of this research. The proof of concept proves to be secure, privacy-preserving and lightweight while resisting distinguishability attacks.

DNSguard: A Raspberry Pi-Based DDoS Mitigation on DNS Server in IoT Networks

IoT development has increased the likelihood of several security attacks, such as Mirai, VPNFilter, and DNS (Domain Name System) flooding attacks. The DNS is a censorious component of smart home IoT infrastructure, a soft target for adversaries to launch DNS flooding attacks and impact poor QoS. This letter proposes DNSguard, a MUD-enabled two-level DNS flooding attack mitigation framework for IoT networks. DNSguard authenticates the behavioral profiles of Manufacturer Usage Description (MUD) compliant IoT devices and monitors traffic to detect and mitigate DNS flooding attacks at the local DNS server. The solution has been implemented using online DNS traffic in Raspberry Pi, and quantified response time decreased to 67.2%.

Transferable Adversarial Attacks against Automatic Modulation Classifier in Wireless Communications

Deep neural network-based automatic modulation recognition (AMR) technology has become an increasingly important area due to the advantages of self-extraction of features and high identification accuracy. Based on the view of security threats to machine learning classifiers, we investigate the influence of adversarial samples on the AMR model in this paper. The traditional method is based on label gradient attack without taking advantage of the feature-level transferability, resulting in the attack effect that is not perfect. So, we exploit the feature-level transferability property that could be met to fulfill realistic imperceptibility and transfer needs. In this paper, firstly, we proposed an AMR scheme with high recognition accuracy as our attack model. Secondly, we proposed a transferable attack method based on a feature gradient-based, which increases perturbation to clean signal based on features space. Finally, we introduce a new attack strategy, in which we select two original and one adversarial target signal sample as the input of triplet loss to achieve higher attack strength and high transferability. Meanwhile, this paper proposes indicators of signal characteristics to test the effectiveness of our proposed attack method. Based on experimental results, our proposed feature gradient-based adversarial attack method outperforms the currently labeled gradient attack methods regarding attack effectiveness and transferability.

Data poisoning attacks against machine learning algorithms

For the past decade, machine learning technology has increasingly become popular and it has been contributing to many areas that have the potential to influence the society considerably. Generally, machine learning is used by various industries to enhance their performances. Moreover, machine learning algorithms are used to solve some hard problems of systems that may contain very critical information. This makes machine learning algorithms a target of adversaries, which is an important problem for systems that use such algorithms. Therefore, it is significant to determine the performance and the robustness of a machine learning algorithm against attacks. In this paper, we analyze empirically the robustness and performances of six machine learning algorithms against two types of adversarial attacks by using four different datasets and three metrics. In our experiments, we analyze the robustness of Support Vector Machine, Stochastic Gradient Descent, Logistic Regression, Random Forest, Gaussian Naive Bayes, and K-Nearest Neighbor algorithms to create learning models. We observe their performances in spam, botnet, malware, and cancer detection datasets when we launch adversarial attacks against these environments. We use data poisoning for manipulating training data during adversarial attacks, which are random label flipping and distance-based label flipping attacks. We analyze the performance of each algorithm for a specific dataset by modifying the amount of poisoned data and analyzing behaviors of accuracy rate, f1-score, and AUC score. Analyses results show that machine learning algorithms have various performance results and robustness under different adversarial attacks. Moreover, machine learning algorithms are affected differently in each stage of an adversarial attacks. Furthermore, the behavior of a machine learning algorithm highly depends on the type of the dataset. On the other hand, some machine learning algorithms have better robustness and performance results against adversarial attacks for almost all datasets.

Synthesis of strategies for autonomous surveillance on adversarial targets

We study the problem of synthesizing a controller for an agent with imperfect sensing and a quantitative surveillance objective, that is, an agent is required to maintain knowledge of the location of a moving, possibly adversarial target. We formulate the problem as a one-sided partial-information game with a winning condition expressed as a temporal logic specification. The specification encodes the quantitative surveillance requirement as well as any additional tasks. Solving a partial-information game typically involves transforming it into a perfect-information belief game using a belief-set construction. Such a transformation leads to a state-space explosion, rendering the belief game computationally intractable to solve for most realistic settings. We present a belief-set abstraction technique to transform the partial-information game to a provably sound abstract belief game that can be solved efficiently using off-the-shelf reactive synthesis tools. We introduce a counterexample-guided refinement approach to automatically achieve the abstraction precision sufficient to synthesize a strategy that is provably winning on the original partial-information game. We evaluate the proposed method on multiple case-studies, implemented on hardware as well as high-fidelity ROS/Gazebo simulations where the agent must respond in real-time to a human-controlled adversary.

Defense-in-depth against insider attacks in cyber-physical systems

In a cyber-physical system (CPS) built on Internet-of-things (IoT) technologies, whenever measurement and control signals are transferred over communication networks across cyber and physical systems, it potentially becomes a target for adversaries. The problem becomes especially serious if the adversaries are insiders. A single layer of defense may not be strong enough in such a case, as it is difficult to assess the extent of knowledge that the inside attackers may have known about the physical and cyber system configurations, communication networks/protocols, and their respective vulnerabilities. Hence, it is paramount to have a reliable and fail-safe defense-in-depth architecture to fence off would-be-attackers. In this paper, a multi-layer defense-in-depth approach has been developed. For an inside attacker with legitimate access, the first line of defense, such as access control, may have already been compromised. Given this fact, the focus of the current paper has been on detection and mitigation. Both data-driven and model-based techniques are considered to catch stealthy attacks and stop them in their tracks. Effective mitigation techniques can then be deployed to minimize the adverse effects. To demonstrate this design philosophy and validate the effectiveness of the developed methodologies, a lab-scale cyber-physical system platform based on industry-grade communication networks and physical sensors has been used for validation.

Secure IIoT-Enabled Industry 4.0

The Industrial Internet of things (IIoT) is the main driving force behind smart manufacturing, industrial automation, and industry 4.0. Conversely, industrial IoT as the evolving technological paradigm is also becoming a compelling target for cyber adversaries. Particularly, advanced persistent threats (APT) and especially botnets are the foremost promising and potential attacks that may throw the complete industrial IoT network into chaos. IIoT-enabled botnets are highly scalable, technologically diverse, and highly resilient to classical and conventional detection mechanisms. Subsequently, we propose a deep learning (DL)-enabled novel hybrid architecture that can efficiently and timely tackle distributed, multivariant, lethal botnet attacks in industrial IoT. The proposed approach is thoroughly evaluated on a current state-of-the-art, publicly available dataset using standard performance evaluation metrics. Moreover, our proposed technique has been precisely verified with our constructed hybrid DL-enabled architectures and current benchmark DL algorithms. Our devised mechanism shows promising results in terms of high detection accuracy with a trivial trade-off in speed efficiency, assuring the proposed scheme as an optimal and legitimate cyber defense in prevalent IIoTs. Besides, we have cross-validated our results to show utterly unbiased performance.

MalDroid: Secure DL‐enabled intelligent malware detection framework

Nowadays, smartphones are provided with an abundance of capabilities. During the last decade, the availability of smartphone users and online mobile payment services and applications have substantially grown. Besides, the Android infotainment market is exponentially growing and thus potentially becoming a primary target for cyber adversaries and attackers. Likewise, varied Android vulnerability exploitation and targeted pervasive malware sophisticated attacks are also becoming a hot spot for both industry and academia. The authors present a secure by design efficient and intelligent Android detection framework against prevalent, sophisticated and persistent malware threats and attacks. A novel and highly proficient Cuda-enabled multi-class malware threat detection and identification Deep Learning (DL)-driven mechanism that leverages ConvLSTM2D and CNN has been proposed. The devised approach has been extensively evaluated on publicly available state-of-the-art datasets of Android applications (i.e. Android Malware Dataset (AMD), Androzoo). Standard and extended assessment metrics have been employed to thoroughly evaluate the proposed technique. Moreover, the performance of the proposed algorithm has been verified both with the constructed hybrid DL-driven algorithms and current benchmarks. Additionally, the proposed scheme is cross validated to explicitly show unbiased results.

Last Line of Defense: Reliability Through Inducing Cyber Threat Hunting With Deception in SCADA Networks

There exists a gap between existing security mechanisms and their ability to detect advancing threats. Antivirus and EDR (End Point Detection and Response) aim to detect and prevent threats; such security mechanisms are reactive. This approach did not prove to be effective in protecting against stealthy attacks. SCADA (Supervisory Control and Data Acquisition) security is crucial for any country. However, SCADA is always an easy target for adversaries due to a lack of security for heterogeneous devices. An attack on SCADA is mainly considered a national-level threat. Recent research on SCADA security has not considered "unknown threats," which has left a gap in security. The proactive approach, such as threat hunting, is the need of the hour. In this research, we investigated that threat hunting in conjunction with cyber deception and kill chain has countervailing effects on detecting SCADA threats and mitigating them. We have used the concept of "decoy farm" in the SCADA network, where all attacks are engaged. Moreover, we present a novel threat detection and prevention approach for SCADA, focusing on unknown threats. To test the effectiveness of approach, we emulated several Linux and Windows-based attacks on a simulated SCADA network. We have concluded that our approach detects and prevents the attacker before using the current reactive approach and security mechanism for SCADA with enhanced protection for heterogeneous devices. The results and experiments show that the proposed threat hunting approach has significantly improved the threat detection ability.

Robustness of Speech Spoofing Detectors Against Adversarial Post-Processing of Voice Conversion

With the development of speech synthesis and voice conversion techniques, the quality of artificially generated speech has been significantly improved and detecting such spoofing speech becomes crucial to practical applications, such as automatic speaker verification (ASV). State-of-the-art neural-network-based spoofing detection models can distinguish most artificial utterances from natural ones effectively in the latest ASVspoof 2019 evaluation. Motivated by recent progresses of adversarial example generation, this paper studies the robustness of neural-network-based speech spoofing detectors against adversarial attacks. To this end, an adversarial post-processing network (APN) is proposed which generates adversarial examples against a white-box anti-spoofing model by post-processing the speech waveforms produced by a baseline voice conversion system. Experimental results demonstrate the adversarial ability of our proposed APNs against the white-box anti-spoofing models which were used as the adversarial targets of APNs at the training stage. For example, the equal error rate (EER) of a fused detection model based on light convolution neural networks (LCNNs) increased from 0.278% to 12.743% under the white-box condition without degrading the subjective quality of converted speech. Furthermore, the trained APNs can also perform against the detectors with either unseen structures or unseen features by raising their EERs in our experiments. All these results indicate the threat of adversarial speech generation to the performance of state-of-the-art spoofing detection models.

Value of data meets IT security – assessing IT security risks in data-driven value chains

Digitalization forces manufacturing companies to shift towards customer-oriented, highly data-driven forms of value creation. This results in a changing IT security risk landscape as data becomes an attractive target for adversaries leading to an increasing number of attacks. In order to successfully protect data, it is essential that it is assessed in an integrated manner. Although IT security and data-based value creation have been studied by large research bodies, the existing literature fails to provide guidance on IT security risk analysis in data-based value chains. To contribute to the closure of this research gap, we propose a modeling approach which allocates different data types to value activities and analyses the data types in relation to the properties of relevant IT security risks. The evaluation, conducted with industry experts, reveals that it is not only a company’s primary assets that are of concern but also less important data types subject to significant levels of exposure that bear considerable IT security risks.