Autonomous offensive agents often fail to transfer beyond the networks on which they are trained. We isolate a minimal but fundamental shift – unseen host/subnet IP reassignment in an otherwise fixed enterprise scenario – and evaluate attacker generalization in the NetSecGame environment. Agents are trained on five IP-range variants and tested on a sixth unseen variant; only the meta-learning agent may adapt at test time. We compare three agent families (traditional RL, adaptation agents, and LLM-based agents) and use action-distribution-based behavioral/XAI analyses to localize failure modes. Some adaptation methods show partial transfer but significant degradation under unseen reassignment, indicating that even address-space changes can break long-horizon attack policies. Under our evaluation protocol and agent-specific assumptions, prompt-driven pretrained LLM agents achieve the highest success on the held-out reassignment, but at the cost of increased inference-time compute, reduced transparency, and practical failure modes such as repetition/invalid-action loops.

This paper explores coordinated deception strategies by synchronizing defenses across coupled cyber and physical systems to mislead attackers and strengthen defense mechanisms. We introduce a Stackelberg game framework to model the strategic interaction between defenders and attackers, where the defender leverages CVSS-based exploit probabilities and real-world vulnerability data from the National Vulnerability Database (NVD) to guide the deployment of deception. Cyber and physical replicas are used to disrupt attacker reconnaissance and enhance defensive effectiveness. We propose a CVE-based utility function to identify the most critical vulnerabilities and demonstrate that coordinated multilayer deception outperforms single-layer and baseline strategies in improving defender utility across both CVSS versions.

Stackelberg Security Games (SSG) assume that an attacker observes the defender's strategy and chooses the target that maximizes their expected utility perfectly. In most realistic applications this is not plausible, and in the case of cyber deception (e.g., using decoys) the purpose of the game is to induce uncertainty and mistakes. Quantal response is a common way to represent noise and mistakes in decision-making; here it replaces perfect best-response with a logit choice with rationality parameter λ and results in a generalized Quantal Stackelberg Equilibrium (QSE), which recovers the classical solution exactly as λ→∞. We conduct a deeper analysis of how QSE can function as a generalized form of insurance against a variety of forms of model specification error/uncertainty; our analysis shows that QSE provides a practical way to address the important role of tie-breaking rules and model uncertainty in SSG from both a theoretical and practical perspective. We conduct an empirical evaluation in a cybersecurity case study with two networks and real vulnerabilities drawn from CVE and scored using the Common Vulnerability Scoring System (CVSS). QSE beats Stackelberg in realized defender utility spanning 144 scenarios with specification errors and 25 parameter configurations, with gains of 46% to 175% showing a substantial advantage in a wide variety of realistic cases.

Out-of-distribution (OOD) detection remains a critical challenge in malware classification due to the substantial intra family variability introduced by polymorphic and metamorphic malware variants. Most existing deep learning–based malware detectors rely on closed-world assumptions and fail to adequately model this intra-class variation, resulting in degraded performance when confronted with previously unseen malware families. This paper presents MADOOD, a novel two-stage, cluster driven deep learning framework for robust OOD malware detection and classification. In the first stage, malware family embeddings are modeled using class conditional spherical decision boundaries derived from Gaussian Discriminant Analysis (GDA), enabling statistically grounded separation of indistribution and OOD samples without requiring OOD data during training. Z score based distance analysis across multiple class centroids is employed to reliably identify anomalous samples in the latent space. In the second stage, a deep neural network integrates cluster-based predictions, refined embeddings, and supervised classifier outputs to enhance final classification accuracy. Extensive evaluations on benchmark malware datasets comprising 25 known families and multiple novel OOD variants demonstrate that MADOOD significantly outperforms state of the art OOD detection methods, achieving an AUC of up to 0.911 on unseen malware families. The proposed framework provides a scalable, interpretable, and statistically principled solution for real-world malware detection and anomaly identification in evolving cybersecurity environments.

Out-of-distribution (OOD) detection in malware classification remains a significant challenge due to the high intra-class variability among malware variants within the same family. Existing deep learning approaches often overlook this intra-family variation, resulting in suboptimal detection performance. This research proposes a two-stage framework that addresses this limitation by incorporating Gaussian discriminant analysis (GDA) into deep neural networks to model spherical decision boundaries around malware families in the embedding space. The first stage employs unsupervised cluster analysis to determine whether a test sample is in-distribution or out-of-distribution, using z-score-based statistical analysis for reliable outlier detection. The second stage introduces a deep learning model trained on refined embeddings from the initial stage, using predictions from both the cluster analysis and a primary classifier to enhance final prediction accuracy. Evaluation on a dataset comprising 25 malware families and novel OOD samples demonstrates superior performance, achieving an AUC of 0.911 for OOD detection. This approach significantly improves the distinguishability of OOD samples and offers a scalable and statistically grounded method for robust out-of-distribution malware classification and anomaly detection in security contexts.

Research on decision making has identified a large number of systematic patterns in how humans process information and make choices; these are often called cognitive biases because they reflect deviations from perfectly rational choices. Cyber attackers make extensive use of vulnerabilities in decision making (e.g., in many types of social engineering attacks), but cyber defenses rarely try to exploit similar vulnerabilities in cyber attackers. We evaluate the role of Denominator Neglect (DN) and the Hot Hand Fallacy (HHF) in decision made by cyber attackers; both of these are effects of using the Representativeness Heuristic. Our experimental design combines established tasks used to measure these effects with Cyber Isomorphs (CI)-novel tasks presented in a cyber context that mirror tasks use in the literature to the extent possible. This allows us to test whether the effects observed in the literature will translate consistently into more realistic cyber attack scenarios. We conducted a survey comprising both tasks with 85 participants screened for basic cybersecurity knowledge. We observe that participants make choices consistent with denominator neglect and the hot hand fallacy in both variants of the task, with statistically significant differences from the rational baseline in most conditions. The effect size for HHF was particularly strong with biased choices in more than 75% of cases. We also observe that the effects are often stronger in the cyber tasks than the original tasks.

The emergence of few-shot learning as a potential approach to address the problem of data scarcity by learning underlying pattern from a few training sample had so far given a mix-result especially on the suitability of model-agnostic meta learning, transfer learning, and optimization strategy to rapidly learn valid information from few sample. In this research, we did an in-depth evaluation of meta- learning to determine their plausibility and suitability for previously unknown cyberattack detection by first retrieving the original research artifacts of current state of the art meta learning to repeat the experiment with original dataset before replicating the experiment with two different malware dataset which had not been previously done with meta-transfer learning. On each of the experiments, meta-transfer learning gave good results on digital character recognition dataset but abysmal result on Malimg and Malevis malware images datasets thereby indicating its unreliability for detecting cyberattacks and the need for an improvement to the state-of-the-art meta transfer learning towards a better attack detection. Transfer learning performance is independent on imbalance and hence does not influence its performance since both malware dataset used for this experiment result in high validation loss and balancing the dataset doesn't result in reduced validation loss,the successful learning transfer seen on digital character recognition dataset is not unconnected to the fact that several languages have similar characters and digits thereby enhancing the successful learning transfer unlike malware datasets, and more importantly the finding that current meta-learning transfer approach doesn't generalize well on malware dataset and hence not suitable for detecting previously unseen out-of-distribution attack.

Cyber deception strategies play a vital role in fortifying computer networks and systems against malicious intrusions by misleading attackers. This study presents a novel approach to enhance cyber deception strategies by integrating multinomial Bayesian logistic regression, data interpretation, and feature deception techniques. Our method leverages data-driven learning to analyze uncertainty in the data, effectively disguising one asset as another. We propose a deception framework aimed at increasing entropy in the recognition of a target by employing data learning in cyber deception to mislead attackers and safeguard critical systems and networks. To evaluate our framework’s performance, we employ various statistical measures and assess its predictive accuracy. Our deception results illustrate a hindrance in achieving high predictive accuracy and underscore the targeted deception strategies facilitated by our approach.

Phishing is one of the most effective ways in which cybercriminals get sensitive details such as credentials for online banking, digital wallets, state secrets, and many more from potential victims. They do this by spamming users with malicious URLs with the sole purpose of tricking them into divulging sensitive information which is later used for various cybercrimes. In this research, we did a comprehensive review of current state-of-the-art machine learning and deep learning phishing detection techniques to expose their vulnerabilities and future research direction. For better analysis and observation, we split machine learning techniques into Bayesian, non-Bayesian, and deep learning. We reviewed the most recent advances in Bayesian and non-Bayesian-based classifiers before exploiting their corresponding weaknesses to indicate future research direction. While exploiting weaknesses in both Bayesian and non-Bayesian classifiers, we also compared each performance with a deep learning classifier. For a proper review of deep learning-based classifiers, we looked at Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), and Long Short Term Memory Networks (LSTMs). We did an empirical analysis to evaluate the performance of each classifier along with many of the proposed state-of-the-art anti-phishing techniques to identify future research directions, we also made a series of proposals on how the performance of the under-performing algorithm can improved in addition to a two-stage prediction model

Honeypots play a crucial role in implementing various cyber deception techniques as they can mislead attackers and divert them away from valuable assets. The topology of tactical networks changes over time due to the topographical environment and node mobility. An impactful strategic placement of honeypots in tactical networks should essentially consider not only network aspects but also attackers' preferences. To this end, we propose a game-theoretic approach that models an attack-defense scenario and develops an optimal honeypot allocation strategy for the defender. Our approach takes into consideration the changes in network connectivity and the specific features and criticality of different nodes. In particular, we introduce a two-player dynamic game model that explicitly incorporates the future state evolution resulting from changes in connectivity. The defender's objective is twofold: to maximize the likelihood of the attacker hitting a honeypot and to minimize the cost associated with deception in terms of the reconfiguration cost of honeypot re-allocation due to mobility. We present an iterative algorithm to find Nash equilibrium strategies. Unfortunately, this class of games suffers from the curse of dimensionality due to the size of the large state space. Therefore, we evaluate the scalability of our algorithm and provide a compact state space that shows a significant reduction in terms of runtime. Finally, we relax the assumption that the defender has full up-to-date knowledge of the network topology. We present a sensitivity-analysis-based approach to quantify the impact of imperfect information in terms of the defender reward. We validate our approach numerically via extensive simulations, demonstrating that our game model successfully enhances network security and is more scalable.

Domain Generation Algorithms (DGAs) are malicious techniques used by malware to dynamically generate seemingly random domain names for communication with Command & Control (C&C) servers. Due to the fast and simple generation of DGA domains, detection methods must be highly efficient and precise to be effective. Large Language Models (LLMs) have demonstrated their proficiency in real-time detection tasks, making them ideal candidates for detecting DGAs. Our work validates the effectiveness of fine-tuned LLMs for detecting DGAs and DNS exfiltration attacks. We developed LLM models and conducted comprehensive evaluation using a diverse dataset comprising 59 distinct real-world DGA malware families and normal domain data. Our LLM model significantly outperformed traditional natural language processing techniques, especially in detecting unknown DGAs. We also evaluated its performance on DNS exfiltration datasets, demonstrating its effectiveness in enhancing cybersecurity measures. To the best of our knowledge, this is the first work that empirically applies LLMs for DGA and DNS exfiltration detection.

N-shot learning has emerge in recent year as potential learning approach to solve the problem of data scarcity by learning underlying pattern from a few training sample. Despite recent state-of-the-art research on model-agnostic metal learning, transfer learning, and optimization strategy to rapidly learn valid information from few sample, there remains a big challenge on an actual out-of-distribution zero-day without any similarity to previously known malware family or new variant of an existing malware family. This ultimately questions the effectiveness of current state-of-the-art few-shot learning approach. In this research, we did an in-depth investigation into the performance of state-of-the-art Zero-shot, Single-shot, and few-shot learning approaches on zero-day out-of-distribution malware attack detection based on their static properties using Malimg and Malevis malware dataset. We ensure our model was aware of an out-of-distribution class during training while varying the number of samples in the out-of-distribution class accordingly zero-shot(no sample), single-shot (1 sample), few-shot(5 samples) while using confusion matrix to get the actual number of correct prediction on out-of-distribution malware validation samples. we assert that the model should be smart enough to detect and classify previously unseen data into an empty family as an out-of-distribution considering that the model was made to be aware of the existence of such distribution during training. Result shows 0, 0, and 3 correct out-of-distribution predictions on Zero-shot, single-shot, and few-shot experiments respectively, thereby showing limitation of the current state-of-the-art N-shot approaches on out-of-distribution attack.

To secure computers and information systems from attackers taking advantage of vulnerabilities in the system to commit cybercrime, several methods have been proposed for real-time detection of vulnerabilities to improve security around information systems. Of all the proposed methods, machine learning had been the most effective method in securing a system with capabilities ranging from early detection of software vulnerabilities to real-time detection of ongoing compromise in a system. As there are different types of cyberattacks, each of the existing state-of-the-art machine learning models depends on different algorithms for training which also impact their suitability for detection of a particular type of cyberattack. In this research, we analyzed each of the current state-of-theart machine learning models for different types of cyberattack detection from the past 10 years with a major emphasis on the most recent works for comparative study to identify the knowledge gap where work is still needed to be done with regard to detection of each category of cyberattack

The ever-evolving ways attacker continues to im prove their phishing techniques to bypass existing state-of-the-art phishing detection methods pose a mountain of challenges to researchers in both industry and academia research due to the inability of current approaches to detect complex phishing attack. Thus, current anti-phishing methods remain vulnerable to complex phishing because of the increasingly sophistication tactics adopted by attacker coupled with the rate at which new tactics are being developed to evade detection. In this research, we proposed an adaptable framework that combines Deep learning and Randon Forest to read images, synthesize speech from deep-fake videos, and natural language processing at various predictions layered to significantly increase the performance of machine learning models for phishing attack detection.

Protecting against adversarial attacks is a common multiagent problem. Attackers in the real world are predominantly human actors, and the protection methods often incorporate opponent models to improve the performance when facing humans. Previous results show that modeling human behavior can significantly improve the performance of the algorithms. However, modeling humans correctly is a complex problem, and the models are often simplified and assume humans make mistakes according to some distribution or train parameters for the whole population from which they sample. In this work, we use data gathered by psychologists who identified personality types that increase the likelihood of performing malicious acts. However, in the previous work, the tests on a handmade game could not show strategic differences between the models. We created a novel model that links its parameters to psychological traits. We optimized over parametrized games and created games in which the differences are profound. Our work can help with automatic game generation when we need a game in which some models will behave differently and to identify situations in which the models do not align.

Many real-world situations are dynamic, with long-term interactions between multiple agents with uncertainty and limited observations. The agents must reason about which actions to take while also predicting and learning about what actions the other agents will take and how their choices will interact. In the most general setting, there is no limitation on the length of the sequence of actions the agent can perform — that is, there is no fixed horizon that can be used as an endpoint for analysis. These settings can be modeled as partially observable stochastic games (POSGs). Many adversarial domains (e.g., security settings) can be modeled as strictly competitive (or zero-sum) variants of these games. While these models are capable of modeling a wide variety of realistic problems, solving general POSGs is computationally intractable, so we focus on a broad subclass of POSGs called one-sided POSGs. In these games, only one agent has imperfect information while their opponent has full knowledge of the current situation. We provide a complete approach for solving zero-sum, one-sided POSGs: we (1) give a theoretical analysis of one-sided POSGs and their value functions, (2) show that a variant of a value-iteration algorithm converges in this setting, (3) adapt the heuristic search value-iteration algorithm for solving one-sided POSGs, (4) describe how to use approximate value functions to derive strategies in the game, and (5) experimentally demonstrate that our algorithm can solve one-sided POSGs of non-trivial sizes and analyze the scalability of our algorithm in three different domains: pursuit-evasion, patrolling, and search games.

Adversaries are often able to penetrate networks and compromise systems by exploiting vulnerabilities in people and systems. The key to the success of these attacks is information that adversaries collect throughout the phases of the cyber kill chain. We summarize and analyze the methods, tactics, and tools that adversaries use to conduct reconnaissance activities throughout the attack process. First, we discuss what types of information adversaries seek and how and when they can obtain this information. Then, we provide a taxonomy and detailed overview of adversarial reconnaissance techniques. The taxonomy introduces a categorization of reconnaissance techniques based on the source as third-party and human-, and system-based information gathering. This article provides a comprehensive view of adversarial reconnaissance that can help in understanding and modeling this complex but vital aspect of cyber attacks as well as insights that can improve defensive strategies, such as cyber deception.

Reconnaissance activities precedent other attack steps in the cyber kill chain. Zero-day attacks exploit unknown vulnerabilities and give attackers the upper hand against conventional defenses. Honeypots have been used to deceive attackers by misrepresenting the true state of the network. Existing work on cyber deception does not model zero-day attacks. In this paper, we address the question of "How to allocate honeypots over the network?" to protect its most valuable assets. To this end, we develop a two-player zero-sum game theoretic approach to study the potential reconnaissance tracks and attack paths that attackers may use. However, zero-day attacks allow attackers to avoid placed honeypots by creating new attack paths. Therefore, we introduce a sensitivity analysis to investigate the impact of different zero-day vulnerabilities on the performance of the proposed deception technique. Next, we propose several mitigating strategies to defend the network against zero-day attacks based on this analysis. Finally, our numerical results validate our findings and illustrate the effectiveness of the proposed defense approach.

The Power Trading Agent Competition (Power TAC) simulates numerous economic multi-agent trading scenarios where brokers trade in multiple power markets. A successful broker makes sound predictions about future energy supplies, demands, and costs in these markets and bids intelligently to gain profit. The broker who maintains the most profitable portfolio wins the competition. In this chapter, we detail some novel wholesale trading strategies of the broker SPOT used in competitions from 2015 to 2020. We present the higher-level ideas of these strategies with experimental results and discuss the experiences gained in the tournaments while playing some of these strategies. SPOT’s MCTS wholesale bidding strategy can procure energy at the lowest cost, which is presently the state-of-the-art strategy in the PDA markets. We also propose a state-of-the-art methodology to explore better strategies (applicable to the Power TAC domain) against a specific set of opponents, minimizing the agent’s regret.