As private 4G and 5G networks become increasingly critical for industrial applications such as smart manufacturing and autonomous systems, ensuring the security and reliability of the Radio Access Network (RAN) is essential. Traditional mobile network defenses rely heavily on centralized analysis at the core network, resulting in latency and limited scalability. To address these challenges, the Open Radio Access Network (O-RAN) architecture introduces modular and programmable components that enable security mechanisms to operate closer to the network edge. In this paper, we propose a hybrid detection framework that integrates a rule-based traffic filter at the O-CU with an intelligent xApp deployed on the Near-RT RIC. The system performs GTP decapsulation and flow-level feature extraction at the O-CU, then transmits the data to the xApp, which employs an LSTM-based detector to identify low-rate Denial-of-Service attacks. Detection results are fed back to dynamically update local filtering rules. Experimental results demonstrate that the proposed framework enables lightweight and near real-time threat mitigation, thereby reducing reliance on core-network-based analysis and enhancing the responsiveness of RAN-level defenses.
The Open Radio Access Network (RAN) represents a significant advancement in the ongoing evolution of mobile networks, transitioning from proprietary physical hardware to virtualised network functions. Open RAN advocates for a disaggregated RAN utilising commercial off-the-shelf (COTS) hardware. The O-RAN Alliance is the preeminent organisation in the Open RAN initiative, guiding the industry towards a vendor-neutral radio access network characterised by open interfaces and protocols. The introduction of RAN Intelligent Controllers (RICs) and the ability to deploy third-party services on these RICs expedite the innovation within the RAN. The two RICs, non-real-time RIC and near-real-time RIC, enhance the operation of RAN by facilitating the deployment of third-party services, either as an rApp for non-real-time RIC or as an xApp for near-real-time RIC. However, this new disaggregated and open RAN expands the threat surface and introduces novel security and privacy challenges that were previously absent, and these issues remain unaddressed. The introduction of new stakeholders, such as third-party application providers and cloud service providers, into the RAN ecosystem presents potential vulnerabilities. This paper proposes a hierarchical management strategy to tackle security challenges in Open RAN, enabling authorisation, authentication, and monitoring for third-party applications. Experimental evaluations across multiple configurations demonstrate that the proposed framework is scalable and imposes minimal overhead, making it a practical solution for securing next-generation RAN deployments.
The deployment of 5G networks and increasing virtualization of network components have expanded the attack surface of telecommunications infrastructures, requiring domain specific expertise for effective and rapid vulnerability detection and assessment. AI-based solutions could be a valid support, but any information regarding 5G configurations or infrastructure architectures is highly sensitive and often subject to data protection policy that prohibit sharing with external services, including Large Language Models (LLMs) hosted in proprietary cloud systems. On the other hand, a local LLM deployment ensures compliance with confidentiality requirements while maintaining control over critical information. Our research investigates the capabilities of LLMs, from lightweight to frontier architectures, for identifying 5G relevant Common Vulnerabilities and Exposures (CVEs), by introducing a framework that executes a multi-model evaluation with progressive prompt complexity. We validate LLM responses using a novel dataset of CVEs that we manually annotated, evaluating performance to identify models, parameters, and strategies for a practical applicability in local environments resources limited. Preliminary results show that enriching the prompt by adding more context using embeddings can lead to encouraging performance in 5G CVE classification in limited settings with models under 14 billion of parameters, indicating potential advantages compared to large and frontier models which require higher inference computational costs.
Radio Frequency Fingerprinting (RFF) enables passive device identification by leveraging hardware-induced signal distortions such as phase noise and frequency offset. This technique introduces relevant privacy risks, as passive observers can re-identify and track devices over time. To address these concerns, recent defenses have proposed injecting random noise to obscure fingerprint features and hinder classification. In this paper, we revisit the threat model from an adversarial standpoint and demonstrate that such noise-based obfuscation is fundamentally vulnerable. We introduce Echo, a lightweight signal-level attack that exploits the statistical independence of additive noise across repeated transmissions; by applying a combination of ensemble averaging with autoencoder-based reconstruction error analysis, Echo systematically suppresses obfuscating noise and restores persistent hardware signatures. We evaluate this approach on a public dataset of ten USRP transmitters under varying noise conditions and three adversarial training scenarios. Our results show that Echo achieves fingerprint recovery across 22 different test scenarios, obtaining a minimum Area Under the Curve (AUC) of 0.80. These findings expose a critical weakness in current obfuscation strategies and call for further research in the area.
Cellular networks, including 5G and 4G, use the radio resource control (RRC) communication to establish and set up the communication channel between a user equipment and a base station. We study the Wireless Blackhole threat, in which a fake base station directs multiple users to connect to itself, thereby depriving them of connectivity availability. The attacker manipulates the standardized cellular protocol by using the broadcasting control communication transmission-in contrast to the previous research targeting a single victim using unicast transmission-hence the name Blackhole. The Wireless Blackhole threat directs the victims' connections to the attacker itself, unlike jamming triggering the victims to avoid the attacker/channel. The threat is effective against user equipment both during connection initiation and when idle with an existing connection, and the threat impact range depends on the attacker's transmission power. Informed by the threat and vulnerability analyses, we design and build a distributed edge defense which detects and recovers from the threat. To enable such defense, our defense leverages a whitelist of the registered base stations (e.g., our implementation is based on that published by FCC) and utilizes a distributed, out-of-band, and offline delivery of the list via blockchain. We implement the Wireless Blackhole threat and defense on software-defined radios and validate and evaluate it against both a srsRAN-based user equipment and mobile phones connected to real-world cellular operators. We facilitate further R&D to secure the cellular networking against fake base station and discuss future directions, including those which can add to our defense but require changes in the cellular infrastructure.
Recently, the development of 6G technology has been increasingly associated with sustainability, which is considered one of the key criteria for the success of next-generation 6G communication networks. The 6G ecosystem is expected to be more complex and vulnerable due to the both existing as well as emerging security threats. 6G-enabled verticals will require long-lasting, robust, resilient, energy-efficient, and holistic sustainable security solutions. However, sustainability in the context of 6G security remains a relatively less explored area in the literature. This work provides an initial contribution towards defining the concept of 6G security and sustainability from two major perspectives, i.e., sustainable security for 6G and security for sustainable 6G. Furthermore, the paper analyzes 6G security from five main sustainability aspects: environmental, economic, social, technological, and legal, and examines various factors and key performance indicators (KPIs) critical to the design of green 6G security.
Advanced Persistent Threats (APTs) are complex, stealthy attacks that involve multiple stages and many attack techniques used in each stage, making them difficult to defend against. Although many solutions can detect APTs, most of them only detect the existence of attack, but cannot produce fine-grained classification over the stage of the APT and the specific attack technique used. Some existing solutions can classify the stages of APT, but few of them do attack technique classification, and existing work do not provide interpretability for the classification or countermeasures for the attack. In this paper, we propose a solution named CAPTure, to detect APT stages and attack techniques, provide interpretability for the detection, and map the attack to MITRE D3FEND countermeasures. Specifically, we introduce a selective undersampling technique to address the imbalanced data distribution problem, and a hybrid graph representation concatenated with numerical flow statistics features to capture both structural and quantitative contexts for detection. Leveraging these features, we further design a stacked hierarchical classifier for APT stage detection and attack technique detection, followed by feature importance generation for interpretability and countermeasure mapping for attack mitigation. Evaluations on two open-source APT datasets show that CAPTure can detect APT stages and attack techniques accurately, with higher performance than baseline schemes. It also has a short latency of below 1.25 seconds in detection, making it capable of real-time detection.
Industrial Cyber-Physical Systems (CPS) face escalating cybersecurity threats capable of causing significant operational disruptions, yet existing risk assessment methods often fail to adequately evaluate the threats posed by sophisticated multi-stage attacks. This paper introduces a novel methodology to identify the most critical cyberattack scenarios targeting industrial CPS operations. Our approach integrates an artificial intelligence (AI) based sequential decision-making optimization engine with a simulation model of the industrial process and a dynamical system based on cyber-states that represent the cyberattack as it evolves according to the adversary's actions. This allows for the discovery of attack scenarios that maximize the disruption of operational key performance indicators of the industrial CPS. A key contribution of this work is the systematic integration of well-accepted cybersecurity data sources and control frameworks to define a principled “cost” for each adversary action. This principled cost, combined with a configurable attacker “budget” representing total adversary effort, enables the AI engine to identify the most critical attack paths and quantify the maximum operational disruption achievable. The resulting information is invaluable for operators, particularly in domains like power systems, offering crucial guidance to achieve an enhanced cybersecurity posture. Our methodology's efficacy is demonstrated through testing and validation on both a synthetic power system scenario, based on a modified IEEE 123-bus network, and a real-world application to the distribution grid operated by Holy Cross Energy in Colorado, United States.
Federated Learning (FL) is a decentralized machine learning technique that trains models locally using only model parameters, reducing data transmission concerns and improving privacy. FL encounters security threats, particularly the Backdoor Attack, which injects targeted patterns by malicious clients during training. This can lead to incorrect predictions when exposed to specific triggers. Besides, the presence of Independent and Identically Distributed (non-IID) data in practical applications poses a significant challenge for FL, as conventional models consider data samples from the same distribution. Distinguishing non-IID distributions from malicious attacks, including model poisoning attacks, is particularly difficult, as they can induce gradient anomalies and affect global model accuracy. The issue of non-IID data and the prevention of misclassification remains a significant challenge. This work proposes an Adaptive Attention-Guided Multi-Teacher Distillation technique for non-IID FL settings. In this work, a student model acquires knowledge by selecting only those teachers who make accurate predictions and by extracting their consensus attention. This process utilizes adaptive attention to avoid the integration of backdoor-related attention, thereby mitigating the risks associated with backdoor attacks. Empirical results show that when testing on datasets comprising exclusively of samples with triggers, our method reliably maintains the Attack Success Rate (ASR) under 10%. In most cases, the proposed work demonstrates an average Accuracy in Trigger data (AiT) that outperforms existing state-of-the-art methods by approximately 9%. Thus, the proposed solution is more suitable for mitigating backdoor attacks in non-IID FL.
5G communication enables digital innovation by integrating a wide range of services, making security, especially primary authentication, critically important. To secure 5G communication, the 3GPP has designed the 5G Authentication and Key Agreement (5G-AKA) protocol. However, current trends show a growing interest in post-quantum security, and 5G-AKA is not resistant to quantum-era threats due to several classical vulnerabilities. To address these limitations, various enhanced versions of 5G-AKA have been proposed, claiming to provide security against attacks that exploit its weaknesses. However, state-of-the-art research reveals that many of these versions remain susceptible to classical threats, such as attacks from malicious serving networks (SNs), ephemeral secret leakage (ESL) attacks, and fail to provide perfect forward secrecy. Additionally, they are not secure against quantum attacks. In response to these challenges, we propose a Post-Quantum Secure, Lattice-Based 5G-AKA Protocol that is resistant to malicious serving networks and ensures perfect forward secrecy. Our proposed protocol is built on lattice-based cryptography, making it one of the first 5G-AKA variants designed with post-quantum security in mind. We validate the security of our protocol using formal verification techniques, including the Real-Or-Random (ROR) model and the Scyther tool. Furthermore, we perform a comparative analysis with existing protocols to highlight the improvements. Finally, we implement our protocol in a testbed environment to demonstrate its practical feasibility for real-time 5G applications.
Deep learning models, particularly those used for im-age classification, have become pervasive in today's technological landscape. However, the increasing demand for privacy, espe-cially when sharing data with external parties, poses significant challenges. Homomorphic encryption (HE) offers a compelling solution by enabling computations on encrypted data without compromising privacy. Despite its potential, the widespread adoption of HE remains limited due to its computational complexity. In this work, we propose an optimized approach to improve the efficiency of a fundamental building block in image classification models: the convolution operation. Our approach is based on the observation that multiple kernel evaluations in a convolution layer can be performed in parallel, provided that their receptive fields do not overlap. Using this property, we reorganize convolution computations into independent groups, significantly reducing redundant operations. This grouping reduces the number of multiplications by up to 75% compared to standard encrypted convolution techniques. Additionally, the approach eliminates the need for costly preprocessing, enabling direct processing of encrypted inputs. To further accelerate homomorphic operations, we leverage a GPU-accelerated HE library, enabling faster op-erations. These contributions make encrypted convolution signifi-cantly more practical for real-world deep learning applications.
We address the problem of data falsification attacks (DFAs) in cooperative autonomous vehicles. In DFAs, compromised or malicious vehicles broadcast false navigation information (acceleration, velocity, position, steering, motion intent) to influence the control algorithm of neighboring vehicles and cause collisions, traffic jams, and passenger discomfort. To maintain safe and efficient driving under DFAs, we propose a trust-aware par-tially observable Markov decision process (POMDP) framework that accounts for the dynamic uncertainty caused by Byzantine behaviors. Compared to traditional POMDP formulations, our framework integrates data fusion between sensing data and V2V messages across different data sampling rates, thereby addressing the challenge of achieving robust real-time control despite delayed data verification. Moreover, the incorporation of trust allows for quickly shifting real-time control between a sensors-only mode (safer but less efficient) to a sensor plus V2V messages mode (less safe but more efficient), depending on the trust level. Extensive simulations with real traffic data demonstrate that our method accurately maintains the vehicle's true state, prevents catastrophic vehicle collisions while ensuring passenger comfort and efficiency.
As cyber threats grow more advanced, real-time classification and prediction of attacker behavior is critical for effective defense. This paper presents a novel approach to classifying and predicting ongoing cyber attacks. Our methodology employs agglomerative clustering based on normalized Levenshtein distance to group similar attack sequences, followed by Markov Chain modeling and Bayesian classification to predict attacker intentions early in the attack. Additionally, we introduce a quantitative risk scoring framework using VirusTotal's IP reputation to assess the severity of identified attack clusters. Among 10,126 test sequences, 99.79% of attacks are correctly clustered after observing the first command. We define a data-driven entropy threshold-“min wrong guess line”-to signal real-time prediction confidence. Our approach enables early-stage threat classification and risk assessment for faster and more informed response to evolving cyberattacks.
Cyber threats present an ongoing challenge for organisations worldwide. Attackers range from cybercriminals to state-funded groups that have a specialised skill set to execute complex attacks and present an Advanced Persistent Threat (APT). Therefore, organisations use security monitoring as a second line of defence to detect attacks based on signatures that raise alarms when an Indicator of Compromise (IoC) is observed. However, current Intrusion Detection Systems (IDS) generate many false positives, leading to alert fatigue. The raised alerts also do not show the whole attack as they need to be reviewed individually. Our work presents a simplified approach that enables efficient and real-time construction of attacks by correlating alerts. Tests with different network datasets suggest that our prioritisation mechanisms can reduce the number of false-positive alerts by 99 %. Our performance evaluation indicates that we can detect multi-stage attacks in real-time with a low memory footprint and short execution time.
Upper-Ilayer cryptographic techniques are commonly used to ensure secure data transmission over wireless networks by encrypting information at the software level. However, wireless transmissions remain vulnerable to passive and active attacks. To address this issue, research has been conducted on physical layer encryption (PLE). Unlike traditional upper-layer encryption, which operates at the bit level, PLE obscures the transmitted signal, making it more challenging for eavesdroppers to decipher. This paper proposes a novel physical layer chaotic encryption method designed to obfuscate the modulation scheme used in orthogonal frequency division multiplexing (OFDM) systems. The proposed scheme leverages a chaotic system, the three-dimensional logistic-sine cascading map (3D-LSCM), to scramble constellation points. The method improves upon previous work by incorporating histogram equalisation to evenly distribute the output of the chaotic system, thereby better obfuscating the modulation scheme. Experimental results obtained using the ADALM-PLUTO with an OFDM signal demonstrate the effectiveness of the encryption scheme in concealing modulation types, such as Quadrature Amplitude Modulation (QAM), from potential eavesdroppers. The encryption demonstrates high key sensitivity, statistical independence, and large key space, highlighting its robustness against brute-force attacks.
The increasing sophistication and volume of Android malware have driven the adoption of Machine Learning (ML) models for scalable and accurate threat detection. However, these models remain vulnerable to adversarial attacks that subtly manipulate input features to evade classification. In this paper, we introduce LAMLAD, a novel adversarial attack framework that exploits the generative and reasoning capabilities of Large Language Models (LLMs) to bypass ML-based malware detectors. LAMLAD employs a two-agent architecture composed of an LLM manipulator, which crafts realistic feature-level perturbations without altering core malicious behaviors, and an LLM analyzer, which guides the modification process to ensure successful evasion. To improve efficiency and context awareness, LAMLAD integrates Retrieval-Augmented Generation (RAG) into the LLM workflow. By targeting Drebin features, LAMLAD enables stealthy and high-confidence attacks against widely used Android malware classifiers. We evaluate LAMLAD against three representative ML-based Android malware detectors and compare it with two state-of-the-art adversarial attack techniques. Results demonstrate that LAMLAD achieves an attack success rate of 97% with an average of 3 attempts per adversarial example, highlighting its potency, efficiency, and adaptability in real-world adversarial scenarios.
iSOCaaS aims to boost European cyber-security resilience across all levels-organizational, sectoral, national, cross-border, and EU-by creating a standards-based, interoperable Security Operations Center (SOC) cluster. This initiative will empower EU member states to better tackle evolving cyber threats and meet the continent's cybersecurity needs.
Integrated communications and sensing (ISAC) has emerged as a core capability in next-generation wireless networks (NextG), providing numerous advantages in spectral efficiency, performance, and hardware cost. However, using communication waveforms for sensing leaves ISAC systems vulnerable to sensing attacks, as such waveforms are not designed with sensing security in mind. In this paper, we investigate the security of ISAC systems against sensing attacks. We focus on OFDM due to its widespread adoption. We theoretically and experimentally demonstrate that OFDM-based sensing is vulnerable to target spoofing and elimination attacks that can alter the perception of the physical environment. Such attacks are launched online by exploiting the known structure of OFDM waveforms. To counter sensing attacks, we design waveform randomization methods that exploit the flexible 5G/NextG numerology to implement a moving target defense strategy. We analyze the security of the proposed attack detection method and demonstrate that online attackers are detected with high probability. We validate the security and communication/sensing performance of our methods in a 28 GHz mm Wave testbed by implementing the ISAC functionality, the sensing attacks, and the proposed countermeasures.
The advancement of wireless networks towards sixth-generation (6G) and pervasive artificial intelligence (AI) underscores critical barriers to trustworthy cybersecurity, notably algorithmic bias, workforce underrepresentation, and the neglect of intersectional gender perspectives. Current cybersecurity research predominantly emphasizes algorithmic performance, often overlooking broader societal impacts such as equity and inclusivity. This paper addresses this gap by synthesizing interdisciplinary lessons from emerging technology deployments, critically examining risks inherent in AI-driven 6G security, and articulating the urgent need for an “Equity by Design” approach. We introduce a comprehensive roadmap to embed intersectional equity into cybersecurity processes—from policy-making and technical development to real-world deployment. Utilizing the BEiNG-WISE COST Action as a best-practice case study, we demonstrate how explicit gender-inclusive and participatory methodologies can effectively operationalize equity, transparency, and diversity. Our findings illustrate that integrating social and technical dimensions is essential for building secure, resilient, and equitable digital futures.
In our information-driven society, the volume of data generated by individuals has grown significantly. Protecting the privacy of individuals is becoming more challenging, as this data can reveal detailed insights into personal preferences and behavior. To address this challenge, we introduce a user-centric, privacy-preserving data-sharing solution that leverages a central data storage service, hereafter referred to as the data intermediary. By integrating local differential privacy with user-in-the-Ioop access control, our system enables data providers to securely and unlinkably store their data at the intermediary. Data consumers can localize and request data via the intermediary. The data providers are included in access decisions without disclosing their identity nor by enabling the linking of their data. We evaluated our approach using theoretical analysis and simulations. Our findings indicate that our system achieves c-privacy and safeguards data providers against external and internal attackers, malicious data consumers, and an honest-but-curious intermediary. Moreover, our method reduces the message overhead for data discovery in our system by more than half compared to existing approaches.
