CyberSec Research

Split Learning (SL) -- splits a model into two distinct parts to help protect client data while enhancing Machine Learning (ML) processes. Though promising, SL has proven vulnerable to different attacks, thus raising concerns about how effective it may be in terms of data privacy. Recent works have shown promising results for securing SL through the use of a novel paradigm, named Function Secret Sharing (FSS), in which servers obtain shares of a function they compute and operate on a public input hidden with a random mask. However, these works fall short in addressing the rising number of attacks which exist on SL. In SplitHappens, we expand the combination of FSS and SL to U-shaped SL. Similarly to other works, we are able to make use of the benefits of SL by reducing the communication and computational costs of FSS. However, a U-shaped SL provides a higher security guarantee than previous works, allowing a client to keep the labels of the training data secret, without having to share them with the server. Through this, we are able to generalize the security analysis of previous works and expand it to different attack vectors, such as modern model inversion attacks as well as label inference attacks. We tested our approach for two different convolutional neural networks on different datasets. These experiments show the effectiveness of our approach in reducing the training time as well as the communication costs when compared to simply using FSS while matching prior accuracy.

This paper aims to propose a novel machine learning (ML) approach incorporating Homomorphic Encryption (HE) to address privacy limitations in Unmanned Aerial Vehicles (UAV)-based face detection. Due to challenges related to distance, altitude, and face orientation, high-resolution imagery and sophisticated neural networks enable accurate face recognition in dynamic environments. However, privacy concerns arise from the extensive surveillance capabilities of UAVs. To resolve this issue, we propose a novel framework that integrates HE with advanced neural networks to secure facial data throughout the inference phase. This method ensures that facial data remains secure with minimal impact on detection accuracy. Specifically, the proposed system leverages the Cheon-Kim-Kim-Song (CKKS) scheme to perform computations directly on encrypted data, optimizing computational efficiency and security. Furthermore, we develop an effective data encoding method specifically designed to preprocess the raw facial data into CKKS form in a Single-Instruction-Multiple-Data (SIMD) manner. Building on this, we design a secure inference algorithm to compute on ciphertext without needing decryption. This approach not only protects data privacy during the processing of facial data but also enhances the efficiency of UAV-based face detection systems. Experimental results demonstrate that our method effectively balances privacy protection and detection performance, making it a viable solution for UAV-based secure face detection. Significantly, our approach (while maintaining data confidentially with HE encryption) can still achieve an accuracy of less than 1% compared to the benchmark without using encryption.

Large Language Models (LLMs) have transformed software development and automated code generation. Motivated by these advancements, this paper explores the feasibility of LLMs in modifying malware source code to generate variants. We introduce LLMalMorph, a semi-automated framework that leverages semantical and syntactical code comprehension by LLMs to generate new malware variants. LLMalMorph extracts function-level information from the malware source code and employs custom-engineered prompts coupled with strategically defined code transformations to guide the LLM in generating variants without resource-intensive fine-tuning. To evaluate LLMalMorph, we collected 10 diverse Windows malware samples of varying types, complexity and functionality and generated 618 variants. Our thorough experiments demonstrate that it is possible to reduce the detection rates of antivirus engines of these malware variants to some extent while preserving malware functionalities. In addition, despite not optimizing against any Machine Learning (ML)-based malware detectors, several variants also achieved notable attack success rates against an ML-based malware classifier. We also discuss the limitations of current LLM capabilities in generating malware variants from source code and assess where this emerging technology stands in the broader context of malware variant generation.

The emergence of quantum computers poses a significant threat to current secure service, application and/or protocol implementations that rely on RSA and ECDSA algorithms, for instance DNSSEC, because public-key cryptography based on number factorization or discrete logarithm is vulnerable to quantum attacks. This paper presents the integration of post-quantum cryptographic (PQC) algorithms into CoreDNS to enable quantum-resistant DNSSEC functionality. We have developed a plugin that extends CoreDNS with support for five PQC signature algorithm families: ML-DSA, FALCON, SPHINCS+, MAYO, and SNOVA. Our implementation maintains compatibility with existing DNS resolution flows while providing on-the-fly signing using quantum-resistant signatures. A benchmark has been performed and performance evaluation results reveal significant trade-offs between security and efficiency. The results indicate that while PQC algorithms introduce operational overhead, several candidates offer viable compromises for transitioning DNSSEC to quantum-resistant cryptography.

Rowhammer is a read disturbance vulnerability in modern DRAM that causes bit-flips, compromising security and reliability. While extensively studied on Intel and AMD CPUs with DDR and LPDDR memories, its impact on GPUs using GDDR memories, critical for emerging machine learning applications, remains unexplored. Rowhammer attacks on GPUs face unique challenges: (1) proprietary mapping of physical memory to GDDR banks and rows, (2) high memory latency and faster refresh rates that hinder effective hammering, and (3) proprietary mitigations in GDDR memories, difficult to reverse-engineer without FPGA-based test platforms. We introduce GPUHammer, the first Rowhammer attack on NVIDIA GPUs with GDDR6 DRAM. GPUHammer proposes novel techniques to reverse-engineer GDDR DRAM row mappings, and employs GPU-specific memory access optimizations to amplify hammering intensity and bypass mitigations. Thus, we demonstrate the first successful Rowhammer attack on a discrete GPU, injecting up to 8 bit-flips across 4 DRAM banks on an NVIDIA A6000 with GDDR6 memory. We also show how an attacker can use these to tamper with ML models, causing significant accuracy drops (up to 80%).

Phishing attacks are becoming increasingly sophisticated, underscoring the need for detection systems that strike a balance between high accuracy and computational efficiency. This paper presents a comparative evaluation of traditional Machine Learning (ML), Deep Learning (DL), and quantized small-parameter Large Language Models (LLMs) for phishing detection. Through experiments on a curated dataset, we show that while LLMs currently underperform compared to ML and DL methods in terms of raw accuracy, they exhibit strong potential for identifying subtle, context-based phishing cues. We also investigate the impact of zero-shot and few-shot prompting strategies, revealing that LLM-rephrased emails can significantly degrade the performance of both ML and LLM-based detectors. Our benchmarking highlights that models like DeepSeek R1 Distill Qwen 14B (Q8_0) achieve competitive accuracy, above 80%, using only 17GB of VRAM, supporting their viability for cost-efficient deployment. We further assess the models' adversarial robustness and cost-performance tradeoffs, and demonstrate how lightweight LLMs can provide concise, interpretable explanations to support real-time decision-making. These findings position optimized LLMs as promising components in phishing defence systems and offer a path forward for integrating explainable, efficient AI into modern cybersecurity frameworks.

As AI models become ubiquitous in our daily lives, there has been an increasing demand for transparency in ML services. However, the model owner does not want to reveal the weights, as they are considered trade secrets. To solve this problem, researchers have turned to zero-knowledge proofs of ML model inference. These proofs convince the user that the ML model output is correct, without revealing the weights of the model to the user. Past work on these provers can be placed into two categories. The first method compiles the ML model into a low-level circuit, and proves the circuit using a ZK-SNARK. The second method uses custom cryptographic protocols designed only for a specific class of models. Unfortunately, the first method is highly inefficient, making it impractical for the large models used today, and the second method does not generalize well, making it difficult to update in the rapidly changing field of machine learning. To solve this, we propose ZKTorch, an open source end-to-end proving system that compiles ML models into base cryptographic operations called basic blocks, each proved using specialized protocols. ZKTorch is built on top of a novel parallel extension to the Mira accumulation scheme, enabling succinct proofs with minimal accumulation overhead. These contributions allow ZKTorch to achieve at least a $3\times$ reduction in the proof size compared to specialized protocols and up to a $6\times$ speedup in proving time over a general-purpose ZKML framework.

Machine learning models, particularly decision trees (DTs), are widely adopted across various domains due to their interpretability and efficiency. However, as ML models become increasingly integrated into privacy-sensitive applications, concerns about their confidentiality have grown, particularly in light of emerging threats such as model extraction and fault injection attacks. Assessing the vulnerability of DTs under such attacks is therefore important. In this work, we present BarkBeetle, a novel attack that leverages fault injection to extract internal structural information of DT models. BarkBeetle employs a bottom-up recovery strategy that uses targeted fault injection at specific nodes to efficiently infer feature splits and threshold values. Our proof-of-concept implementation demonstrates that BarkBeetle requires significantly fewer queries and recovers more structural information compared to prior approaches, when evaluated on DTs trained with public UCI datasets. To validate its practical feasibility, we implement BarkBeetle on a Raspberry Pi RP2350 board and perform fault injections using the Faultier voltage glitching tool. As BarkBeetle targets general DT models, we also provide an in-depth discussion on its applicability to a broader range of tree-based applications, including data stream classification, DT variants, and cryptography schemes.

Malware developers exploit the fact that most detection models focus on the entire binary to extract the feature rather than on the regions of potential maliciousness. Therefore, they reverse engineer a benign binary and inject malicious code into it. This obfuscation technique circumvents the malware detection models and deceives the ML classifiers due to the prevalence of benign features compared to malicious features. However, extracting the features from the potential malicious regions enhances the accuracy and decreases false positives. Hence, we propose a novel model named PotentRegion4MalDetect that extracts features from the potential malicious regions. PotentRegion4MalDetect determines the nodes with potential maliciousness in the partially preprocessed Control Flow Graph (CFG) using the malicious strings given by StringSifter. Then, it extracts advanced features of the identified potential malicious regions alongside the features from the completely preprocessed CFG. The features extracted from the completely preprocessed CFG mitigate obfuscation techniques that attempt to disguise malicious content, such as suspicious strings. The experiments reveal that the PotentRegion4MalDetect requires fewer entries to save the features for all binaries than the model focusing on the entire binary, reducing memory overhead, faster computation, and lower storage requirements. These advanced features give an 8.13% increase in SHapley Additive exPlanations (SHAP) Absolute Mean and a 1.44% increase in SHAP Beeswarm value compared to those extracted from the entire binary. The advanced features outperform the features extracted from the entire binary by producing more than 99% accuracy, precision, recall, AUC, F1-score, and 0.064% FPR.

Machine learning-based supervised classifiers are widely used for security tasks, and their improvement has been largely focused on algorithmic advancements. We argue that data challenges that negatively impact the performance of these classifiers have received limited attention. We address the following research question: Can developments in Generative AI (GenAI) address these data challenges and improve classifier performance? We propose augmenting training datasets with synthetic data generated using GenAI techniques to improve classifier generalization. We evaluate this approach across 7 diverse security tasks using 6 state-of-the-art GenAI methods and introduce a novel GenAI scheme called Nimai that enables highly controlled data synthesis. We find that GenAI techniques can significantly improve the performance of security classifiers, achieving improvements of up to 32.6% even in severely data-constrained settings (only ~180 training samples). Furthermore, we demonstrate that GenAI can facilitate rapid adaptation to concept drift post-deployment, requiring minimal labeling in the adjustment process. Despite successes, our study finds that some GenAI schemes struggle to initialize (train and produce data) on certain security tasks. We also identify characteristics of specific tasks, such as noisy labels, overlapping class distributions, and sparse feature vectors, which hinder performance boost using GenAI. We believe that our study will drive the development of future GenAI tools designed for security tasks.

Privacy protection laws, such as the GDPR, grant individuals the right to request the forgetting of their personal data not only from databases but also from machine learning (ML) models trained on them. Machine unlearning has emerged as a practical means to facilitate model forgetting of data instances seen during training. Although some existing machine unlearning methods guarantee exact forgetting, they are typically costly in computational terms. On the other hand, more affordable methods do not offer forgetting guarantees and are applicable only to specific ML models. In this paper, we present \emph{efficient unlearning with privacy guarantees} (EUPG), a novel machine unlearning framework that offers formal privacy guarantees to individuals whose data are being unlearned. EUPG involves pre-training ML models on data protected using privacy models, and it enables {\em efficient unlearning with the privacy guarantees offered by the privacy models in use}. Through empirical evaluation on four heterogeneous data sets protected with $k$-anonymity and $\epsilon$-differential privacy as privacy models, our approach demonstrates utility and forgetting effectiveness comparable to those of exact unlearning methods, while significantly reducing computational and storage costs. Our code is available at https://github.com/najeebjebreel/EUPG.

Advanced Encryption Standard (AES) is a widely adopted cryptographic algorithm, yet its practical implementations remain susceptible to side-channel and fault injection attacks. In this work, we propose a comprehensive framework that enhances AES-128 encryption security through controlled anomaly injection and real-time anomaly detection using both statistical and machine learning (ML) methods. We simulate timing and fault-based anomalies by injecting execution delays and ciphertext perturbations during encryption, generating labeled datasets for detection model training. Two complementary detection mechanisms are developed: a threshold-based timing anomaly detector and a supervised Random Forest classifier trained on combined timing and ciphertext features. We implement and evaluate the framework on both CPU and FPGA-based SoC hardware (PYNQ-Z1), measuring performance across varying block sizes, injection rates, and core counts. Our results show that ML-based detection significantly outperforms threshold-based methods in precision and recall while maintaining real-time performance on embedded hardware. Compared to existing AES anomaly detection methods, our solution offers a low-cost, real-time, and accurate detection approach deployable on lightweight FPGA platforms.

Cyber Threat Intelligence (CTI) has emerged as a vital complementary approach that operates in the early phases of the cyber threat lifecycle. CTI involves collecting, processing, and analyzing threat data to provide a more accurate and rapid understanding of cyber threats. Due to the large volume of data, automation through Machine Learning (ML) and Natural Language Processing (NLP) models is essential for effective CTI extraction. These automated systems leverage Open Source Intelligence (OSINT) from sources like social networks, forums, and blogs to identify Indicators of Compromise (IoCs). Although prior research has focused on adversarial attacks on specific ML models, this study expands the scope by investigating vulnerabilities within various components of the entire CTI pipeline and their susceptibility to adversarial attacks. These vulnerabilities arise because they ingest textual inputs from various open sources, including real and potentially fake content. We analyse three types of attacks against CTI pipelines, including evasion, flooding, and poisoning, and assess their impact on the system's information selection capabilities. Specifically, on fake text generation, the work demonstrates how adversarial text generation techniques can create fake cybersecurity and cybersecurity-like text that misleads classifiers, degrades performance, and disrupts system functionality. The focus is primarily on the evasion attack, as it precedes and enables flooding and poisoning attacks within the CTI pipeline.

Privacy-preserving machine learning (ML) seeks to balance data utility and privacy, especially as regulations like the GDPR mandate the anonymization of personal data for ML applications. Conventional anonymization approaches often reduce data utility due to indiscriminate generalization or suppression of data attributes. In this study, we propose an interactive approach that incorporates human input into the k-anonymization process, enabling domain experts to guide attribute preservation based on contextual importance. Using the UCI Adult dataset, we compare classification outcomes of interactive human-influenced anonymization with traditional, fully automated methods. Our results show that human input can enhance data utility in some cases, although results vary across tasks and settings. We discuss limitations of our approach and suggest potential areas for improved interactive frameworks in privacy-aware ML.

The digital evolution of connected vehicles and the subsequent security risks emphasize the critical need for implementing in-vehicle cyber security measures such as intrusion detection and response systems. The continuous advancement of attack scenarios further highlights the need for adaptive detection mechanisms that can detect evolving, unknown, and complex threats. The effective use of ML-driven techniques can help address this challenge. However, constraints on implementing diverse attack scenarios on test vehicles due to safety, cost, and ethical considerations result in a scarcity of data representing attack scenarios. This limitation necessitates alternative efficient and effective methods for generating high-quality attack-representing data. This paper presents a context-aware attack data generator that generates attack inputs and corresponding in-vehicle network log, i.e., controller area network (CAN) log, representing various types of attack including denial of service (DoS), fuzzy, spoofing, suspension, and replay attacks. It utilizes parameterized attack models augmented with CAN message decoding and attack intensity adjustments to configure the attack scenarios with high similarity to real-world scenarios and promote variability. We evaluate the practicality of the generated attack-representing data within an intrusion detection system (IDS) case study, in which we develop and perform an empirical evaluation of two deep neural network IDS models using the generated data. In addition to the efficiency and scalability of the approach, the performance results of IDS models, high detection and classification capabilities, validate the consistency and effectiveness of the generated data as well. In this experience study, we also elaborate on the aspects influencing the fidelity of the data to real-world scenarios and provide insights into its application.

Location information serves as the fundamental element for numerous Internet of Things (IoT) applications. Traditional indoor localization techniques often produce significant errors and raise privacy concerns due to centralized data collection. In response, Machine Learning (ML) techniques offer promising solutions by capturing indoor environment variations. However, they typically require central data aggregation, leading to privacy, bandwidth, and server reliability issues. To overcome these challenges, in this paper, we propose a Federated Learning (FL)-based approach for dynamic indoor localization using a Deep Neural Network (DNN) model. Experimental results show that FL has the nearby performance to Centralized Model (CL) while keeping the data privacy, bandwidth efficiency and server reliability. This research demonstrates that our proposed FL approach provides a viable solution for privacy-enhanced indoor localization, paving the way for advancements in secure and efficient indoor localization systems.

Machine learning (ML) frameworks rely heavily on pseudorandom number generators (PRNGs) for tasks such as data shuffling, weight initialization, dropout, and optimization. Yet, the statistical quality and reproducibility of these generators-particularly when integrated into frameworks like PyTorch, TensorFlow, and NumPy-are underexplored. In this paper, we compare the statistical quality of PRNGs used in ML frameworks (Mersenne Twister, PCG, and Philox) against their original C implementations. Using the rigorous TestU01 BigCrush test suite, we evaluate 896 independent random streams for each generator. Our findings challenge claims of statistical robustness, revealing that even generators labeled ''crush-resistant'' (e.g., PCG, Philox) may fail certain statistical tests. Surprisingly, we can observe some differences in failure profiles between the native and framework-integrated versions of the same algorithm, highlighting some implementation differences that may exist.

Several recent works focused on the best practices for applying machine learning to cybersecurity. In the context of malware, TESSERACT highlighted the impact of concept drift on detection performance and suggested temporal and spatial constraints to be enforced to ensure realistic time-aware evaluations, which have been adopted by the community. In this paper, we demonstrate striking discrepancies in the performance of learning-based malware detection across the same time frame when evaluated on two representative Android malware datasets used in top-tier security conferences, both adhering to established sampling and evaluation guidelines. This questions our ability to understand how current state-of-the-art approaches would perform in realistic scenarios. To address this, we identify five novel temporal and spatial bias factors that affect realistic evaluations. We thoroughly evaluate the impact of these factors in the Android malware domain on two representative datasets and five Android malware classifiers used or proposed in top-tier security conferences. For each factor, we provide practical and actionable recommendations that the community should integrate in their methodology for more realistic and reproducible settings.