Optimal Attack Research Articles

Dense Overload Subgraph Induced by Cyber–Physical Attacks in Smart Grid

With the tight couplings between the power grid and communication networks, smart grid is vulnerable to cyber-physical attacks, which can inject imperceptive perturbations to cause cascading failures. Nevertheless, as a dynamic and evolving process, it remains challenging to precisely lock the susceptible components that initiate a cascading event. To address this issue, we formulate a bi-level optimization problem in the context of cyber-physical attacks to model the attacker-operator adversarial game and generate an optimal attack strategy for electricity vulnerability assessments. Specifically, we merge the electrical overloading, topological proximity and outage correlations to construct a dense overload subgraph, which is critical for cascading propagation, by making use of fractional program to capture the density of subgraph with maximum average degree. We then compute exact solutions by adopting multiple linearization techniques to convert this bi-level fractional program into a mixed-integer linear program. Numerical studies in IEEE different-scale power systems verify the effectiveness of our model in understanding the cascading risk for a smart grid.

Improving the adversarial transferability with relational graphs ensemble adversarial attack.

In transferable black-box attacks, adversarial samples remain adversarial across multiple models and are more likely to attack unknown models. From this view, acquiring and exploiting multiple models is the key to improving transferability. For exploiting multiple models, existing approaches concentrate on differences among models but ignore the underlying complex dependencies. This exacerbates the issue of unbalanced and inadequate attacks on multiple models. To this problem, this paper proposes a novel approach, called Relational Graph Ensemble Attack (RGEA), to exploit the dependencies among multiple models. Specifically, we redefine the multi-model ensemble attack as a multi-objective optimization and create a sub-optimization problem to compute the optimal attack direction, but there are serious time-consuming problems. For this time-consuming problem, we define the vector representation of the model, extract the dependency matrix, and then equivalently simplify the sub-optimization problem by utilizing the dependency matrix. Finaly, we theoretically extend to investigate the connection between RGEA and the traditional multiple gradient descent algorithm (MGDA). Notably, combining RGEA further enhances the transferability of existing gradient-based attacks. The experiments using ten normal training models and ten defensive models on the labeled face in the wild (LFW) dataset demonstrate that RGEA improves the success rate of white-box attacks and further boosts the transferability of black-box attacks.

Optimal energy constrained deception attacks in cyber–physical systems with multiple channels: A fusion attack approach

This paper studies the issue of developing the optimal deception attacks on the multiple channels in cyber–physical systems, where the attackers are limited by energy constraints. To fully utilize the eavesdropped data, by linearly combining the innovations from the different channels, a fusion attack model is proposed under the stealthiness condition. According to the statistical characteristics of the correlated stochastic variables and the orthogonality principle, the state estimation error is quantified and analyzed by deriving the iteration of the error covariance matrices of the remote estimators under the proposed attack framework. Moreover, by analyzing the correlations of the decision variables in the objective function, it is shown that the attack parameters and energy allocation strategy can be derived by two steps without loss of optimality, such that the optimal attack scheme is acquired by solving a multivariate semi-definite programming (SDP) problem and a linear 0–1 programming problem respectively. Finally, simulation examples are provided to illustrate the effectiveness of the proposed method.

Optimal stealthy false data injection attacks for networked linear quadratic Gaussian control systems with two channels

SummaryIn this paper, a security problem in networked control systems which are modeled as linear quadratic Gaussian control systems is investigated from the perspective of attackers. We consider the scenario that the attackers can inject false data into two channels: sensor‐to‐controller and controller‐to‐actuator channels. The goal of attackers is to design the optimal stealthy false data injection attacks to maximize the linear quadratic Gaussian control cost function, where Kullback‐Leiber divergence is used as a stealthiness metric. Firstly, the cost function is divided into a fixed part and a part which is related to a remote estimation error covariance. Then, when the two channels are attacked by attackers, the recursion of remote estimation error covariance is derived. In the presence of attacks, the optimization problem with multiple decision variables, which may not be a convex problem, is transformed into a convex optimization problem and the corresponding optimal solution is obtained. In addition, an upper bound of benefit loss when using the optimal solution of the convex optimization problem is presented. Finally, a simulation example is presented to show the effectiveness of the obtained results.

An Optimal Combining Attack Strategy Against Economic Dispatch of Integrated Energy System

An optimal attack strategy combining denial-of-service (DoS) attacks and false data injection (FDI) attacks against the economic dispatch of integrated energy system (IES) is proposed in this brief. Firstly, the operation process of the distributed double consensus algorithm (DDCA) under DoS attacks is given, and the impact of DoS attacks on the economic dispatch of IES is qualitatively analyzed. Subsequently, an optimal stealthy DoS attack scheme consisting of the optimal attack sequence and the optimal attacked nodes is proposed. Furthermore, in order to enhance the economic destruction ability of DoS attacks, FDI attacks against measurement devices and communication links, respectively, are employed. The additional FDI attacks can minimize the power valley value in the optimal DoS attack scheme, and the explicit form of the vector to inject initial values is deduced. Finally, the proposed combining attack strategy that can maximize the operating cost is demonstrated by simulation cases on a 39–32 power-heat integrated energy system.

Statistical-Based Optimal $\epsilon$-Stealthy Attack Under Stochastic Communication Protocol: An Application to Networked Permanent Magnet Synchronous Machine Systems

This article presents the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\epsilon$</tex-math></inline-formula> -stealthy attack against cyber-physical systems under the stochastic communication protocol (SCP), which is used to avoid data collisions. The aim is to develop an optimal attack strategy to degrade the estimation performance maximally, while remaining a certain level of stealthiness characterized by the Kullback–Leibler divergence. Taking the SCP-introduced effects into account, a novel attack model is proposed, where only a part of the transmitted data is utilized to construct the attack signal at each time instant. Subsequently, the optimization problem is formulated. Different from the existing results, where only the numerical solution is given, the analytical expression of the optimal attack parameters is presented, where the analytical solution has more advantage in analyzing how the parameters affect the attack performance. Finally, the numerical simulations on a three-tank system and the experiments performed on a networked permanent magnet synchronous machine prototype system are provided to validate our results.

A Guessing Entropy-Based Framework for Deep Learning-Assisted Side-Channel Analysis

Recently deep-learning (DL) techniques have been widely adopted in side-channel power analysis. A DL-assisted SCA generally consists of two phases: a deep neural network (DNN) training phase and a follow-on attack phase using the trained DNN. However, currently the two phases are not well aligned, as there is no conclusion on what metric used in the training can result in the most effective attack in the second phase. When traditional loss functions such as negative log-likelihood (NLL) are used in training a DNN, the trained model does not yield optimal follow-on attack. Recently some information theoretical SCA leakage metrics are proposed, either as the validation metric to stop the DNN training with traditional loss functions, or as both the validation metric and the training loss function. None of those proposed metrics, however, directly measures the SCA effectiveness. We propose to conduct DNN training directly with a common SCA effectiveness metric, Guessing Entropy (GE). We overcome the prior practical difficulty of using GE in DNN training by utilizing the GEEA estimation algorithm introduced in CHES 2020. We show that using GEEA as either the validation metric or the loss function produces DNN models that lead to much more effective follow-on attacks. Our work consolidates the DL-assisted SCA framework with a consistent metric, which shows great potential to be adopted as the universal SCA-oriented DNN training framework.

Optimal Stealthy Joint Attacks against Distributed State Estimation in Cyber-Physical Systems

Optimal Stealthy Joint Attacks against Distributed State Estimation in Cyber-Physical Systems

Distributed Detection Over Blockchain-Aided Internet of Things in the Presence of Attacks

Distributed detection over a blockchain-aided Internet of Things (BIoT) network in the presence of attacks is considered, where the integrated blockchain is employed to secure data exchanges over the BIoT as well as data storage at the agents of the BIoT. We consider a general adversary model where attackers jointly exploit the vulnerability of IoT devices and that of the blockchain employed in the BIoT. The optimal attacking strategy which minimizes the Kullback-Leibler divergence is pursued. It can be shown that this optimization problem is nonconvex, and hence it is generally intractable to find the globally optimal solution to such a problem. To overcome this issue, we first propose a relaxation method that can convert the original nonconvex optimization problem into a convex optimization problem, and then the analytic expression for the optimal solution to the relaxed convex optimization problem is derived. The optimal value of the relaxed convex optimization problem provides a detection performance guarantee for the BIoT in the presence of attacks. In addition, we develop a coordinate descent algorithm which is based on a capped water-filling method to solve the relaxed convex optimization problem, and moreover, we show that the convergence of the proposed coordinate descent algorithm can be guaranteed.

THE PLACE OF OSINT IN THE CYBER KILL CHAIN

In recent years, there has been a trend towards an increase in the number of cyber attacks on organizations and individual users. In many cases, a key factor in the implementation of an information security incident is the attacker's effective preparation for a cyber attack: target selection, reconnaissance, i.e. obtaining any information that may be needed when planning an attack, weaponization based on discovered defense mechanisms, software and hardware, etc. and delivery, i.e. choosing how the malware will reach the victim and what steps will be required to activate it further. Having a significant amount of important and critical information for the organization from the point of view of ensuring security provides the attacker with the opportunity to choose the optimal attack scenario and significantly increases the chances of its success. The problem is that today's OSINT methods and tools allow you to find almost any information that is not protected in a real way, which significantly increases the risks, especially for organizations that find it difficult to control all the information that their employees post on social networks, disclose in interviews or accidentally enters the Internet. However, most intelligence tools are not only available to attackers, so ethical hackers and penetration testers can also use OSINT tools to examine an organization's vulnerabilities and improve its defenses before attackers exploit those vulnerabilities. The article examines the main methods of intelligence based on open sources, considers the most common and most often used OSINT tools, describes the life cycle of a cyber attack and defines the stages that require the use of OSINT tools when conducting an audit of the organization's information security and penetration tests.

Evaluating Criticality of Nodes in Consensus Network Under False Data Injection Attack

In this paper, a finite-duration, magnitude-bounded false data injection (FDI) attack on consensus network is considered. The aim of the attacker is to induce disagreement between nodes and consequently, influence the convergence of consensus algorithm. In order to measure the induced disagreement, a metric, namely induced terminal disagreement (ITD), is defined. The objective of this study is to determine the criticality of individual nodes in terms of the worst-case ITD resulting from attack on them. To achieve that, for every node, the closed-form expressions for the optimal attack input which results in the maximum ITD and the corresponding value of ITD, are obtained. Based on that, criticality ranks are assigned to all nodes. These ranks are beneficial in allocating security resources and designing resilient architecture. Further, the effect of varying attack duration on the worst-case ITDs and criticality ranks, is analyzed. Finally, it is shown that the criticality ranks of nodes have strong negative correlation with their degrees. A numerical example and simulations are presented to illustrate the proposed results.

An approach to generate damage strategies for inter-domain routing systems based on multi-objective optimization.

Inter-domain routing systems are important complex networks on the Internet. It has been paralyzed several times in recent years. The researchers pay close attention to the damage strategy of inter-domain routing systems and think it is related to the attacker's behavior. The key to the damage strategy is knowing how to select the optimal attack node group. In the process of selecting nodes, the existing research seldom considers the attack cost, and there are some problems, such as an unreasonable definition of attack cost and an unclear optimization effect. To solve the above problems, we designed an algorithm to generate damage strategies for inter-domain routing systems based on multi-objective optimization (PMT). We transformed the damage strategy problem into a double-objective optimization problem and defined the attack cost related to the degree of nonlinearity. In PMT, we proposed an initialization strategy based on a network partition and a node replacement strategy based on partition search. Compared with the existing five algorithms, the experimental results proved the effectiveness and accuracy of PMT.

Recursive reasoning-based training-time adversarial machine learning

The training process of a machine learning (ML) model may be subject to adversarial attacks from an attacker who attempts to undermine the test performance of the ML model by perturbing the training minibatches, and thus needs to be protected by a defender. Such a problem setting is referred to as training-time adversarial ML. We formulate it as a two-player game and propose a principled Recursive Reasoning-based Training-Time adversarial ML (R2T2) framework to model this game. R2T2 models the reasoning process between the attacker and the defender and captures their bounded reasoning capabilities (due to bounded computational resources) through the recursive reasoning formalism. In particular, we associate a deeper level of recursive reasoning with the use of a higher-order gradient to derive the attack (defense) strategy, which naturally improves its performance while requiring greater computational resources. Interestingly, our R2T2 framework encompasses a variety of existing adversarial ML methods which correspond to attackers (defenders) with different recursive reasoning capabilities. We show how an R2T2 attacker (defender) can utilize our proposed nested projected gradient descent-based method to approximate the optimal attack (defense) strategy at an arbitrary level of reasoning. R2T2 can empirically achieve state-of-the-art attack and defense performances on benchmark image datasets.

Asymmetric Vulnerability of Measurement and Control Channels in Closed-Loop Systems

This article studies the vulnerability of measurement and control channels in closed-loop systems. The problem of optimal stealthy attacks on linear quadratic Gaussian (LQG) control is analyzed. Different types of stealthiness are considered, including the probability distribution of innovations and their whiteness. For stealthy attacks on two different channels, the worst-case attacks that maximize LQG costs while satisfying different stealthiness constraints are obtained and compared. It is shown that worst-case attacks on the two channels have different properties, which depend on system parameters as well as the threshold of the relaxed stealthiness constraint. The asymmetric vulnerability of the two channels is illustrated by numerical examples.

Optimal deception attacks on remote state estimators equipped with interval anomaly detectors

This paper studies the problem of optimal deception attacks on remote state estimation, where an interval χ2 detector is deployed to reveal anomalies. The information-based attack policy that can bypass the anomaly detector and cause the maximum estimation quality degradation is derived. For both attacks with strict and relaxed stealthiness, the optimal compromised measurements can be designed with three steps: obtain the minimum mean-square error estimation of the prediction error, de-correlate the estimate with historical compromised innovations, and design the compromised innovation as an optimal linear transformation. All available information for attackers is fully utilized for performance maximization while the stealthiness constraint is satisfied precisely to deceive the anomaly detector. The attack effect depends on both the amount of online information and the duration of detection interval. Contrary to well-studied innovation-based attacks using static linear combinations, the information-based deception policy is shown to be generated by a linear time-varying system, whose coefficients can be completely determined offline. The optimality of the proposed attack is verified with numerical examples and comparative studies.

Cascaded Reinforcement Learning Agents for Large Action Spaces in Autonomous Penetration Testing

Organised attacks on a computer system to test existing defences, i.e., penetration testing, have been used extensively to evaluate network security. However, penetration testing is a time-consuming process. Additionally, establishing a strategy that resembles a real cyber-attack typically requires in-depth knowledge of the cybersecurity domain. This paper presents a novel architecture, named deep cascaded reinforcement learning agents, or CRLA, that addresses large discrete action spaces in an autonomous penetration testing simulator, where the number of actions exponentially increases with the complexity of the designed cybersecurity network. Employing an algebraic action decomposition strategy, CRLA is shown to find the optimal attack policy in scenarios with large action spaces faster and more stably than a conventional deep Q-learning agent, which is commonly used as a method for applying artificial intelligence to autonomous penetration testing.

Upper bound on device-independent quantum key distribution with two way classical postprocessing under individual attack

Device-independent quantum key distribution (DI-QKD) can guarantee the security even with untrusted devices. Unfortunately, conventional DI-QKD protocols can tolerate low noises only and require high detection efficiencies to achieve positive key rates. To improve the noise robustness, one promising solution is implementing the two-way classical postprocessing, which has the advantage of reducing the bit errors. In this paper, we study the DI-QKD with B-step two-way classical postprocessing under individual attacks. We adopt the tool of convex combination attack, i.e. an optimal individual attack, to upper bound the Devetak-Winter key rate. We show that, by using the B-step procedure, our protocol can tolerate detection efficiencies as low as and depolarising noise of 0.799, which is better than the thresholds for the protocol with one-way error correction. This result can serve as the lower bounds on the critical noise and detection efficiency for the scenarios under general attacks. Our work justifies the advantage of two-way classical postprocessing for DI-QKD, thus offering a step towards its applications.

Redundancy and Synergy of an Entangling Cloner in Continuous-Variable Quantum Communication

We address minimization of information leakage from continuous-variable quantum channels. It is known, that regime of minimum leakage can be accessible for the modulated signal states with variance equivalent to a shot noise, i.e., vacuum fluctuations, in the case of collective attacks. Here we derive the same condition for the individual attacks and analytically study the properties of the mutual information quantities in and out of this regime. We show that in such regime a joint measurement on the modes of a two-mode entangling cloner, being the optimal individual eavesdropping attack in a noisy Gaussian channel, is no more effective that independent measurements on the modes. Varying variance of the signal out of this regime, we observe the nontrivial statistical effects of either redundancy or synergy between the measurements of two modes of the entangling cloner. The result reveals the non-optimality of entangling cloner individual attack for sub-shot-noise modulated signals. Considering the communication between the cloner modes, we show the advantage of knowing the residual noise after its interaction with the cloner and extend the result to a two-cloner scheme.

Antibody production in mice requires neither vitamin D, nor the vitamin D receptor.

The vitamin D receptor as well as its ligand have been localized to various immune tissues and cells. These observations have led researchers to hypothesize a role for vitamin D in the immune system. However, a specific role for vitamin D in immunity has yet to be clearly delineated. The work in this report was undertaken to determine if mounting an antibody response is altered in the face of vitamin D-deficiency or when the signaling pathway is eliminated by removal of the nuclear receptor. This investigation provides direct evidence vitamin D is not necessary for producing antibodies, a process paramount for optimal attack against many foreign organisms. The idea that vitamin D plays a significant role in immunity has been proposed repeatedly for many years. To address this important idea we have carried out studies in mice to determine if vitamin D plays a significant role in antibody production. Two animal models were utilized: mice depleted of vitamin D and mice devoid of the vitamin D receptor. Further, a possible role of hypocalcemia resulting from vitamin D deficiency in antibody production was determined. Neither the absence of vitamin D or the vitamin D receptor nor hypocalcemia affected the ability of mice to mount an antibody response to an antigen challenge. Thus, we found no evidence that vitamin D or normal serum calcium is required for this major form of immunity.

Cyber-Attack Detection in Autonomous Vehicle Networks by Energy Aware Optimal Data Transmission with Game Fuzzy Q-Learning based Heuristic Routing Protocol

The automotive sector has seen a dramatic transition due to rapid technological advancement. Network connection has improved, enabling the transfer of the cars' technologies from being fully machine- to software-controlled. Controller area network (CAN) bus protocol manages network for autonomous vehicles. However, due to the intricacy of data and traffic patterns that facilitate unauthorised access to a can bus and many sorts of assaults, the autonomous vehicle network still has security flaws as well as vulnerabilities. This research proposes novel technique in cyber attack detection in autonomous vehicle networks enhanced data transmission based optimization and routing technique. Here the autonomous vehicle network optimal data transmission has been carried out using energy aware lagrangian multipliers based optimal data transmission. The cyber attack detection has been carried out using fuzzy q-learning based heuristic routing protocol. The experimental results has been carried out based on optimal data transmission and attack detection in terms of throughput of 95%, PDR of 94%, End-end delay of 46%, energy efficiency of 96%, network lifetime of 95%, attack detection rate of 88%.