5G slices are susceptible to indirect Distributed Denial of Service (DDoS) attacks, where overwhelming traffic directed to one slice can also disrupt other slices sharing the same infrastructure Many current mitigation methods rely on a detection phase, which may not be effective against unknown or sophisticated attacks. Moving Target Defense (MTD) is a security mechanism that invalidates the adversary's collected information, and it can be deployed without the detection phase. In this paper, we propose a Slice Mutation technique based on Reinforcement Learning (SMRL) that reduces the impact of DDoS attacks on 5G slices while keeping the number of allocated slices acceptable. SMRL proposes a general RL model that considers ternary and ranking numbers to improve learning performance. We tested SMRL on computer networks attacked by a real botnet called Mirai and assessed its performance using various measures, including a new functionality analysis method The results indicate that SMRL decreases the number of slices impacted by a DDoS attack and enhances the distribution of slices among infrastructure resources by 46 % and 20 %, respectively.
Phishing attacks are a significant threat to cybersecurity, particularly among university students who are frequent targets due to their extensive online activities and limited cybersecurity awareness. This study explores the impact of various factors, including threat susceptibility, phishing avoidance behavior, and the use of anti-phishing tools, on students' awareness of phishing attacks. Using a quantitative approach, data were collected from 715 university students worldwide through a structured questionnaire. The findings reveal that while students exhibit a moderate level of awareness about phishing attacks, their reliance on anti-phishing tools remains insufficient. The study identifies a significant positive relationship between the use of anti-phishing tools and increased phishing awareness and avoidance behaviors. Additionally, the research highlights the mediating role of anti-phishing tools in enhancing students' cybersecurity awareness. The results underscore the importance of integrating educational programs and advanced anti-phishing tools to improve students' resilience against phishing attacks. Recommendations for enhancing cybersecurity education and practices among university students are also provided.
This study explores how Microsoft AirSim and OpenAI's Natural Language Processing capabilities can enable drone navigation within a campus simulation. Utilizing Unreal Engine, we create a 3D simulation of Georgia State University's campus to investigate language-based drone control. Our implementation integrates three key technologies: (1) Microsoft's AirSim platform for simulating drone physics, (2) OpenAI's ChatGPT API for natural language interpretation and command processing, and (3) a detailed campus environment within Unreal Engine. This integration replaces traditional drone control interfaces, allowing users to operate simulated drones through natural language instructions. By translating user commands into navigation directions, this technology showcases the practical applications of language models. Our findings indicate that this approach enhances campus navigation simulations and provides a secure environment for testing drone operations in urban settings. This study highlights the potential of combining language processing with drone control systems, particularly in educational simulations.
In current times, network slicing in a 5G context is a significant study field. But it might be difficult to meet network slice requests' requirements. Network slices need to share limited resources; therefore energy efficiency and security are crucial. Additionally, it is essential to establish secured network slicing for Software-Defined Network/Network Function Virtualization (SDN/NFV). As attackers have developed to become more skilled and frequently use different attacking approaches, security is a crucial problem in network slicing. We address security, ineffective network slicing, and overloading using load balancing and Deep Learning (DL) based network slicing algorithms in edge enabled SDN/NFV assisted 5G settings in this research. Here, we mainly concentrate on secure and efficient network slicing in SDN/NFV assisted 5G systems. Initially, slicing of network is performed based on UniqueNet which includes lightweight convolutional layers that reduce the processing time and increase accuracy. For authentication of users, we employ the Improved Mersenne Twister (IMT) algorithm and role-based access control is performed using Improved Deep Q Network (ImDQN) algorithm for authorization purpose. Clustering is done by using k-means clustering (KMC) algorithm. Here, Cluster Head (CH) performed intrusion detection using Enhanced Bidirectional Generative Adversarial Network (E-BiGAN) algorithm. After detected intrusions, the Kangaroo-based IDS (KIDS) jump and send the notification to all the nodes in the CH. For efficient load balancing, we perform optimal switch selection using Dove Swarm Optimization (DSO). The performance of the suggested framework is then evaluated in terms of different metrics and compared with existing approaches to prove the efficacy of the proposed system.
Cloud computing has become a significant part of people's daily lives over the last decade due to its cost-effective services, such as Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). As a result, the cloud environment contains a massive amount of sensitive and confidential data, thus becoming a crucial target for attackers. Additionally, cloud forensics over the cloud environment has become complex and difficult due to the cloud's several challenges. These challenges include collecting volatile data, dependency on cloud service providers (CSPs), secure logs storage, and more. This paper proposes a Cloud-Based Data Volatility and CSP Reliance Eradication Framework (CDCEF) to mitigate data volatility and dependency on CSP in cloud forensics in IaaS. We also performed experiments on the Amazon Web Service (AWS) Lightsail - a widely used cloud platform by people globally, based on hypothetical cybercrime scenarios to support our framework. The significant benefits of this framework for investigators are that it allows them to collect volatile data without losing the integrity of the evidence and eradicates CSP reliance. Additionally, the framework provides another benefit of minimizing the usage of forensic tools.
Image manipulation threatens data integrity and public trust, making reliable authenticity tools essential. The development of a publicly evaluatable perceptual hash framework enables various applications, including private image search resilient to image alterations. Despite the potential of such a framework, little research has systematically analyzed the performance of various perceptual hash algorithms within it. In this paper, we assess the performance of several leading perceptual hash methods, including aHash, pHash, dHash, wHash, and DCT, across five diverse image datasets and examine how cryptographic techniques impact the effectiveness of the algorithms. Integrating advanced encryption techniques with perceptual hashing in this approach is instrumental in advancing data security. It improves the security, privacy, and computational efficiency of perceptual hashing, solidifying its importance within the overall methodology.
The Internet of Things (IoT) is a network of interconnected devices and systems that collect and exchange data. Resource-constrained IoT devices are particularly susceptible to attacks due to their widespread use and often inadequate security setups, leading to potential vulnerabilities in individual devices. Traditional Signature-based Intrusion Detection Systems (SIDS) are insufficient for the dynamic IoT landscape, where new devices and protocols continually emerge. Anomaly-based IDS systems(AIDS), which detect deviations from normal behavior, are better suited for IoT environments as they provide continuous monitoring and real-time detection of malicious activities, enhancing threat intelligence and reducing false positives. In this research, we empirically analyze anomalybased IDS systems using various machine learning techniques deployed on Raspberry Pi. The effectiveness of the system is evaluated in terms of detection accuracy, computational efficiency, and resource utilization. Power consumption is measured using a source meter and CPU usage is monitored with the Glances software. This study demonstrates that Random Forest is the most balanced machine learning algorithm for anomaly-based IDS on IoT devices, offering high accuracy of 98.2% with efficient resource utilization (an average energy consumption of 40.94 Joules, peak CPU utilization of 35%, and average power consumption of 3.24 watts), paving the way for future research in adaptive and scalable intrusion detection.
To detect malicious URLs more timely, machine learning based malicious URL detection methods have replaced traditional blacklist methods. These studies aim to improve the accuracy and speed of detection from various aspects such as URL segmentation, URL embedding methods, machine learning models, etc. However, the security issues inherent in these machine learning based malicious URL detection methods have been overlooked. Adversarial example attacks are one of the security issues faced by machine learning based malicious URL detectors. In this paper, we proposed a new adversarial example attack method against malicious URL detection based on machine learning and it has better performance than existing methods. Besides, we compared the robustness of different URL embedding methods and machine learning models with our attack methods and existing attack methods. At last, we analyzed the reasons why our proposed method performs better and the reasons why the context-considered embedding method has high resistance to adversarial example attacks.
Cross-lingual sentiment analysis has developed as a significant area of research in linguistics, especially for languages having diverse syntactic and morphological structures. The objective of this study emphasizes creating a sophisticated sentiment analysis model that connects English and Arabic datasets, two languages with distinct linguistic problems. Using cutting-edge transformer architectures, we utilize pre-trained models—BERT for English and AraBERT for Arabic—to address the challenges of morphologically rich but resource-limited languages such as Arabic. The foundation of this study is the IMDB movie review dataset, which is similarly structured and large for both languages. To find the best deep learning architecture, we conducted extensive experiments using Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and attention methods. While LSTM-based models produced competitive results, transformer-based models such as proposed BERT-LA that included bidirectional and attention layers outperformed them substantially, particularly on Arabic and English data. Furthermore, ablation research was conducted to evaluate the models' performance using important measures such as accuracy, precision, recall, and the F1-score. Our model got an impressive 97.04% accuracy on the English dataset and 98.02% on the Arabic dataset. This study contributes to understanding how language-specific embeddings and transformer models affect under-represented languages.
In recent years, the rapid improvement of deep learning technologies, particularly Generative Adversarial Net-works, has led to the proliferation of high-quality synthetic facial images and videos, commonly known as deepfakes. This study aims to evaluate and compare the performance of three prominent deep learning models - ResN et, EfficientNet, and Xception - in detecting synthetic faces. Using the Deepfake Detection Challenge and FaceForensics++ datasets, we system-atically assess each model's capability to handle diverse and challenging scenarios, including blurred and dark images. Data augmentation techniques, such as random blurring, brightness adjustment, and contrast enhancement, were employed to im-prove the models' robustness. Additionally, we applied model- specific optimizations, including the integration of Squeeze-and- Excitation blocks in Res Net, compound scaling in EfficientNet, and multi-scale feature fusion in Xception. These enhancements significantly improved the models' accuracy and resilience against low-quality synthetic data. Our results indicate that EfficientNet and Xception outperform ResNet in both general and adverse conditions, with EfficientNet excelling in high-resolution image processing and Xception showing superior performance in fine- grained feature extraction. Furthermore, the introduction of pre- trained weights, multitask learning frameworks, and dynamic learning rate adjustments during training contributed to the models' enhanced performance.
While deep learning-based recommendation systems have achieved great success, recommendation system models are also at serious risk of intellectual property infringement. Current model watermarking research faces significant challenges in terms of fidelity, invisibility, and efficiency. Additionally, existing model watermarking techniques are predominantly applied to image data, with limited applicability to tabular data. In this paper, we introduce an innovative watermarking framework designed to safeguard the ownership of recommendation system models. Specifically, we verify recommendation system model ownership by embedding a type of backdoor watermark into the training dataset, which does not affect model performance. We have conducted experiments on several classical datasets to validate the reliability and effectiveness of our approach.
This work presents a highly effective strategy for attacking image captioning models through the use of prompt engineering. The objective of this approach is to deliberately guiding the output of LLMs and introduce dynamic noise into the original clean image captions, causing them to be categorized as a different class. Consequently, when the image captioning model is fine-tuned using adversarial captions, it will deteriorate and produce inaccurate captions for clean photos. The novelty of this attack is that it does not require the attacker to perform any model training and only require to prompt the LLMs to generate only a small amount of captions for the attack to be effective. Comprehensive experiments using GPT-3.5 indicate that with only 100 captions created by LLMs with malicious intent can significantly worsen picture captioning model performance by up to over 50% in BLEU metric and over 25% in ROUGE-L and METEOR scores.
Manned-Unmanned Teaming (MUM-T) systems integrate manned aircraft and unmanned aerial vehicles (UAVs) to enhance mission effectiveness, allowing a single pilot to coordinate multiple UAVs for tasks like reconnaissance, communication, and targeting. However, the complexity and operational demands of MUM-T systems introduce significant security challenges, particularly for mission-critical data integrity and real-time communication. In this paper, we propose a new framework for adaptive blockchain cryptography that combines Proof of Authority (PoA)-based blockchain with XOR-based lightweight authentication. The blockchain component, with the manned aircraft that serves as the sole validator, ensures tamper-resistant logging of key mission data. Additionally, it supports accountability and traceability through an efficient PoA consensus algorithm. In parallel, XOR-based lightweight authentication secures control and telemetry signals with minimal computational overhead that enables low-latency and a real-time communication. Analyses results show that the proposed framework achieves a better transaction throughput with acceptable latency, which meets the stringent security and performance requirements of MUM-T operations. The proposed framework offers a scalable and resilient solution for secure communications in complex military environments.
Integrating Groundbreaking advancements in AI, like language models, interpretable AI, and machine learning, opens up a world of exciting new possibilities. The Evolving face of cybersecurity and Modern cyber threats are complex and well crafted; hence, conventional cybersecurity mechanisms show difficulty in staying relevant. LLMs, especially based on Transformer architecture will noticeably increase the accuracy and speed of detecting threats. Transparency and trust are increased by XAI approaches like SHAP and LIME, which offer facts about ML model predictions. This paper explores the literature that demonstrates the integration between XAI and LLMs in cybersecurity, exemplifying how this trinity of models has the potential to help attenuate errors producing reduced false positives and improve how we detect threats. Thinking about the possibilities the challenges including performance Explainability trade-offs, the need for common evaluation metrics, and the black-box nature of AI Models, remain in place. Solving these will help to enhance AI-driven solutions in cybersecurity.
Cyber security is becoming more complex due to the exponential growth of interconnected systems and the global threat landscape. To mitigate those risks, there is a need for a skilled cyber security workforce that can navigate the complex decision-making in rapidly evolving cyberspace. Artificial intelligence (AI) is rapidly adopted into cyber defence operations, but we can not effectively train human and AI-assisted cyber defence operators without understanding the underlying learning theory and eco-systems. Cyber security exercises (CSXs) are popular teaching methods for cyber-readiness. However, applying learning analytics (LA) methods and AI-based approaches to exercise design and implementation is still in the early stages. We propose a holistic human-AI interaction model within the LA and CSX context. The model brings together elements and processes of human-AI interactions, as well as cyber ranges, cyber security, and LA tools, and a wider lens of multimodal learning analytics, exercise life-cycle, and overall pedagogical approach. We also discuss the opportunities and challenges for LA and AI in the context of cyber security training. We analyse the role of AI from the learning, instruction, and administration lens in cyber security training, specifically in the exercises. We aim to stimulate further discussions on the future of human-AI collaboration and how to enhance cyber security training with novel LA and AI capabilities.
The rise of Autonomous Vehicles (AVs) brings with it the need for secure and privacy-preserving machine learning models. Federated Learning (FL) allows AVs to collaboratively train models while keeping raw data localized. However, traditional FL systems are vulnerable to security threats, including adversarial attacks, data breaches, and dependency on a central aggregator, which can be a single point of failure. To address these concerns, this paper introduces a peer-to-peer decentralized federated learning system that integrates lightweight blockchain technology and Binius Zero- Knowledge Proofs (ZKPs) to enhance security and privacy. In this system, Binius ZKPs ensure that model updates are cryptographically verified without exposing sensitive information, guaranteeing data confidentiality and integrity during the learning process. The lightweight blockchain framework secures the network by creating an immutable, decentralized record of all model updates, thus preventing tampering, fraud, or unauthorized modifications. This decentralized approach eliminates the need for a central aggregator, significantly enhancing system resilience to attacks and making it suitable for dynamic environments like AV networks. Additionally, the system's design includes Byzantine resilience, providing protection against adversarial nodes and ensuring that the global model aggregation process remains robust even in the presence of malicious actors. Extensive performance evaluations demonstrate that the system achieves low-latency, scalability, and efficient resource usage while maintaining strong security and privacy guarantees, making it an ideal solution for real-time federated learning in autonomous vehicle networks. The proposed framework not only ensures privacy but also fosters trust among participants in a fully decentralized environment.
Protecting information flow, data and assets is paramount to every establishment. Therefore, enterprise security architecture design is essential in achieving this protection as it directly implements enterprise security policies. Existing research revealed that researchers have made little effort to investigate inference security challenges to enterprise security architecture design and to assess how the existing security architecture models fare against inference attacks. It was also discovered that existing security architecture models are too old and susceptible to inference attacks. Hence, this research explores a novel solution for designing effective enterprise security architecture and addressing inference attacks.
OpenRAN is revolutionizing wireless telecommunications, enabling more flexible and innovative network architectures. Within this framework, near-real-time applications in RAN Intelligent Controllers (near-RT RIC) are pushing the boundaries of ultra-reliable low latency communications. However, security concerns challenge their adoption. This paper investigates vulnerabilities in near-RT RIC AI xApps through systematic experiments, focusing on a Handover AI xApp. Using four distinct attack strategies, we demonstrate that current security measures are inadequate, exposing these Ultra-Reliable Low Latency Communications (URLLC) AI xApps to various attacks. Our findings highlight the potential for malicious exploitation, emphasizing the need for robust security frameworks in OpenRAN deployments utilizing near-RT applications.
This study explores advancements in AI-generated image detection, emphasizing the increasing realism of images, including deepfakes, and the need for effective detection meth-ods. Traditional Convolutional Neural Networks (CNNs) have shown success but face limitations in generalization and accu-racy, particularly with newer technologies like Diffusion Models. With the evolution of AI image generation models, from CNNs to Generative Adversarial Networks (GANs) and Diffusion Models, detecting synthetic images has become more challenging. Issues include dataset diversity, adversarial attacks, and inconsistencies in pre-processing methods. While state-of-the-art models like CNNs, Vision Transformers (ViTs), and hybrid approaches exist, their accuracy in detecting increasingly sophisticated fake images remains suboptimal. This research proposes a novel hybrid detection model combining CNNs and ViTs with an additional attention mechanism layer. This structure aims to improve the interaction between local and global features, enhancing detection accuracy. The model was trained using the CIFAKE dataset, which contains 120,000 real and AI -generated images. The added attention mechanism enhances feature extraction, addressing limitations in existing models when faced with next-generation synthetic images. The hybrid CNNNiT +Attention model demonstrated improved detection accuracy, achieving 99.77%, surpassing previous methods. This research lays a foundation for stronger AI -generated image detection, helping to mitigate the risks of synthetic image fraud.
In many machine learning scenarios, training occurs outside the control of the model sponsor or the entity using the model. A growing concern in such settings revolves around model poisoning and data poisoning-how training is conducted and which data contributes to the process. This paper introduces a protective scheme against model and data poisoning attacks. Leveraging cryptographic primitives such as hashes, signature schemes, and zero-knowledge proofs, the scheme ensures the integrity of the training process. Hashing maintains the continuity of data from authenticated sensors, while signatures validate the data. In the end, zero-knowledge proofs verify the correct model computation by the entity carrying out the training process. By adopting this approach, model sponsors can securely delegate training tasks, guaranteeing the authenticity of the results. Implementation and testing demonstrate the scheme's feasibility, effectively countering data and model poisoning threats.
