Cyber security is a critical problem that causes data breaches, identity theft, and harm to millions of people and businesses. As technology evolves, new security threats emerge as a result of a dearth of cyber security specialists equipped with up-to-date information. It is hard for security firms to prevent cyber-attacks without the cooperation of senior professionals. However, by depending on artificial intelligence to combat cyber-attacks, the strain on specialists can be lessened. as the use of Artificial Intelligence (AI) can improve Machine Learning (ML) approaches that can mine data to detect the sources of cyberattacks or perhaps prevent them as an AI method, it enables and facilitates malware detection by utilizing data from prior cyber-attacks in a variety of methods, including behavior analysis, risk assessment, bot blocking, endpoint protection, and security task automation. However, deploying AI may present new threats, therefore cyber security experts must establish a balance between risk and benefit. While AI models can aid cybersecurity experts in making decisions and forming conclusions, they will never be able to make all cybersecurity decisions and judgments.

Anomaly detection is a challenge well-suited to machine learning and in the context of information security, the benefits of unsupervised solutions show significant promise. Recent attention to Graph Neural Networks (GNNs) has provided an innovative approach to learn from attributed graphs. Using a GNN encoder-decoder architecture, anomalous edges between nodes can be detected during the reconstruction phase. The aim of this research is to determine whether an unsupervised GNN model can detect anomalous network connections in a static, attributed network. Network logs were collected from four corporate networks and one artificial network using endpoint monitoring tools. A GNN-based anomaly detection system was designed and employed to score and rank anomalous connections between hosts. The model was validated against four realistic experimental scenarios against the four large corporate networks and the smaller artificial network environment. Although quantitative metrics were affected by factors including the scale of the network, qualitative assessments indicated that anomalies from all scenarios were detected. The false positives across each scenario indicate that this model in its current form is useful as an initial triage, though would require further improvement to become a performant detector. This research serves as a promising step for advancing this methodology in detecting anomalous network connections. Future work to improve results includes narrowing the scope of detection to specific threat types and a further focus on feature engineering and selection.

There will be a billion smart devices with processing, sensing, and actuation capabilities that can be connected to the Internet under the IoT paradigm. The level of convenience, effectiveness, and automation for consumers is expected to rise owing to promising IoT applications. Privacy is a significant concern in IoT systems, and it is essential to provide users with full awareness and control over the data collected by these systems. The use of privacy-enhancing technologies can help to minimise the risks associated with data collection and processing and ensure that user privacy is protected. Lack of standards for devices with limited resources and heterogeneous technologies intensifies the security issue. There are various emerging and existing technologies that can help to address the security risks in the IoT sector and achieve a high degree of trust in IoT applications. By implementing these technologies and countermeasures, it is possible to improve the security and reliability of IoT systems, ensuring that they can be used safely and effectively in a wide range of applications. This article s intent is to provide a comprehensive investigation of the threats and risks in the IoT industry and to examine some potential countermeasures.

An IC used in a safety-critical application such as automotive often requires a long lifetime of more than 10 years. Previously, stress test has been used as a means to establish the accelerated aging model for an IC product under a harsh operating condition. Then, the accelerated aging model is time-stretched to predict an IC’s normal lifetime. However, such a long-stretching prediction may not be very trustworthy. In this work, we present a more refined method to provide higher credibility in the IC lifetime prediction. We streamline in this paper a progressive lifetime prediction method with two phases – the training phase and the inference phase. During the training phase, we collect the aging histories of some training devices under various stress levels. During the inference phase, the extrapolation is performed on the “stressed lifetime” versus the “stress level” space and thereby leading to a more trustworthy prediction of the lifetime.

In the realm of Internet of Things (IoT) devices, the trust management system (TMS) has been enhanced through the utilisation of diverse machine learning (ML) classifiers in recent times. The efficacy of training machine learning classifiers with pre-existing datasets for establishing trustworthiness in IoT devices is constrained by the inadequacy of selecting suitable features. The current study employes a subset of the UNSW-NB15 dataset to compute additional features such as throughput, goodput, packet loss. These features may be combined with the best discriminatory features to distinguish between trustworthy and non-trustworthy IoT networks. In addition, the transformed dataset undergoes filter-based and wrapper-based feature selection methods to mitigate the presence of irrelevant and redundant features. The evaluation of classifiers is performed utilising diverse metrics, including accuracy, precision, recall, F1-score, true positive rate (TPR), and false positive rate (FPR). The performance assessment is conducted both with and without the application of feature selection methodologies. Ultimately, a comparative analysis of the machine learning models is performed, and the findings of the analysis demonstrate that our model s efficacy surpasses that of the approaches utilised in the existing literature.

IoT scenarios face cybersecurity concerns due to unauthorized devices that can impersonate legitimate ones by using identical software and hardware configurations. This can lead to sensitive information leaks, data poisoning, or privilege escalation. Behavioral fingerprinting and ML/DL techniques have been used in the literature to identify devices based on performance differences caused by manufacturing imperfections. In addition, using Federated Learning to maintain data privacy is also a challenge for IoT scenarios. Federated Learning allows multiple devices to collaboratively train a machine learning model without sharing their data, but it requires addressing issues such as communication latency, heterogeneity of devices, and data security concerns. In this sense, Trustworthy Federated Learning has emerged as a potential solution, which combines privacy-preserving techniques and metrics to ensure data privacy, model integrity, and secure communication between devices. Therefore, this work proposes a trustworthy federated learning framework for individual device identification. It first analyzes the existing metrics for trustworthiness evaluation in FL and organizes them into six pillars (privacy, robustness, fairness, explainability, accountability, and federation) for computing the trustworthiness of FL models. The framework presents a modular setup where one component is in charge of the federated model generation and another one is in charge of trustworthiness evaluation. The framework is validated in a real scenario composed of 45 identical Raspberry Pi devices whose hardware components are monitored to generate individual behavior fingerprints. The solution achieves a 0.9724 average F1-Score in the identification on a centralized setup, while the average F1-Score in the federated setup is 0.8320. Besides, a 0.6 final trustworthiness score is achieved by the model on state-of-the-art metrics, indicating that further privacy and robustness techniques are required to improve this score.

The prediction of human trust in machines within decision-aid systems is crucial for improving system performance. However, previous studies have only measured machine performance based on its decision history, failing to account for the machine’s current decision state. This delay in evaluating machine performance can result in biased trust predictions, making it challenging to enhance the overall performance of the human-machine system. To address this issue, this paper proposes incorporating machine estimated performance scores into a human-machine trust prediction model to improve trust prediction accuracy and system performance. We also provide an explanation for how this model can enhance system performance.To estimate the accuracy of the machine’s current decision, we employ the KNN(K-Nearest Neighbors) method and obtain a corresponding performance score. Next, we report the estimated score to humans through the human-machine interaction interface and obtain human trust via trust self-reporting. Finally, we fit the trust prediction model parameters using data and evaluate the model’s efficacy through simulation on a public dataset. Our ablation experiments show that the model reduces trust prediction bias by 3.6\% and significantly enhances the overall accuracy of human-machine decision-making.

In recent times, the research looks into the measures taken by financial institutions to secure their systems and reduce the likelihood of attacks. The study results indicate that all cultures are undergoing a digital transformation at the present time. The dawn of the Internet ushered in an era of increased sophistication in many fields. There has been a gradual but steady shift in attitude toward digital and networked computers in the business world over the past few years. Financial organizations are increasingly vulnerable to external cyberattacks due to the ease of usage and positive effects. They are also susceptible to attacks from within their own organisation. In this paper, we develop a machine learning based quantitative risk assessment model that effectively assess and minimises this risk. Quantitative risk calculation is used since it is the best way for calculating network risk. According to the study, a network s vulnerability is proportional to the number of times its threats have been exploited and the amount of damage they have caused. The simulation is used to test the model s efficacy, and the results show that the model detects threats more effectively than the other methods.

Malware detection constitutes a fundamental step in safe and secure computational systems, including industrial systems and the Internet of Things (IoT). Modern malware detection is based on machine learning methods that classify software samples as malware or benign, based on features that are extracted from the samples through static and/or dynamic analysis. State-of-the-art malware detection systems employ Deep Neural Networks (DNNs) whose accuracy increases as more data are analyzed and exploited. However, organizations also have significant privacy constraints and concerns which limit the data that they share with centralized security providers or other organizations, despite the malware detection accuracy improvements that can be achieved with the aggregated data. In this paper we investigate the effectiveness of federated learning (FL) methods for developing and distributing aggregated DNNs among autonomous interconnected organizations. We analyze a solution where multiple organizations use independent malware analysis platforms as part of their Security Operations Centers (SOCs) and train their own local DNN model on their own private data. Exploiting cross-silo FL, we combine these DNNs into a global one which is then distributed to all organizations, achieving the distribution of combined malware detection models using data from multiple sources without sample or feature sharing. We evaluate the approach using the EMBER benchmark dataset and demonstrate that our approach effectively reaches the same accuracy as the non-federated centralized DNN model, which is above 93\%.

IBMD(Intelligent Behavior-Based Malware Detection) aims to detect and mitigate malicious activities in cloud computing environments by analyzing the behavior of cloud resources, such as virtual machines, containers, and applications.The system uses different machine learning methods like deep learning and artificial neural networks, to analyze the behavior of cloud resources and detect anomalies that may indicate malicious activity. The IBMD system can also monitor and accumulate the data from various resources, such as network traffic and system logs, to provide a comprehensive view of the behavior of cloud resources. IBMD is designed to operate in a cloud computing environment, taking advantage of the scalability and flexibility of the cloud to detect malware and respond to security incidents. The system can also be integrated with existing security tools and services, such as firewalls and intrusion detection systems, to provide a comprehensive security solution for cloud computing environments.

Malwares have been being a major security threats to enterprises, government organizations and end-users. Beside traditional malwares, such as viruses, worms and trojans, new types of malwares, such as botnets, ransomwares, IoT malwares and crypto-jackings are released daily. To cope with malware threats, several measures for monitoring, detecting and preventing malwares have been developed and deployed in practice, such as signature-based detection, static and dynamic file analysis. This paper proposes 2 malware detection models based on statistics and machine learning using opcode n-grams. The proposed models aim at achieving high detection accuracy as well as reducing the amount of time for training and detection. Experimental results show that our proposed models give better performance measures than previous proposals. Specifically, the proposed statistics-based model is very fast and it achieves a high detection accuracy of 92.75\% and the random forest-based model produces the highest detection accuracy of 96.29\%.

The term Internet of Things(IoT) describes a network of real-world items, gadgets, structures, and other things that are equipped with communication and sensors for gathering and exchanging data online. The likelihood of Android malware attacks on IoT devices has risen due to their widespread use. Regular security precautions might not be practical for these devices because they frequently have limited resources. The detection of malware attacks on IoT environments has found hope in ML approaches. In this paper, some machine learning(ML) approaches have been utilized to detect IoT Android malware threats. This method uses a collection of Android malware samples and good apps to build an ML model. Using the Android Malware dataset, many ML techniques, including Naive Bayes (NB), K-Nearest Neighbour (KNN), Decision Tree (DT), and Random Forest (RF), are used to detect malware in IoT. The accuracy of the DT model is 95\%, which is the highest accuracy rate, while that of the NB, KNN, and RF models have accuracy rates of 84\%, 89\%, and 92\%, respectively.

With the development of network technologies, network intrusion has become increasing complex which makes the intrusion detection challenging. Traditional intrusion detection algorithms detect intrusion traffic through intrusion traffic characteristics or machine learning. These methods are inefficient due to the dependence of manual work. Therefore, in order to improve the efficiency and the accuracy, we propose an intrusion detection method based on deep learning. We integrate the Transformer and LSTM module with intrusion detection model to automatically detect network intrusion. The Transformer and LSTM can capture the temporal information of the traffic data which benefits to distinguish the abnormal data from normal data. We conduct experiments on the publicly available NSL-KDD dataset to evaluate the performance of our proposed model. The experimental results show that the proposed model outperforms other deep learning based models.

Network intrusion detection is a crucial task in ensuring the security and reliability of computer networks. In recent years, machine learning algorithms have shown promising results in identifying anomalous activities indicative of network intrusions. In the context of intrusion detection systems, novelty detection often receives limited attention within machine learning communities. This oversight can be attributed to the historical emphasis on optimizing performance metrics using established datasets, which may not adequately represent the evolving landscape of cyber threats. This research aims to compare four widely used novelty detection algorithms for network intrusion detection, namely SGDOneClassSVM, LocalOutlierDetection, EllipticalEnvelope Covariance, and Isolation Forest. Our experiments with the UNSW-NB15 dataset show that Isolation Forest was the best-performing algorithm with an F1-score of 0.723. The result shows that network-based intrusion detection systems are still challenging for novelty detection algorithms.

In the face of a large number of network attacks, intrusion detection system can issue early warning, indicating the emergence of network attacks. In order to improve the traditional machine learning network intrusion detection model to identify the behavior of network attacks, improve the detection accuracy and accuracy. Convolutional neural network is used to construct intrusion detection model, which has better ability to solve complex problems and better adaptability of algorithm. In order to solve the problems such as dimension explosion caused by input data, the albino PCA algorithm is used to extract data features and reduce data dimensions. For the common problem of convolutional neural networks in intrusion detection such as overfitting, Dropout layers are added before and after the fully connected layer of CNN, and Sigmoid is selected as the intrusion classification prediction function. This reduces the overfitting, improves the robustness of the intrusion detection model, and enhances the fault tolerance and generalization ability of the model to improve the accuracy of the intrusion detection model. The effectiveness of the proposed method in intrusion detection is verified by comparison and analysis of numerical examples.

The use of computers and the internet has spread rapidly over the course of the past few decades. Every day, more and more peopleare coming to rely heavily on the internet. When it comes to the field of information security, the subject of security is one that is becoming an increasingly important focus. It is vital to design a powerful intrusion detection system in order to prevent computer hackers and other intruders from effectively getting into computer networks or systems. This can be accomplished by: (IDS). The danger and attack detection capabilities of the computer system are built into the intrusion detection system. Abuse has occurred and can be used to identify invasions when there is a deviation between a preset pattern of intrusion and an observedpattern of intrusion. An intrusion detection system (IDS) is a piece of hardware (or software) that is used to generate reports for a Management Station as well as monitor network and/or system activities for unethical behaviour or policy violations. In the current study, an approach known as machine learning is suggested as a possible paradigm for the development of a network intrusion detection system. The results of the experiment show that the strategy that was suggested improves the capability of intrusion detection.

Intelligent environments rely heavily on the Internet of Things, which can be targeted by malicious attacks. Therefore, the autonomous capabilities of agents in intelligent health-care environments, and the agents’ characteristics (accuracy, reliability, efficiency and responsiveness), should be exploited to devise an autonomous intelligent agent that can safeguard the entire environment from malicious attacks. Hence, this paper contributes to achieving this aim by selecting the eight most valuable features out of 50 features from the adopted dataset using the Chi-squared test. Then, three wellknown machine learning classifiers (i.e. naive Bayes, random forest and logistic regression) are compared in classifying malicious attacks from non-attacks in an intelligent health-care environment. The highest achieved classification accuracy was for the random forest classifier (99.92\%).

Organizations strive to secure their valuable data and minimise potential damages, recognising that critical operations are susceptible to attacks. This research paper seeks to elucidate the concept of proactive cyber threat hunting. The proposed framework is to help organisations check their preparedness against upcoming threats and their probable mitigation plan. While traditional threat detection methods have been implemented, they often need to address the evolving landscape of advanced cyber threats. Organisations must adopt proactive threat-hunting strategies to safeguard business operations and identify and mitigate unknown or undetected network threats. This research proposes a conceptual model based on a review of the literature. The proposed framework will help the organisation recover from the attack. As the recovery time is less, the financial loss for the company will also be reduced. Also, the attacker might need more time to gather data, so there will be less stealing of confidential information. Cybersecurity companies use proactive cyber defence strategies to reduce an attacker s time on the network. The different frameworks used are SANS, MITRE, Hunting ELK, Logstash, Digital Kill Chain, Model in Diamonds, and NIST Framework for Cybersecurity, which proposes a proactive approach. It is beneficial for the defensive security team to assess their capabilities to defend against Advanced Threats Persistent (ATP) and a wide range of attack vectors.

Advanced persistent threats (APTs) have novel features such as multi-stage penetration, highly-tailored intention, and evasive tactics. APTs defense requires fusing multi-dimensional Cyber threat intelligence data to identify attack intentions and conducts efficient knowledge discovery strategies by data-driven machine learning to recognize entity relationships. However, data-driven machine learning lacks generalization ability on fresh or unknown samples, reducing the accuracy and practicality of the defense model. Besides, the private deployment of these APT defense models on heterogeneous environments and various network devices requires significant investment in context awareness (such as known attack entities, continuous network states, and current security strategies). In this paper, we propose a few-shot multi-domain knowledge rearming (FMKR) scheme for context-aware defense against APTs. By completing multiple small tasks that are generated from different network domains with meta-learning, the FMKR firstly trains a model with good discrimination and generalization ability for fresh and unknown APT attacks. In each FMKR task, both threat intelligence and local entities are fused into the support/query sets in meta-learning to identify possible attack stages. Secondly, to rearm current security strategies, an finetuning-based deployment mechanism is proposed to transfer learned knowledge into the student model, while minimizing the defense cost. Compared to multiple model replacement strategies, the FMKR provides a faster response to attack behaviors while consuming less scheduling cost. Based on the feedback from multiple real users of the Industrial Internet of Things (IIoT) over 2 months, we demonstrate that the proposed scheme can improve the defense satisfaction rate.

As the use of machine learning continues to grow in prominence, so does the need for increased knowledge of the threats posed by artificial intelligence. Now more than ever, people are worried about poison attacks, one of the many AI-generated dangers that have already been made public. To fool a classifier during testing, an attacker may "poison" it by altering a portion of the dataset it utilised for training. The poison-resistance strategy presented in this article is novel. The approach uses a recently developed basic called the keyed nonlinear probability test to determine whether or not the training input is consistent with a previously learnt Ddistribution when the odds are stacked against the model. We use an adversary-unknown secret key in our operation. Since the caveats are kept hidden, an adversary cannot use them to fool a keyed nonparametric normality test into concluding that a (substantially) modified dataset really originates from the designated dataset (D).

This survey paper provides an overview of the current state of AI attacks and risks for AI security and privacy as artificial intelligence becomes more prevalent in various applications and services. The risks associated with AI attacks and security breaches are becoming increasingly apparent and cause many financial and social losses. This paper will categorize the different types of attacks on AI models, including adversarial attacks, model inversion attacks, poisoning attacks, data poisoning attacks, data extraction attacks, and membership inference attacks. The paper also emphasizes the importance of developing secure and robust AI models to ensure the privacy and security of sensitive data. Through a systematic literature review, this survey paper comprehensively analyzes the current state of AI attacks and risks for AI security and privacy and detection techniques.

AI technology is widely used in different fields due to the effectiveness and accurate results that have been achieved. The diversity of usage attracts many attackers to attack AI systems to reach their goals. One of the most important and powerful attacks launched against AI models is the label-flipping attack. This attack allows the attacker to compromise the integrity of the dataset, where the attacker is capable of degrading the accuracy of ML models or generating specific output that is targeted by the attacker. Therefore, this paper studies the robustness of several Machine Learning models against targeted and non-targeted label-flipping attacks against the dataset during the training phase. Also, it checks the repeatability of the results obtained in the existing literature. The results are observed and explained in the domain of the cyber security paradigm.

Federated learning is proposed as a typical distributed AI technique to protect user privacy and data security, and it is based on decentralized datasets that train machine learning models by sharing model gradients rather than sharing user data. However, while this particular machine learning approach safeguards data from being shared, it also increases the likelihood that servers will be attacked. Joint learning models are sensitive to poisoning attacks and can effectively pose a threat to the global model when an attacker directly contaminates the global model by passing poisoned gradients. In this paper, we propose a joint learning poisoning attack method based on feature selection. Unlike traditional poisoning attacks, it only modifies important features of the data and ignores other features, which ensures the effectiveness of the attack while being highly stealthy and can bypass general defense methods. After experiments, we demonstrate the feasibility of the method.

Machine learning models are susceptible to a class of attacks known as adversarial poisoning where an adversary can maliciously manipulate training data to hinder model performance or, more concerningly, insert backdoors to exploit at inference time. Many methods have been proposed to defend against adversarial poisoning by either identifying the poisoned samples to facilitate removal or developing poison agnostic training algorithms. Although effective, these proposed approaches can have unintended consequences on the model, such as worsening performance on certain data sub-populations, thus inducing a classification bias. In this work, we evaluate several adversarial poisoning defenses. In addition to traditional security metrics, i.e., robustness to poisoned samples, we also adapt a fairness metric to measure the potential undesirable discrimination of sub-populations resulting from using these defenses. Our investigation highlights that many of the evaluated defenses trade decision fairness to achieve higher adversarial poisoning robustness. Given these results, we recommend our proposed metric to be part of standard evaluations of machine learning defenses.

Wireless Sensor Networks (WSN s) have gained prominence in technology for diverse applications, such as environmental monitoring, health care, smart agriculture, and industrial automation. Comprising small, low-power sensor nodes that sense and collect data from the environment, process it locally, and communicate wirelessly with a central sink or gateway, WSN s face challenges related to limited energy resources, communication constraints, and data processing requirements. This paper presents a comprehensive review of the current state of research in WSN s, focusing on aspects such as network architecture, communication protocols, energy management techniques, data processing and fusion, security and privacy, and applications. Existing solutions are critically analysed regarding their strengths, weaknesses, research gaps, and future directions for WSNs.