This paper presents Map Reduce Graph (MRG), a novel unsupervised method for modeling and securing HTTP REST APIs. MRG learns API structure from real-world traffic without prior knowledge or labels, automatically generating OpenAPI-compliant documentation by reconstructing routes, methods, and parameter formats. MRG enables real-time updates, explainable visualization, and anomaly detection, helping identify undocumented or evolving behaviors. It detects malformed requests, structural deviations, and injection attacks using graph-based validation and a deep autoencoder for payload analysis. Compared to state-of-the-art methods like HRAL and FT-ANN, MRG achieves up to 11.4

The growing dominance of encrypted network traffic and modern encryption protocols (TLS 1.3, QUIC, DoH) poses significant challenges for accurate network classification, particularly as many existing approaches rely on text- or image-based representations, which fail to adequately capture the inherent structural relationships present in network communication-relationships that are more naturally represented as graphs. In this work, we introduce GraphMux, a graph-based framework that leverages line graph transformations to fuse multiple graph views into a unified representation. We also present three graph-based flow representations (TIG+Chain, StarBurst, and 2Chain) designed to capture both temporal burst dynamics and client-server interaction patterns, using only packet time, direction, and length information, without incorporating any unencrypted statistical features. We evaluate our approach on three datasets: two academic datasets (UTMobileNetTraffic2021 and QUIC PCAP) and a commercial dataset (Flash), using four graph embedding architectures. Across all datasets, GraphMux consistently achieves superior performance, and the proposed graph constructions often yield the best results. Additional experiments examining attribute-selection strategies reveal a strong positive relationship between well-aligned feature assignments and classification accuracy, underscoring the importance of principled attribute design when constructing graph representations for encrypted traffic.

Model repositories such as Hugging Face increasingly distribute machine learning artifacts serialized with Python's pickle format, exposing users to remote code execution (RCE) risks during model loading. Recent defenses, such as PickleBall, rely on per-library policy synthesis that requires complex system setups and verified benign models, which limits scalability and generalization. In this work, we propose a lightweight, machine-learning-based scanner that detects malicious Pickle-based files without policy generation or code instrumentation. Our approach statically extracts structural and semantic features from Pickle bytecode and applies supervised and unsupervised models to classify files as benign or malicious. We construct and release a labeled dataset of 727 Pickle-based files from Hugging Face and evaluate our models on four datasets: our own, PickleBall (out-of-distribution), Hide-and-Seek (9 advanced evasive malicious models), and synthetic joblib files. Our method achieves 90.01

The dominance of encryption and the relentless growth of network traffic have rendered traditional deep packet inspection ineffective and made retraining-based machine learning classifiers operationally costly, as every new application or threat class requires a full model rebuild. In order to overcome this challenge, we present an ultra-fast encrypted traffic classification system based on Classification By Retrieval (CBR): each flow is represented by a fixed-length feature vector extracted from only the first few packets, indexed in a HNSW approximate nearest-neighbor structure, and classified by retrieving its nearest labeled neighbor. New classes are absorbed by inserting a handful of labeled vectors into the running index, with no retraining and no downtime, and out-of-distribution traffic is detected within the same retrieval step using a single global distance threshold. We evaluate handcrafted statistical features and learned CNN and MLP embeddings against a Random Forest baseline on the public CESNET-TLS22 dataset (187 application classes) and a proprietary cellular dataset (Radcom). With CNN embeddings, CBR matches Random Forest accuracy (macro F1 of 0.923 versus 0.925 on CESNET) while training roughly ten times faster, adapts to new classes two to three orders of magnitude faster (seconds versus thousands of seconds), and detects out-of-distribution flows with up to 0.90 AUROC. These results show that retrieval-based classification is a practical and adaptable alternative for real-time encrypted traffic analysis in dynamic network environments.

AI model repositories such as Hugging Face and TensorFlow Hub have become an attractive surface for steganographic malware: attackers exploit the redundancy in float32 weights to embed payloads while preserving model accuracy. Existing AI-model steganalysis methods require tens of thousands of labeled training samples and only detect attacks at high embedding rates (>= 50%), limiting their practical utility. We address both gaps with a few-shot learning approach. We propose a novel parameter-position-stable image representation, Grayscale-Fourpart (GF), that maps float32 weights to a square grayscale image, and pair it with a metric-learning few-shot CNN detector. The detector trains from as few as 6 model files and consistently flags attacks down to 25% embedding rate, with 6% in some cases. We benchmark against a seven-baseline matrix spanning two prior academic method, the canonical raw-byte 1D-CNN paradigm, and four threshold-based statistics, and identify the conjoint conditions under which the simpler baselines collapse and ours retains accuracy. The trained detectors transfer to novel out-of-distribution spread-spectrum attacks despite training only on LSB perturbations. A deployment-feasibility study shows that GF feature extraction scales linearly to 108 parameters at approximate to 0.49 s and similar to 1.77 GiB peak memory, similar to 352 & times; faster than the strongest prior baseline at the same scale, making this, to our knowledge, the first AI-model steganalysis pipeline practical for repository-scale deployment. The full code framework, including baseline reproductions, is released as open-source.

Microservices have become the backbone of large IT enterprises due to their ability to scale, recover from failures, and adapt to dynamic workloads within cloud-native architectures. Ensuring the reliability of such systems requires accurate and efficient Root Cause Analysis (RCA) to identify and address faults promptly, minimizing disruptions and maintaining service quality. However, current RCA techniques often focus on isolated levels, such as metrics or services, and rely heavily on predefined thresholds or statistical methods, limiting their effectiveness in complex and interconnected systems. We propose a novel unsupervised RCA method based on Temporal Graph Networks (TGNs) to address these limitations. Our approach models the temporal and structural relationships within microservices through the integrated examination of multimodal data to accurately identify faults. Evaluated on an open-source dataset, our method shows superior accuracy compared to SOTA approaches with at least 2% improvement in PR@1, PR@2, and PR@5.

Application Programming Interfaces (APIs) are essential in software development, enabling web services, mobile apps, and microservices. However, their widespread use introduces significant security risks, highlighting the importance of API security. This paper presents HTTP REST API Learning (HRAL), a novel unsupervised anomaly detection approach that models the structure and behavior of API endpoints directly from network traffic, without relying on predefined rules or documentation. HRAL enables robust detection of malicious activity by understanding how APIs behave and flagging deviations as potential threats. We evaluate HRAL across varying levels of OpenAPI documentation detail and compare it with existing techniques. HRAL achieves strong performance, with an average recall of 82.07

The increasing sophistication and frequency of cyberattacks have made Network Intrusion Detection Systems (NIDS) a critical component of modern cybersecurity. This work presents D-MAGIC, a novel real-time NIDS that leverages zero-shot learning and graph-based dynamic clustering to detect both known and unknown threats. Unlike traditional systems that rely on labeled datasets and predefined attack signatures, D-MAGIC operates unsupervised, identifying anomalies by detecting deviations from normal network behavior. D-MAGIC employs dynamic clustering to identify coordinated attacks and emerging threats by embedding network flow relationships into a graph structure and clustering similar patterns. Additionally, a second anomaly detection method, called SAGA, is introduced. SAGA enhances detection by using approximate nearest neighbor (ANN) analysis to evaluate isolated nodes, which clustering alone cannot classify. By leveraging information from a node’s neighbors, SAGA improves the identification of subtle or emerging threats. This hybrid approach ensures more comprehensive detection, capturing both clustered anomalies and isolated outliers. Experimental results on the CIC-IDS-2017 and CSE-CIC-IDS-2018 datasets demonstrate that D-MAGIC achieves up to 12% improvement in F1 score over state-of-the-art methods, significantly reduces false positives, and ensures rapid, real-time detection with minimal latency. Furthermore, on the CIC-IDS-2017 dataset, SAGA achieves up to a 6% improvement in accuracy, showcasing its effectiveness in handling diverse attack scenarios.

The increasing sophistication and frequency of cyberattacks have made Network Intrusion Detection Systems (NIDS) a critical component of modern cybersecurity. This work presents D-MAGIC, a novel real-time NIDS that leverages zero-shot learning and graph-based dynamic clustering to detect known and unknown threats. Unlike traditional systems that rely on labeled datasets and predefined attack signatures, D-MAGIC operates unsupervised, identifying anomalies by detecting deviations from normal network behavior. By embedding the relationships between network flows into a graph structure and dynamically clustering similar patterns, D-MAGIC can detect coordinated attacks and emerging threats with minimal delay. Experimental results on the CIC-IDS-2017 and CSE-CIC-IDS-2018 datasets demonstrate that D-MAGIC achieves an improvement of up to 12 % based on the standard F1 score compared to state-of-the-art methods, while significantly reducing false positives and ensuring rapid, real-time detection with minimal detection latency.

Latency is the most impactful on fairness and Quality of Experience (QoE) in First-Person Shooter (FPS) games. High latency degrades the QoE of players, who may leave the game if unsatisfied with their QoE. Modern FPS games make great efforts to maintain an excellent QoE even under a poor Internet connection with high latency. Those efforts include the wide distribution of game servers and many software optimizations to smooth the effect of lags in the games. This study aims to provide insights into QoE estimation for network-intensive applications by examining one of the most prominent FPS games of the past two decades: Call of Duty. We observed that the game dynamically adjusts its network traffic behavior, including packet size and transmission rate, in response to variations in network quality. However, the ISP does not have this capability since the network traffic is encrypted; observing the game’s network traffic does not expose its nature and most certainly does not expose the game player’s intensity, latency, or QoE. We propose a novel technique for estimating latency and QoE in FPS games from an ISP-level perspective. In our evaluation, the model detected problematic latency in near real-time with 81% accuracy and an 80% F1 on a 10-second window, highlighting a trade-off between responsiveness and predictive performance. The dataset generated for this study is publicly available to support further research.

The increasing popularity of online services has made Internet Traffic Classification a critical field of study. However, the rapid development of internet protocols and encryption limits usable data availability. This paper addresses the challenges of classifying encrypted internet traffic, focusing on the scarcity of open-source datasets and limitations of existing ones. We propose two Data Augmentation (DA) techniques to synthetically generate data based on real samples: Average augmentation and MTU augmentation. Both augmentations are aimed to improve the performance of the classifier, each from a different perspective: The Average augmentation aims to increase dataset size by generating new synthetic samples, while the MTU augmentation enhances classifier robustness to varying Maximum Transmission Units (MTUs). Our experiments, conducted on two well-known academic datasets and a commercial dataset, demonstrate the effectiveness of these approaches in improving model performance and mitigating constraints associated with limited and homogeneous datasets. Our findings underscore the potential of data augmentation in addressing the challenges of modern internet traffic classification. Specifically, we show that our augmentation techniques significantly enhance encrypted traffic classification models. This improvement can positively impact user Quality of Experience (QoE) by more accurately classifying traffic as video streaming (e.g., YouTube) or chat (e.g., Google Chat). Additionally, it can enhance Quality of Service (QoS) for file downloading activities (e.g., Google Docs).

The exponential growth of network traffic in modern telecommunications has made the task of analyzing flow data for effective bandwidth estimation increasingly complex and resource-intensive. Accurate estimation of effective throughput is essential for a wide range of network management tasks, including dynamic traffic shaping, congestion control, quality optimization, and anomaly detection. However, traditional packet-based solutions, which rely on inspecting packets across the entire flow duration, demand substantial computational resources and memory, thereby reducing system performance and limiting scalability under high traffic volumes. This paper introduces an efficient sampling method based on linear regression and error reduction to accurately estimate effective throughput. Unlike general sampling methods such as random or systematic sampling, which do not account for actual network conditions or flow dynamics, our approach leverages real-time flow behavior to guide the sampling process. By focusing only on the most informative and impactful portions of each flow, the method significantly reduces the amount of data that needs to be processed while maintaining high estimation accuracy, allowing accurate throughput estimation using less than 15% of the available data, with an estimation error of no more than 10%. These advantages make the proposed method highly suitable for scalable and efficient deployment in diverse real-time network monitoring and traffic analysis environments.

File type classification is critical in digital forensics, and file carving. However, the increasing diversity of file formats challenges accurate classification. Traditional methods rely on hand-crafted features or compact neural networks but face long training times, limited training data, and lower accuracy. This paper introduces three novel, content-based file-type classification approaches to address these challenges. These approaches improve accuracy and streamline the integration of new file types using pre-trained models, enhancing both speed and reliability. The first approach utilizes Natural Language Processing (NLP) with a transformer architecture, while the second combines statistical features with a pre-trained model via transfer learning. These methods achieved accuracy rates of 72.4 % and 69.2 %, respectively, surpassing state-of-the-art Convolutional Neural Network (CNN) models. The third approach employs one-shot learning, achieving 100 % accuracy in several scenarios, enabling efficient training with minimal data.

Pretrained deep learning model sharing holds tremendous value for researchers and enterprises alike. It allows them to apply deep learning by fine-tuning models at a fraction of the cost of training a brand-new model. However, model sharing exposes end-users to cyber threats that leverage the models for malicious purposes. Attackers can use model sharing by hiding self-executing malware inside neural network parameters and then distributing them for unsuspecting users to unknowingly directly execute them, or indirectly as a dependency in another software. In this work, we propose NeuPerm, a simple yet effec- tive way of disrupting such malware by leveraging the theoretical property of neural network permutation symmetry. Our method has little to no effect on model performance at all, and we empirically show it successfully disrupts state-of-the-art attacks that were only previously addressed using quantization, a highly complex process. NeuPerm is shown to work on LLMs, a feat that no other previous similar works have achieved. The source code is available at https://github.com/danigil/NeuPerm.git.

This paper examines the challenges in distributing AI models through file transfer mechanisms. Despite advancements in security measures, vulnerabilities persist, necessitating a multi-layered approach to mitigate risks effectively. The physical security of model files is critical, requiring stringent access controls and attack prevention solutions. This paper proposes a novel solution architecture that protects the model architecture and weights from attacks by using Moving Target Defense (MTD), which obfuscates the model, preventing unauthorized access, and enabling detection of changes to the model. Our method is shown to be effective at detecting alterations to the model, such as steganography; it is faster than encryption (0.1 seconds to obfuscate vs. 18 seconds to encrypt for a 2500 MB model), and it preserves the accessibility of the original model file format, unlike encryption. Finally, our code is available at https://github.com/ArielCyber/AI-model-MTD.git.

Application Programming Interface (API) Injection attacks refer to the unauthorized or malicious use of APIs, which are often exploited to gain access to sensitive data or manipulate online systems for illicit purposes. Identifying actors that deceitfully utilize an API poses a demanding problem. Although there have been notable advancements and contributions in the field of API security, there remains a significant challenge when dealing with attackers who use novel approaches that don't match the well-known payloads commonly seen in attacks. Also, attackers may exploit standard functionalities unconventionally and with objectives surpassing their intended boundaries. Thus, API security needs to be more sophisticated and dynamic than ever, with advanced computational intelligence methods, such as machine learning models that can quickly identify and respond to abnormal behavior. In response to these challenges, we propose a novel unsupervised few-shot anomaly detection framework composed of two main parts: First, we train a dedicated generic language model for API based on FastText embedding. Next, we use Approximate Nearest Neighbor search in a classification-by-retrieval approach. Our framework allows for training a fast, lightweight classification model using only a few examples of normal API requests. We evaluated the performance of our framework using the CSIC 2010 and ATRDF 2023 datasets. The results demonstrate that our framework improves API attack detection accuracy compared to the state-of-the-art (SOTA) unsupervised anomaly detection baselines.

This paper examines the challenges in distributing AI models through model zoos and file transfer mechanisms. Despite advancements in security measures, vulnerabilities persist, necessitating a multi-layered approach to mitigate risks effectively. The physical security of model files is critical, requiring stringent access controls and attack prevention solutions. This paper proposes a novel solution architecture composed of two prevention approaches. The first is Content Disarm and Reconstruction (CDR), which focuses on disarming serialization attacks that enable attackers to run malicious code as soon as the model is loaded. The second is protecting the model architecture and weights from attacks by using Moving Target Defense (MTD), alerting the model structure, and providing verification steps to detect such attacks. The paper focuses on the highly exploitable Pickle and PyTorch file formats. It demonstrates a 100 repositories and actual malware attacks from the HuggingFace model zoo.

Post-quantum cryptography (PQC) is expected to revolutionize secure communications in next-generation digital ecosystems. Previous and ongoing activities demonstrate that different PQC algorithms significantly impact traffic latency, but they do not yet provide a scheme to assess the existence of the PQC algorithm or its identification when encrypted traffic is analyzed for traffic engineering purposes. Hence, this work is the first to propose a novel PQClass pipeline for classifying encrypted Internet traffic of recently NIST-approved PQC algorithms. Hence, it establishes solid grounds for enabling engineers to optimize their networks and, in parallel, for cybersecurity practitioners to familiarise themselves with PQC algorithmic properties for enhancing or devising security architectures in diverse setups. Our pipeline demonstrates impressive performance on real-world data, achieving 86% accuracy in detecting the presence of a PQC algorithm and 91% and 98% accuracy in identifying the browser and OS, respectively, based on PQC-based traffic.

In today’s digital landscape, ensuring the security of cloud environments is critical for organizational resilience, growth, and operational efficiency. As cloud services become more prevalent, so do sophisticated attacks targeting cloud users, making early detection essential. This paper introduces a novel time-based embedding approach for Cloud Services Graph-based Anomaly Detection (CS-GAD) that leverages a Graph Neural Network (GNN) to detect anomalous user behavior. We propose a dynamic tripartite graph to model interactions among users, actions, and cloud services over time. Using behavioral patterns, our GNN generates user embeddings to enable early detection of anomalies. We evaluate this approach on a novel dataset simulating five real-world attacks: cryptojacking, billing abuse, lateral movement, monitor exploitation, and service targeting. The dataset comprises 107,116 Application Programming Interface (API) calls over 32 days, tracking 79 AWS services, with attacks embedded within legitimate cloud traffic. Our results demonstrate that the proposed method achieves a lower false positive rate and higher detection accuracy than a prevailing method, as evidenced by improved accuracy, precision, recall, and F1-score.

Recent advancements in Internet protocols, including DNS over HTTPS (DoH) and Encrypted Service Name Indicators (ESNI), are making traditional Deep Packet Inspection (DPI) engines obsolete. Consequently, there is a growing need for next-generation traffic classification using artificial intelligence (AI). While DPI automatically categorizes unknown traffic as ’other,’ AI-based models cannot automatically handle unknown or Out-of-Distribution (OOD) traffic. AI models must effectively detect and classify OOD traffic to ensure robustness, reliability, and accuracy in real-world applications; however, current research often fails to address the challenges of OOD detection.In this paper, we evaluate various state-of-the-art OOD detection techniques for internet traffic classification and explore the drawbacks and advantages of using different threshold levels for the model’s tolerance for OOD. Our findings reveal that varying rejection rates have distinct effects on OOD techniques, leading to a change in the optimal strategy for achieving dependable and precise detection across diverse OOD scenarios. We demonstrate that adjusting rejection rates from 10% to 30% can significantly improve the True Detection Rate (TDR) by up to 50%, while the False Detection Rate (FDR) may increase by less than 10%. Moreover, we emphasize that rejection-rate-based evaluation is pivotal for next-generation flow classification, promising a substantial reduction in FDR through rigorous methodological assessment.