arXiv subjects

Quanyan Zhu

Publications and source records attributed to Quanyan Zhu.

At least 55 records · Page 3Linked to original sources

arXiv2025

Agentic AI for Cyber Resilience: A New Security Paradigm and Its System-Theoretic Foundations

Cybersecurity is being fundamentally reshaped by foundation-model-based artificial intelligence. Large language models now enable autonomous planning, tool orchestration, and strategic adaptation at scale, challenging security architectures built on static rules, perimeter defenses, and human-centered workflows. This chapter argues for a shift from prevention-centric security toward agentic cyber resilience. Rather than seeking perfect protection, resilient systems must anticipate disruption, maintain critical functions under attack, recover efficiently, and learn continuously. We situate this shift within the historical evolution of cybersecurity paradigms, culminating in an AI-augmented paradigm where autonomous agents participate directly in sensing, reasoning, action, and adaptation across cyber and cyber-physical systems. We then develop a system-level framework for designing agentic AI workflows. A general agentic architecture is introduced, and attacker and defender workflows are analyzed as coupled adaptive processes, and game-theoretic formulations are shown to provide a unifying design language for autonomy allocation, information flow, and temporal composition. Case studies in automated penetration testing, remediation, and cyber deception illustrate how equilibrium-based design enables system-level resiliency design.

cs.CR↗

arXiv2025

Cyber Resilience in Next-Generation Networks: Threat Landscape, Theoretical Foundations, and Design Paradigms

The evolution of networked systems, driven by innovations in software-defined networking (SDN), network function virtualization (NFV), open radio access networks (O-RAN), and cloud-native architectures, is redefining both the operational landscape and the threat surface of critical infrastructures. This book offers an in-depth, interdisciplinary examination of how resilience must be re-conceptualized and re-engineered to address the multifaceted challenges posed by these transformations. Structured across six chapters, this book begins by surveying the contemporary risk landscape, identifying emerging cyber, physical, and AI-driven threats, and analyzing their implications for scalable, heterogeneous network environments. It then establishes rigorous definitions and evaluation frameworks for resilience, going beyond robustness and fault-tolerance to address adaptive, anticipatory, and retrospective mechanisms across diverse application domains. The core of the book delves into advanced paradigms and practical strategies for resilience, including zero trust architectures, game-theoretic threat modeling, and self-healing design principles. A significant portion is devoted to the role of artificial intelligence, especially reinforcement learning and large language models (LLMs), in enabling dynamic threat response, autonomous network control, and multi-agent coordination under uncertainty.

cs.NI↗

arXiv2025

Timing-Aware Two-Player Stochastic Games with Self-Triggered Control

We study self-triggered two-player stochastic games on Piecewise Deterministic Markov Processes (PDMPs) where each agent decides when to observe and which open-loop action to hold. Augmenting the state with clocks and committed controls yields flow regions (both hold) and trigger surfaces (at least one updates). The framework covers both blind simultaneous (Nash) timing and observable sequential (Stackelberg) commitments; the former leads to coupled, intractable QVIs, while the latter admits a nested Hamilton-Jacobi-Bellman quasi-variational inequality and a tractable dynamic-programming decomposition. We outline a computational scheme based on implicit differentiation of the follower's fixed point. A pursuit-evasion example illustrates the strategic timing interaction.

eess.SY↗

arXiv2025

Bayesian Holonic Systems: Equilibrium, Uniqueness, and Computation

This paper addresses the challenge of modeling and control in hierarchical, multi-agent systems, known as holonic systems, where local agent decisions are coupled with global systemic outcomes. We introduce the Bayesian Holonic Equilibrium (BHE), a concept that ensures consistency between agent-level rationality and system-wide emergent behavior. We establish the theoretical soundness of the BHE by showing its existence and, under stronger regularity conditions, its uniqueness. We propose a two-time scale learning algorithm to compute such an equilibrium. This algorithm mirrors the system's structure, with a fast timescale for intra-holon strategy coordination and a slow timescale for inter-holon belief adaptation about external risks. The convergence of the algorithm to the theoretical equilibrium is validated through a numerical experiment on a continuous public good game. This work provides a complete theoretical and algorithmic framework for the principled design and analysis of strategic risk in complex, coupled control systems.

eess.SY↗

arXiv2025

A First Order Meta Stackelberg Method for Robust Federated Learning (Technical Report)

Recent research efforts indicate that federated learning (FL) systems are vulnerable to a variety of security breaches. While numerous defense strategies have been suggested, they are mainly designed to counter specific attack patterns and lack adaptability, rendering them less effective when facing uncertain or adaptive threats. This work models adversarial FL as a Bayesian Stackelberg Markov game (BSMG) between the defender and the attacker to address the lack of adaptability to uncertain adaptive attacks. We further devise an effective meta-learning technique to solve for the Stackelberg equilibrium, leading to a resilient and adaptable defense. The experiment results suggest that our meta-Stackelberg learning approach excels in combating intense model poisoning and backdoor attacks of indeterminate types.

cs.CR↗

arXiv2025

Guarding Against Malicious Biased Threats (GAMBiT): Experimental Design of Cognitive Sensors and Triggers with Behavioral Impact Analysis

This paper introduces GAMBiT (Guarding Against Malicious Biased Threats), a cognitive-informed cyber defense framework that leverages deviations from human rationality as a new defensive surface. Conventional cyber defenses assume rational, utility-maximizing attackers, yet real-world adversaries exhibit cognitive constraints and biases that shape their interactions with complex digital systems. GAMBiT embeds insights from cognitive science into cyber environments through cognitive triggers, which activate biases such as loss aversion, base-rate neglect, and sunk-cost fallacy, and through newly developed cognitive sensors that infer attackers' cognitive states from behavioral and network data. Three rounds of human-subject experiments (total n=61) in a simulated small business network demonstrate that these manipulations significantly disrupt attacker performance, reducing mission progress, diverting actions off the true attack path, and increasing detectability. These results demonstrate that cognitive biases can be systematically triggered to degrade the attacker's efficiency and enhance the defender's advantage. GAMBiT establishes a new paradigm in which the attacker's mind becomes part of the battlefield and cognitive manipulation becomes a proactive vector for cyber defense.

cs.CR↗

arXiv2025

Attributing Responsibility in AI-Induced Incidents: A Computational Reflective Equilibrium Framework for Accountability

The pervasive integration of Artificial Intelligence (AI) has introduced complex challenges in the responsibility and accountability in the event of incidents involving AI-enabled systems. The interconnectivity of these systems, ethical concerns of AI-induced incidents, coupled with uncertainties in AI technology and the absence of corresponding regulations, have made traditional responsibility attribution challenging. To this end, this work proposes a Computational Reflective Equilibrium (CRE) approach to establish a coherent and ethically acceptable responsibility attribution framework for all stakeholders. The computational approach provides a structured analysis that overcomes the limitations of conceptual approaches in dealing with dynamic and multifaceted scenarios, showcasing the framework's traceability, coherence, and adaptivity properties in the responsibility attribution process. We examine the pivotal role of the initial activation level associated with claims in equilibrium computation. Using an AI-assisted medical decision-support system as a case study, we illustrate how different initializations lead to diverse responsibility distributions. The framework offers valuable insights into accountability in AI-induced incidents, facilitating the development of a sustainable and resilient system through continuous monitoring, revision, and reflection.

cs.AI↗

arXiv2025

Bi-Level Game-Theoretic Planning of Cyber Deception for Cognitive Arbitrage

Cognitive vulnerabilities shape human decision-making and arise primarily from two sources: (1) cognitive capabilities, which include disparities in knowledge, education, expertise, or access to information, and (2) cognitive biases, such as rational inattention, confirmation bias, and base rate neglect, which influence how individuals perceive and process information. Exploiting these vulnerabilities allows an entity with superior cognitive awareness to gain a strategic advantage, a concept referred to as cognitive arbitrage. This paper investigates how to exploit the cognitive vulnerabilities of Advanced Persistent Threat (APT) attackers and proposes cognition-aware defenses that leverage windows of superiority to counteract attacks. Specifically, the proposed bi-level cyber warfare game focuses on "strategic-level" design for defensive deception mechanisms, which then facilitates "operational-level" actions and tactical-level execution of Tactics, Techniques, and Procedures (TTPs). Game-theoretic reasoning and analysis play a significant role in the cross-echelon quantitative modeling and design of cognitive arbitrage strategies. Our numerical results demonstrate that although the defender's initial advantage diminishes over time, strategically timed and deployed deception techniques can turn a negative value for the attacker into a positive one during the planning phase, and achieve at least a 40% improvement in total rewards during execution. This demonstrates that the defender can amplify even small initial advantages, sustain a strategic edge over the attacker, and secure long-term objectives, such as protecting critical assets throughout the attacker's lifecycle.

cs.GT↗

arXiv2025

Guarding Against Malicious Biased Threats (GAMBiT) Experiments: Revealing Cognitive Bias in Human-Subjects Red-Team Cyber Range Operations

We present three large-scale human-subjects red-team cyber range datasets from the Guarding Against Malicious Biased Threats (GAMBiT) project. Across Experiments 1-3 (July 2024-March 2025), 19-20 skilled attackers per experiment conducted two 8-hour days of self-paced operations in a simulated enterprise network (SimSpace Cyber Force Platform) while we captured multi-modal data: self-reports (background, demographics, psychometrics), operational notes, terminal histories, keylogs, network packet captures (PCAP), and NIDS alerts (Suricata). Each participant began from a standardized Kali Linux VM and pursued realistic objectives (e.g., target discovery and data exfiltration) under controlled constraints. Derivative curated logs and labels are included. The combined release supports research on attacker behavior modeling, bias-aware analytics, and method benchmarking. Data are available via IEEE Dataport entries for Experiments 1-3.

cs.CR↗

arXiv2025

Dynamic Information Manipulation Game

We propose a dynamic information manipulation game (DIMG) to investigate the incentives of an information manipulator (IM) to influence the transition rules of a partially observable Markov decision process (POMDP). DIMG is a hierarchical game where the upper-level IM stealthily designs the POMDP's joint state distributions to influence the lower-level controller's actions. DIMG's fundamental feature is characterized by a stagewise constraint that ensures the consistency between the unobservable marginals of the manipulated and the original kernels. In an equilibrium of information distortion, the IM minimizes cumulative cost that depends on the controller's informationally manipulated actions generated by the optimal policy to the POMDP. We discuss ex ante and interim manipulation schemes and show their connections. The effect of manipulation on the performance of control policies is analyzed through its influence on belief distortion.

math.OC↗

arXiv2025

Game Theory Meets LLM and Agentic AI: Reimagining Cybersecurity for the Age of Intelligent Threats

Protecting cyberspace requires not only advanced tools but also a shift in how we reason about threats, trust, and autonomy. Traditional cybersecurity methods rely on manual responses and brittle heuristics. To build proactive and intelligent defense systems, we need integrated theoretical frameworks and software tools. Game theory provides a rigorous foundation for modeling adversarial behavior, designing strategic defenses, and enabling trust in autonomous systems. Meanwhile, software tools process cyber data, visualize attack surfaces, verify compliance, and suggest mitigations. Yet a disconnect remains between theory and practical implementation. The rise of Large Language Models (LLMs) and agentic AI offers a new path to bridge this gap. LLM-powered agents can operationalize abstract strategies into real-world decisions. Conversely, game theory can inform the reasoning and coordination of these agents across complex workflows. LLMs also challenge classical game-theoretic assumptions, such as perfect rationality or static payoffs, prompting new models aligned with cognitive and computational realities. This co-evolution promises richer theoretical foundations and novel solution concepts. Agentic AI also reshapes software design: systems must now be modular, adaptive, and trust-aware from the outset. This chapter explores the intersection of game theory, agentic AI, and cybersecurity. We review key game-theoretic frameworks (e.g., static, dynamic, Bayesian, and signaling games) and solution concepts. We then examine how LLM agents can enhance cyber defense and introduce LLM-driven games that embed reasoning into AI agents. Finally, we explore multi-agent workflows and coordination games, outlining how this convergence fosters secure, intelligent, and adaptive cyber systems.

cs.CR↗

arXiv2025

LLM-Stackelberg Games: Conjectural Reasoning Equilibria and Their Applications to Spearphishing

We introduce the framework of LLM-Stackelberg games, a class of sequential decision-making models that integrate large language models (LLMs) into strategic interactions between a leader and a follower. Departing from classical Stackelberg assumptions of complete information and rational agents, our formulation allows each agent to reason through structured prompts, generate probabilistic behaviors via LLMs, and adapt their strategies through internal cognition and belief updates. We define two equilibrium concepts: reasoning and behavioral equilibrium, which aligns an agent's internal prompt-based reasoning with observable behavior, and conjectural reasoning equilibrium, which accounts for epistemic uncertainty through parameterized models over an opponent's response. These layered constructs capture bounded rationality, asymmetric information, and meta-cognitive adaptation. We illustrate the framework through a spearphishing case study, where a sender and a recipient engage in a deception game using structured reasoning prompts. This example highlights the cognitive richness and adversarial potential of LLM-mediated interactions. Our results show that LLM-Stackelberg games provide a powerful paradigm for modeling decision-making in domains such as cybersecurity, misinformation, and recommendation systems.

cs.AI↗

arXiv2025

Reasoning and Behavioral Equilibria in LLM-Nash Games: From Mindsets to Actions

We introduce the LLM-Nash framework, a game-theoretic model where agents select reasoning prompts to guide decision-making via Large Language Models (LLMs). Unlike classical games that assume utility-maximizing agents with full rationality, this framework captures bounded rationality by modeling the reasoning process explicitly. Equilibrium is defined over the prompt space, with actions emerging as the behavioral output of LLM inference. This approach enables the study of cognitive constraints, mindset expressiveness, and epistemic learning. Through illustrative examples, we show how reasoning equilibria can diverge from classical Nash outcomes, offering a new foundation for strategic interaction in LLM-enabled systems.

cs.AI↗

arXiv2025

A Multi-Resolution Dynamic Game Framework for Cross-Echelon Decision-Making in Cyber Warfare

Cyber warfare has become a critical dimension of modern conflict, driven by society's increasing dependence on interconnected digital and physical infrastructure. Effective cyber defense often requires decision-making at different echelons, where the tactical layer focuses on detailed actions such as techniques, tactics, and procedures, while the strategic layer addresses long-term objectives and coordinated planning. Modeling these interactions at different echelons remains challenging due to the dynamic, large-scale, and interdependent nature of cyber environments. To address this, we propose a multi-resolution dynamic game framework in which the tactical layer captures fine-grained interactions using high-resolution extensive-form game trees, while the strategic layer is modeled as a Markov game defined over lower-resolution states abstracted from those game trees. This framework supports scalable reasoning and planning across different levels of abstraction through zoom-in and zoom-out operations that adjust the granularity of the modeling based on operational needs. A case study demonstrates how the framework works and its effectiveness in improving the defender's strategic advantage.

cs.CR↗

arXiv2025

When to Deceive: A Cross-Layer Stackelberg Game Framework for Strategic Timing of Cyber Deception

Cyber deception is an emerging proactive defense strategy to counter increasingly sophisticated attacks such as Advanced Persistent Threats (APTs) by misleading and distracting attackers from critical assets. However, since deception techniques incur costs and may lose effectiveness over time, defenders must strategically time and select them to adapt to the dynamic system and the attacker's responses. In this study, we propose a Stackelberg game-based framework to design strategic timing for cyber deception: the lower tactical layer (follower) captures the evolving attacker-defender dynamics under a given deception through a one-sided information Markov game, while the upper strategic layer (leader) employs a stopping-time decision process to optimize the timing and selection of deception techniques. We also introduce a computational algorithm that integrates dynamic programming and belief-state updates to account for the attacker's adaptive behavior and limited deception resources. Numerical experiments validate the framework, showing that strategically timed deceptions can enhance the defender's expected utility and reduce the risk of asset compromise compared to baseline strategies.

cs.GT↗

arXiv2025

PACT: A Contract-Theoretic Framework for Pricing Agentic AI Services Powered by Large Language Models

Agentic AI, often powered by large language models (LLMs), is becoming increasingly popular and adopted to support autonomous reasoning, decision-making, and task execution across various domains. While agentic AI holds great promise, its deployment as services for easy access raises critical challenges in pricing, due to high infrastructure and computation costs, multi-dimensional and task-dependent Quality of Service (QoS), and growing concerns around liability in high-stakes applications. In this work, we propose PACT, a Pricing framework for cloud-based Agentic AI services through a Contract-Theoretic approach, which models QoS along both objective (e.g., response time) and subjective (e.g., user satisfaction) dimensions. PACT accounts for computational, infrastructure, and potential liability costs for the service provider, while ensuring incentive compatibility and individual rationality for the user under information asymmetry. Through contract-based selection, users receive tailored service offerings aligned with their needs. Numerical evaluations demonstrate that PACT improves QoS alignment between users and providers and offers a scalable, liable approach to pricing agentic AI services in the future.

cs.GT↗

arXiv2025

Improving Network Threat Detection by Knowledge Graph, Large Language Model, and Imbalanced Learning

Network threat detection has been challenging due to the complexities of attack activities and the limitation of historical threat data to learn from. To help enhance the existing practices of using analytics, machine learning, and artificial intelligence methods to detect the network threats, we propose an integrated modelling framework, where Knowledge Graph is used to analyze the users' activity patterns, Imbalanced Learning techniques are used to prune and weigh Knowledge Graph, and LLM is used to retrieve and interpret the users' activities from Knowledge Graph. The proposed framework is applied to Agile Threat Detection through Online Sequential Learning. The preliminary results show the improved threat capture rate by 3%-4% and the increased interpretabilities of risk predictions based on the users' activities.

cs.LG↗

arXiv2025

Bilateral Cognitive Security Games in Networked Control Systems under Stealthy Injection Attacks

This paper studies a strategic security problem in networked control systems under stealthy false data injection attacks. The security problem is modeled as a bilateral cognitive security game between a defender and an adversary, each possessing cognitive reasoning abilities. The adversary with an adversarial cognitive ability strategically attacks some interconnections of the system with the aim of disrupting the network performance while remaining stealthy to the defender. Meanwhile, the defender with a defense cognitive ability strategically monitors some nodes to impose the stealthiness constraint with the purpose of minimizing the worst-case disruption caused by the adversary. Within the proposed bilateral cognitive security framework, the preferred cognitive levels of the two strategic agents are formulated in terms of two newly proposed concepts, cognitive mismatch and cognitive resonance. Moreover, we propose a method to compute the policies for the defender and the adversary with arbitrary cognitive abilities. A sufficient condition is established under which an increase in cognitive levels does not alter the policies for the defender and the adversary, ensuring convergence. The obtained results are validated through numerical simulations.

eess.SY↗