Kubernetes supports multi-tenancy by allowing multiple users to share physical infrastructure. However, this design leads to co-location attacks, where an attacker deploys attack pods onto the same node as victim pods to exploit shared hardware resources. While previous work demonstrates the feasibility of such attacks, it assumes that the attacker has sufficient knowledge of the victim pod’s configuration. This paper presents PodBeater, a novel co-location attack that exploits Kubernetes’s multi-value affinity feature to steer the scheduler’s pod placement decisions. Unlike previous work, PodBeater is designed to operate even under a realistic threat model where the attacker lacks knowledge of the victim pod’s configuration. We evaluate PodBeater on both a small-scale cluster with 20 nodes and 350 pods and a large-scale cluster with 40 nodes and 800 pods, demonstrating its effectiveness under both previous and more realistic threat models. Our results show that PodBeater achieves co-location with fewer attack pods compared to previous works.
Generative AI is a technology that automatically generates various forms of content, such as text, images, and code, and is rapidly spreading with the recent popularization of services such as ChatGPT. While these technologies have greatly improved productivity and convenience, they are also emerging as a security threat, with reports of them being used to commit cybercrime, such as ransomware creation and writing phishing emails. Nevertheless, digital forensic analysis of generative AI usage has been largely limited to mobile application environments, and there is a lack of research on Web browser-based usage traces, which are most commonly utilized by real-world users. This study aims to analyze digital artifacts generated in Web browser usage environments, with a focus on ChatGPT, and to present user activity data that can be utilized in investigations.
Malicious JavaScript remains a significant cybersecurity threat. The challenge is that the detection rate of unknown malicious JavaScript is low. On the other hand, large language models (LLMs), which excel in the ability to explain codes, are expected to be effective in the detection field. However, quantitative evaluation and detailed analysis in the detection of malicious JavaScript using LLMs is still insufficient. This study proposes an offline-based LLMs approach for malicious JavaScript detection, specifically evaluating LLaMA2, LLaMA3, and the code-specialized CodeLLaMA. Our results show CodeLLaMA achieved a high F-measure of 0.89. Furthermore, we thoroughly analyzed the impact of individual tokens on detection using both gradient-based and attention-based methods. This analysis revealed that the gradient-based approach offers more explainable features for malicious tokens. These findings highlight the strong potential of LLMs, particularly code-specialized models, for effective malicious JavaScript detection and improve the interpretability of their decisions.
The number and complexity of cyber attacks are increasing, making it very important to teach people of all ages and skill levels about cybersecurity. To address this issue, teaching methods are changing to include both basic ideas and practical, hands-on activities. Following well-known guidelines like CAE-C, REWIRE, and CSEC2017, we created a hands-on course that helps trainees learn practical skills, such as analyzing network data, finding website weaknesses, detecting attacks, making systems stronger, analyzing malware, and doing digital forensics. In this paper, we introduce a simple and organized way to teach beginner-level cybersecurity using open-source tools. We developed a portable and easy-to-use practice system called TALOS, which helps people learn important topics in operating systems and network security. TALOS was used in 19 training sessions with 367 participants, including university students and the general public. Survey results showed that most trainees were very satisfied, with 81.7
Evaluating machine learning models in cybersecurity often involves datasets with an equal ratio of benign and malicious samples. Previous studies employing conventional evaluation methods have assessed performance at a single, fixed class ratio. However, this evaluation approach has a critical flaw. It fails to show how its performance degrades when the class imbalance becomes severe. This issue causes a mismatch when deploying the model in a real environment. Therefore, we propose a new evaluation methodology: the Real Environment Tolerance Curve (RETC). RETC is a systematic method to assess model robustness under varying degrees of class imbalance. It involves generating test datasets with an increasing proportion of benign data and plotting key performance metrics against the imbalance ratio. The resulting visualization, the RETC plot, enables a quantitative evaluation of the ability of a model to maintain performance in benign-dominated environments. In this study, we first establish the theoretical foundation of the RETC method through mathematical analysis, clarifying how each metric behaves as class ratios change. Next, empirical experiments on network traffic and URL datasets demonstrate that RETC reveals critical performance differences missed by conventional single-ratio evaluations. Experiments with URL datasets revealed a model whose false positive rate surged by more than three times (2.5
In modern software supply chains, Software Bill of Materials(SBOM) based scanners can lead to Alert Fatigue. The Vulnerability Exploitability eXchange(VEX) standard involves a manual process for establishing justifications such as vulnerable_code_not_in_execute_path, which states that a component is not affected because the vulnerable code is not in the execution path. Automation of this process is limited due to Node.js’s complex module system. This study proposes an automated VEX generation and vulnerability prioritization framework based on an Inter-Module Symbol Graph constructed using the static analysis tool CodeQL. This graph goes beyond package dependencies, considers Node.js’s module resolution rules, and analyzes import and export information extracted from code to precisely track which component’s API symbol calls another component’s API symbol at the code level. The proposed system uses this graph to filter out vulnerabilities unreachable from the top-level application, reducing alerts by an average of 91.1
LiDAR sensors play a critical role in obstacle recognition and path planning within autonomous driving systems. On ROS 2-based platforms, LiDAR data is delivered through a multi-stage topic structure. While existing sensor attack research has primarily focused on physical spoofing using laser reflections, such methods require high-precision control and are subject to environmental constraints. This paper proposes two LiDAR spoofing attack scenarios that exploit the open communication architecture of ROS 2 by software-level falsification of sensor node topics, disrupting the sensor data flow. The first attack injects random data into the raw input topic, distorting the entire LiDAR processing pipeline and causing collision by preventing obstacle recognition. The second attack manipulates intermediate processed data to generate fake obstacles, leading the vehicle to stop unnecessarily. These attacks demonstrate that substantial physical impacts can be induced without directly tampering with sensor signals. This study experimentally verifies the vulnerability in the sensor data structure of a real ROS 2-based UGV environment and provides foundational insights for designing countermeasures to ensure sensor data integrity and system security.
This work demonstrates that IoT botnets evade detection through covert channels and proposes a botnet detection scheme based on DoT queries. The results indicate that this scheme effectively distinguishes between website domain names, IoT domain names, and DGA-generated domain names. This approach enables network administrators to identify botnet traffic and devices within encrypted DNS traffic in real-time, helping to prevent related attacks and the spread of infection.
With the advancement of artificial intelligence, the risk of personal information breaches has become a significant concern, underscoring the importance of technologies that ensure data privacy. This study proposes an adaptive logistic regression method to enhance the performance of privacy-preserving machine learning using homomorphic encryption. The proposed technique simultaneously performs model training and training parameter optimization, enabling higher accuracy even when operating on encrypted data. Experimental results demonstrate that the adaptive homomorphic encryption logistic regression method achieves 55.92
Our study presents a bootstrap-based framework for evaluating the clustering suitability of TCP payload-derived datasets constructed under different configurations. Specifically, we investigate whet- her combining the first payload with packet header information offers a more effective proxy for client behavior than header-centric or all-payload-centric approaches. Utilizing the UNSW-NB15 dataset, we extract payload features using BIGBIRD and apply UMAP for dimensionality reduction. Using these features, we construct three types of datasets: HeaderOnly (HO), HeaderWithFirstPayload (HFP) and HeaderWithAllPayload (HAP). Clustering is performed using HDBSCAN and clustering suitability is evaluated by observing the stability of clustering results across bootstrap replicas using the Adjusted Rand Index. Our results show that HFP consistently achieves the highest clustering stability across a broad range of configurations. Interestingly, while HAP contains more information, it performs worse than HFP due to noise and overly complex features. HO, though lightweight, lacks representational capacity unless paired with specialized analysis techniques. These findings suggest that first-payload features strike the balance between information richness and resource friendliness. In addition, our work emphasizes the value of bootstrap-based approach in statistical analysis under scarce-data operating conditions. Future work will investigate the generalizability of our pipeline to real-world traffic and explore its use in designing lightweight intrusion detection systems.
The widespread use of encrypted traffic poses challenges for traditional anomaly detection systems that rely on payload inspection. While unsupervised learning methods offer scalable solutions by detecting statistical deviations in NetFlow-like data, they often suffer from high false positive rates due to their lack of interpretability. In this paper, we propose an LLM-assisted reinterpretation framework that augments anomaly detection outputs with natural language explanations. These explanations are designed to help human analysts, particularly junior-level security operators, to re-evaluate detection outcomes and correct misclassifications. Notably, our approach is detection-model-agnostic and can complement various anomaly detection backends. An empirical study with junior participants shows that LLM-guided interpretation reduces false positives significantly and increases confidence in the detection process. This demonstrates the potential of LLMs to serve as a cognitive aid in SOC environments where expert resources are limited.
For the 5G primary authentication, 5G-AKA (5G Authentication and Key Agreement) has been designed and standardized in the 3GPP consortium. In this paper, we propose a stealth mode of 5G-AKA (for short, Stealth-5G-AKA) protocol that is compatible with the current 3GPP standard. In the Stealth-5G-AKA protocol, UE and SN share a stealth anchor key in addition to a (regular) anchor key. Also, we discuss several security properties of the Stealth-5G-AKA protocol. Moreover, we compare the 5G-AKA [3] with relevant 3GPP-compatible protocols (5G-AKA’ [22], 5G-AKA-FS [23] and Stealth-5G-AKA) with respect to security properties and efficiency in terms of computation costs and communication costs.
We propose a foundational and theoretical amendment in the understanding of sender anonymity within the notion of anonymity sets with implications in schemes for anonymous connections. We introduce our extension, of the anonymity metric developed by Díaz et al., creating the new metric degree of undetectability. Using this metric, we construct and prove conditions establishing singleton set anonymity. We show yet another approach to singleton set anonymity through unlinkable pseudonyms. Thus, we demonstrate that anonymity is attainable even for singleton sets under right assumptions, countering the unquestioned core belief in this field, and opening possibilities for paradigm shifts in the design of schemes for anonymity connections that can remain robust under singleton participation, and in largly sparse networks.
This study presents a framework for assessing nation-state cyber attack capabilities by integrating MITRE ATT CK with Large Language Models. We developed a multidimensional assessment framework comprising operational capabilities, technical capabilities, and tactical maturity, enhanced by LLM-based sophistication evaluation and synergy effect models. We conducted validation using five models (GPT-4.1-nano, GPT-4o-mini, o3, o3-mini, o4-mini) with three repeated measurements per model, selecting o3 based on statistical consistency. The framework was validated using 36 versions of ATT CK data from 2018 to 2025, tracking 89 APT groups across 9 nations. This research provides security professionals and policymakers a practical tool for systematic cyber capability assessment.
The convergence of the Internet of Things (IoT) and Software-Defined Networking (SDN) has enabled scalability, dynamic resource allocation, and centralized management of SDN-IoT infrastructures. However, the integration introduces critical security challenges such as cross-layer threats and IoT-specific exploits. While Machine Learning (ML)–based Intrusion Detection Systems (IDS) offer promising defense, their effectiveness is hindered by inadequate and outdated datasets that lack realistic SDN control-plane dynamics and IoT heterogeneity within integrated SDN-IoT environments. Critically, no existing dataset simultaneously captures three dimensions, including SDN control-plane behaviour, IoT device traffic, and cross-layer attack vectors essential for securing modern converged networks. To address the gap, we propose a novel dataset generated from a real-world SDN-IoT testbed. The dataset is validated using five ML models with two classification schemes. The experimental results show high detection performance, with 100
In this paper, we propose PoC Hash, a post-quantum cryptographic hash function that operates natively in Hamilton’s quaternion algebra over a finite field, with security based on the hardness of the Multivariate Cubic (MC) problem. Each compression step takes two input quaternions, forms their non-abelian commutator, multiplies the result by the conjugate of a third quaternion, raises the product to the third power, and finally adds a fixed non-central offset. This procedure costs only five quaternion multiplications (roughly 80 finite-field multiplies), so software performance is expected to be comparable to the SHA-2 family while scaling efficiently on SIMD and GPU hardware. Finding a pre-image for a t-block message reduces to solving an (n,m,q)=(6t,4t,q≈ 2^128) multivariate cubic system. PoC Hash thus combines prime-field efficiency with a non-commutative hardness foundation and meets the 256-bit security goal.
The widespread use of text-to-image diffusion models like Stable Diffusion raises concerns about inadvertent leakage of confidential information through generated images. While prior studies mainly focused on pretraining data leakage using captions provided during training, they often overlooked realistic scenarios. This paper experimentally demonstrates that fine-tuning Stable Diffusion with a small set of confidential images can lead to unintended and intentional Data extraction, even when general prompts without explicit object keywords are used. We design a multi-modal detection framework that analyzes input prompts, latent representations, and output images to identify potential data leaks. Furthermore, we propose an output-stage Leakguard system that filters confidential images in real-time, reducing the leakage rate by over 71.5
SAND is an AND-RX-based lightweight block cipher proposed by Chen et al. There are two variants of SAND, namely SAND-64 and SAND-128, due to structural differences. In this paper, we search for impossible differential distinguishers of SAND-64 using Constraint Programming (CP) and reveal 56 types of impossible differential distinguishers up to 11 rounds. Furthermore, we demonstrate a key recovery attack on 16-round SAND-64. The complexities for the attack require 2^63 data, 2^96 encryptions, and 2^62 bytes of memory, respectively.
The increasing number of mobile devices and the expansion of IoT (Internet of Things) and digital systems have made them targets for more sophisticated Android malicious apps. Existing detection techniques often use randomly generated events using ‘monkey’ tools, but this has limitations for apps that rely on user input for malicious behavior. In this paper, we extracted system calls and changed them to sequences to classify families of malicious apps by type with the DTW(Dynamic Time Warping) Algorithm. We evaluated using the AndroZoo dataset and found that family classification is possible for types of Trojan, Adware, and Exploit. In this study, we used view tree-based interaction for classification without using the ‘monkey’ tool to extract system call logs. We also extracted all logs through system call filtering to extract more malware behavior-focused system call logs. In the future, we plan to use more sophisticated interaction tools and ML and DL for classification and detection.
Messenger, a messaging service, is widely used in the United States and around the world, with 947 million monthly active users as of February 2025. In December 2023, a default implementation of end-to-end encryption in Messenger and two accompanying technical white papers were released. Labyrinth Encrypted Message Storage Protocol is one of the protocols presented in these technical white papers and is intended to securely share message history between devices associated with each Messenger account. However, there has been no rigorous discussion of the security of Labyrinth to date, and the security model claimed in the technical white paper has some ambiguity. In this paper, we give the first formal verification of Labyrinth. First, we verify the confidentiality and integrity of the stored messages as the security claim in the specification (i.e., only considering outside adversaries). Next, considering the case of a device that has been revoked from an account and abused, we exhaustively verify under what conditions it is possible to break the confidentiality and integrity of a message. As a result of the verification, we clarify the conditions under which the confidentiality and integrity of stored messages can be broken by abusing a revoked device and the specific procedures for doing so while satisfying all the security claims made by the specification.
