The convergence of post-quantum cryptography and AI-driven threat hunting: a review of quantum-resilient defense frameworks

1 Introduction

The security of modern digital infrastructures relies heavily on cryptographic mechanisms and monitoring capabilities that were designed under classical computational assumptions (Olaluwe et al., 2025). Public-key cryptography, secure communication protocols, and digital trust models form the backbone of confidentiality, integrity, and authentication across cyberspace. At the same time, security operations increasingly rely on data-driven and federated detection models that must remain effective even as adversarial capabilities expand into the quantum domain (Namakshenas et al., 2024). These assumptions are increasingly challenged by two parallel developments: the emergence of quantum computing and the growing use of artificial intelligence by adversaries. Quantum computing introduces a new computational paradigm that fundamentally alters the feasibility of solving certain mathematical problems underlying widely deployed cryptographic schemes (Sharma and Rani, 2025). Cryptosystems that derive their security from problems such as integer factorization or discrete logarithms are no longer guaranteed to provide long-term protection once sufficiently capable quantum adversaries are considered. This shift does not merely represent a gradual degradation of security margins but a structural break in cryptographic trust assumptions. As a result, post-quantum cryptography has emerged as a critical research direction, aiming to design cryptographic primitives that remain secure against both classical and quantum-enabled attackers (Saeed, 2025). In parallel, the cyber threat landscape has evolved toward highly adaptive, stealthy, and persistent attack strategies. Advanced adversaries increasingly leverage automation, learning-based techniques, and dynamic behaviors to evade static defenses. Quantum-resilient anomaly detection frameworks that combine multi-layer learning with post-quantum cryptographic protections have therefore gained prominence as a proactive security paradigm, enabling continuous behavioral analysis and contextual threat inference across complex distributed environments (Şahin, 2026). Despite significant advances in both post-quantum cryptography and AI-based security analytics, these domains are often treated in isolation, with cryptographic resilience and intelligent detection remaining inadequately co-designed. Cryptographic research primarily focuses on mathematical hardness assumptions, protocol design, and formal security proofs, while AI-driven threat hunting emphasizes data-driven inference, pattern recognition, and adaptive decision-making (Alyas et al., 2025). This separation overlooks the reality that future cyber defense systems must operate in environments where cryptographic resilience and intelligent threat detection are tightly interdependent. Cryptographic strength alone cannot prevent misuse, misconfiguration, or insider-driven exploitation, while intelligent monitoring without quantum-resilient cryptographic foundations remains vulnerable to future cryptanalytic breakthroughs. This paper argues that the convergence of post-quantum cryptography and AI-driven threat hunting is a necessary step toward building quantum-resilient defense frameworks (Alluhaibi, 2024). By integrating cryptographic assurance with learning-based threat inference, security architectures can move beyond static protection models toward adaptive, context-aware, and future-ready defenses. The contribution of this work lies in providing a structured and analytically grounded review of this convergence, synthesizing existing research across cryptography, artificial intelligence, and cybersecurity into a coherent conceptual framework. Figure 1 presents a high-level conceptual view of the transition from classical security models to quantum-resilient defense architectures. It highlights how traditional cryptographic mechanisms and reactive security monitoring, designed under classical assumptions, are increasingly challenged by quantum computing capabilities and AI-enabled adversarial behaviors. The figure further illustrates the limitations of treating post-quantum cryptography and AI-driven threat hunting as isolated solutions, and emphasizes the need for their convergence into integrated, adaptive, and context-aware defense frameworks.

Specifically, the paper makes three primary contributions. First, it systematically reviews the foundational principles of post-quantum cryptography and AI-driven threat hunting, highlighting their respective roles and limitations in isolation. Second, it analyzes emerging architectural and trust models that support quantum-resilient and intelligence-driven security operations. Third, it identifies open challenges and research gaps that must be addressed to enable the effective co-design of cryptographic and AI-based defense mechanisms. The remainder of the paper is organized to progressively build this narrative, moving from technical foundations to integrated frameworks and future research. The proposed AI–PQC defense perspective is particularly relevant to smart grid communication infrastructures, where advanced metering systems, distributed control channels, and real-time grid monitoring rely on secure, low-latency, and resilient communication mechanisms that must remain protected against both quantum-enabled cryptanalysis and intelligent adaptive cyber threats.

Paper Organization: The remainder of this paper is structured as follows. Section 2 establishes the technical foundations, covering quantum computing impacts on cryptography, post-quantum cryptographic families, and AI-driven threat hunting paradigms. Section 3 motivates the convergence of PQC and AI-driven threat hunting. Section 4 presents the review methodology including databases, search strategy, and inclusion/exclusion criteria. Section 5 examines AI techniques in quantum-resilient threat hunting. Section 6 analyzes architectural models, formal security properties, and trust modeling. Section 7 presents the integrated AI-PQC defense framework. Section 8 identifies open challenges and research gaps. Section 9 outlines future research directions. Section 10 concludes the paper.

2 Background and technical foundations

This section establishes the technical foundations necessary to understand the convergence of post-quantum cryptography and AI-driven threat hunting (Rahgozar et al., 2025). It reviews the key principles of quantum computing that directly impact cryptographic security, introduces post-quantum cryptography as a response to emerging quantum threats, and examines the evolution of threat hunting toward intelligence-driven and AI-enabled security practices (Darzi and Yavuz, 2024a). The purpose of this section is to provide conceptual clarity and a common technical baseline without relying on unverifiable quantitative claims.

2.1 Quantum computing and cryptographic disruption

Quantum computing introduces a computational model that departs fundamentally from classical deterministic and probabilistic computation (Nguyen et al., 2025). By exploiting quantum mechanical properties such as superposition and entanglement, quantum systems enable new computational approaches to complex classification and optimization problems, including those relevant to stealth attack detection in distributed security environments (Ogiesoba-Eguakun and Rath, 2025). From a cryptographic perspective, the critical implication of this paradigm shift lies not in general computational acceleration, but in its impact on specific mathematical problem classes that form the security foundations of modern cryptography (Mehmood et al., 2024). Classical public-key cryptographic schemes rely on the assumed intractability of problems such as integer factorization and discrete logarithms under classical computational models. These assumptions implicitly define the adversarial capabilities against which security guarantees are established (Hussain et al., 2025). When adversaries are modeled as quantum-capable, the underlying hardness assumptions must be revisited, as the feasibility of solving these problems can change qualitatively rather than incrementally (Alharbi and Almazmomi, 2025). This constitutes a structural disruption of cryptographic trust, as the validity of long-standing security guarantees becomes contingent on new and evolving computational models. Symmetric cryptographic primitives are affected in a more nuanced manner (Shamoo, 2025). While their design does not rely on the same algebraic hardness assumptions as public-key systems, their effective security level must be evaluated in the context of quantum-capable adversarial models, which motivates quantum-safe approaches to the full security stack including federated learning frameworks (Kannan et al., 2024). Consequently, long-term security planning must account for adversarial models that differ fundamentally from classical assumptions. These considerations motivate the transition toward cryptographic designs whose security arguments explicitly account for quantum-enabled adversaries (Nayak et al., 2025).

2.2 Post-quantum cryptography

Post-quantum cryptography encompasses cryptographic constructions designed to maintain security in the presence of both classical and quantum adversaries. Unlike quantum cryptographic approaches that depend on specialized physical infrastructure, post-quantum cryptographic schemes are intended for deployment on conventional computing platforms while relying on mathematical problems believed to remain hard under known quantum attack models (Mujlid and Alshahrani, 2025). Post-quantum cryptographic research has produced several families of schemes, each grounded in distinct mathematical foundations and security assumptions. These families differ not only in their underlying problem structures, but also in their implications for system integration, key management, and protocol design. Importantly, the diversity of post-quantum constructions reflects the absence of a single dominant hardness assumption in the post-quantum setting, reinforcing the need for careful cryptographic agility and system-level planning. Table 1 provides a conceptual taxonomy of major post-quantum cryptographic families, emphasizing their mathematical foundations and deployment considerations rather than quantitative performance metrics (Darzi and Yavuz, 2024b).

TABLE 1. Conceptual taxonomy of major post-quantum cryptographic families.

NIST Post-Quantum Cryptography Standardization: A critical milestone in the operationalization of post-quantum cryptography was achieved in 2024 when NIST finalized three primary PQC standards. ML-KEM (Module Lattice-based Key Encapsulation Mechanism, formerly CRYSTALS-Kyber) was standardized as the primary key encapsulation mechanism. ML-DSA (Module Lattice-based Digital Signature Algorithm, formerly CRYSTALS-Dilithium) was standardized as the primary digital signature scheme. SLH-DSA (Stateless Hash-based Digital Signature Algorithm, formerly SPHINCS+) was standardized as a hash-based signature alternative offering conservative security assumptions. Additionally, FN-DSA (formerly FALCON) was selected as a compact lattice-based signature scheme for bandwidth-constrained environments. These standards provide the concrete cryptographic primitives that quantum-resilient defense frameworks must integrate. Their selection reflects deliberate diversification across lattice-based and hash-based hardness assumptions, acknowledging that no single post-quantum family yet commands universal confidence. For practitioners designing integrated AI-PQC systems, trade-offs in key size, signature size, and computational overhead all interact with the performance requirements of AI-based threat monitoring pipelines.

From a defensive standpoint, post-quantum cryptography addresses the cryptanalytic dimension of quantum threats by preserving confidentiality, integrity, and authentication guarantees under revised adversarial assumptions. However, these guarantees operate at the cryptographic layer and do not directly address adaptive attack strategies, operational misuse, or behavioral anomalies (Ahmed et al., 2025). As a result, cryptographic resilience must be complemented by higher-layer security mechanisms capable of reasoning about system behavior and threat evolution.

2.3 AI-driven threat hunting

Threat hunting has evolved from rule-based and signature-driven detection toward intelligence-driven and learning-based security analysis (Bera et al., 2025a). Traditional monitoring systems assume that malicious activity can be reliably identified through predefined indicators derived from historical observations. While effective against known threats, this assumption becomes increasingly fragile in the presence of adaptive adversaries that intentionally evade static detection mechanisms (Baazeem, 2026). AI-driven threat hunting reframes security monitoring as a continuous inference process over system behavior (Al Khaldy, 2024). Instead of relying on fixed rules, learning-based models construct representations of normal operational patterns and identify deviations that may indicate malicious intent. This approach enables proactive discovery of previously unseen threats and supports adaptive response strategies as adversarial behaviors evolve (Shahzad and Unalp, 2025). At an abstract level, AI-driven threat hunting can be formalized as a decision process that maps observed system states to threat hypotheses. Let x ∈ X denote an observed behavioral state and h ∈ H denote a threat hypothesis. Threat inference can be expressed as a decision function:

h * = arg ⁡ max ⁡ h ∈ H P h ∣ x

This formulation does not prescribe a specific learning algorithm, but captures the core objective of AI-driven threat hunting: inferring the most plausible threat explanation given observed behavior. Such abstractions are particularly valuable in complex environments where explicit rule construction is infeasible (Lamichhane and Rawat, 2025). In practice, this probabilistic inference framework can be instantiated through several well-established AI paradigms. Bayesian networks offer principled probabilistic reasoning over observed behavioral states, enabling transparent uncertainty quantification aligned with the threat hypothesis formulation above. LSTM (Long Short-Term Memory) networks and Transformer-based models capture temporal dependencies in sequential security event logs, making them well-suited for modeling the evolving behavior of advanced persistent threats. Isolation Forest and One-Class SVM provide effective unsupervised anomaly detection where labeled threat data is scarce. Reinforcement learning offers an adaptive response selection framework, mapping inferred threat hypotheses to defensive actions under uncertainty. In post-quantum environments, the choice among these paradigms must account for the cryptographic overhead of PQC-protected channels, as high-latency key operations may affect the timeliness of threat inference in real-time security contexts. Despite its strengths, AI-driven threat hunting depends critically on the trustworthiness of the underlying data and security mechanisms it observes. If cryptographic protections are compromised or rendered obsolete by future computational advances, the integrity of behavioral observations and threat inference processes may be undermined (Divyajyothi et al., 2025). This dependency highlights the necessity of integrating AI-driven threat hunting with cryptographic mechanisms that remain robust under evolving adversarial models.

Summary of Research Gaps: The review of quantum computing impacts, PQC families, and AI-driven threat hunting in this section reveals three interconnected gaps motivating this study. First, PQC and AI detection methods have been studied extensively in isolation, but their joint behavior under unified adversarial models is largely unexplored. Second, existing cryptographic frameworks do not adequately address the integrity of AI monitoring data pipelines, creating an attack surface PQC alone cannot close. Third, the transition to quantum-resilient architectures lacks formal guidance on safely co-deploying AI-based monitoring during migration periods. These gaps define three research objectives: (i) to characterize interdependencies between PQC and AI-driven threat hunting; (ii) to analyze emerging architectural and trust models supporting their convergence; and (iii) to identify concrete research directions that close the identified gaps.

3 Motivation for convergence: post-quantum cryptography and AI

The accelerating evolution of both computational capabilities and adversarial strategies exposes fundamental limitations in security approaches that treat cryptographic resilience and threat detection as independent concerns (Sai et al., 2025). While post-quantum cryptography addresses the cryptanalytic risks introduced by quantum-capable adversaries, it operates primarily at the level of mathematical assurance and protocol correctness. Conversely, AI-driven threat hunting focuses on behavioral analysis, adaptive inference, and operational awareness, yet often assumes the continued reliability of underlying cryptographic protections (Ali et al., 2025). When considered in isolation, each approach leaves critical gaps that can be exploited by advanced and hybrid threat actors.

Cryptographic mechanisms, regardless of their mathematical strength, do not provide visibility into how systems are used, misused, or deliberately manipulated (Waheed et al., 2025). Even cryptographic primitives that remain secure under quantum adversarial models cannot prevent threats arising from misconfiguration, insider activity, or protocol abuse (Dhivyasree and Prabu, 2025). In such scenarios, the absence of intelligent monitoring limits the defender’s ability to detect early indicators of compromise or to reason about evolving attack intent. As systems grow in scale and complexity, these blind spots become increasingly difficult to manage through static security controls alone (Fathalla and Azab, 2024). At the same time, AI-driven threat hunting cannot operate effectively without trustworthy security foundations. Learning-based detection and inference rely on the integrity, authenticity, and confidentiality of the data they analyze (Shakil et al., 2024). If cryptographic protections are weakened or rendered obsolete under future computational models, adversaries may manipulate observation channels, poison training data, or obscure malicious activity in ways that undermine the reliability of AI-based security analytics (Micheal, 2025). This dependency reveals a structural coupling between cryptographic assurance and intelligent threat detection that is often underappreciated in current security architectures. Emerging threat scenarios further reinforce the need for convergence. Adversaries may combine quantum-enabled cryptanalytic capabilities with AI-assisted automation to accelerate reconnaissance, adapt attack strategies in real time, and exploit weaknesses across multiple layers of the security stack (Abbas et al., 2025). Such hybrid threat models challenge defenses that are designed around single-dimensional assumptions, whether purely cryptographic or purely behavioral. Addressing these threats requires defense strategies that integrate long-term cryptographic resilience with continuous, context-aware threat inference (Khan and Khan, 2025). The convergence of post-quantum cryptography and AI-driven threat hunting enables a shift toward holistic defense frameworks that are both structurally robust and operationally adaptive (Hanafi et al., 2024). In such frameworks, cryptographic mechanisms establish a trusted substrate that preserves the integrity of security signals, while AI-driven threat hunting provides the adaptive reasoning necessary to interpret those signals and respond to evolving adversarial behavior (Baseri et al., 2025). This integration supports security architectures capable of anticipating, rather than merely reacting to, future threat dynamics. By motivating the co-design of cryptographic and intelligent defense mechanisms, this section establishes the conceptual foundation for quantum-resilient security frameworks. The convergence of post-quantum cryptography and AI-driven threat hunting is not presented as an incremental enhancement, but as a necessary evolution in cybersecurity design philosophy, aligned with the realities of emerging computational and adversarial paradigms (Choudhury et al., 2025).

4 Review methodology

This review adopts a structured and analytically driven methodology designed to synthesize existing research on post-quantum cryptography and AI-driven threat hunting without relying on empirical performance comparisons or speculative quantitative claims. The methodological focus is placed on conceptual rigor, consistency of security assumptions, and the identification of design patterns and research gaps relevant to quantum-resilient defense frameworks. Figure 2 illustrates the analytical review methodology employed in this study. The process begins with the selection of peer-reviewed literature spanning cryptography, artificial intelligence, and cybersecurity, followed by relevance and assumption screening to ensure conceptual alignment with the study’s scope and threat models. The retained literature is then thematically classified and subjected to conceptual mapping and analysis, enabling the identification of recurring design principles, interactions, and security assumptions. The final analytical synthesis consolidates observed patterns, research gaps, and integration opportunities, forming a structured foundation for the integrated frameworks discussed in subsequent sections.

The literature selection process prioritizes peer-reviewed research published in established scientific venues in the fields of cryptography, artificial intelligence, and cybersecurity. Rather than pursuing exhaustive coverage, the review emphasizes relevance, conceptual contribution, and alignment with the research objectives of this study. Publications were selected based on their ability to inform one or more of the following dimensions: quantum-resistant cryptographic design, AI-based threat inference, security architecture, or trust and risk modeling under advanced adversarial assumptions. Selected studies were systematically categorized according to thematic focus, underlying security assumptions, and methodological orientation. This categorization enabled the identification of recurring patterns, complementary approaches, and points of divergence across the literature. In particular, the review distinguishes between work that addresses cryptographic resilience at the mathematical or protocol level and work that emphasizes intelligence-driven security analytics at the operational or behavioral level. This distinction is essential for analyzing how these research streams interact and where integration remains underexplored. The analytical phase of the review employs conceptual mapping to synthesize findings across domains. Rather than aggregating results or comparing numerical metrics, the methodology focuses on relationships between design principles, threat models, and security objectives. This approach supports the extraction of higher-level insights that transcend individual implementations and remain robust under evolving technological and adversarial conditions. To ensure methodological rigor and transparency, inclusion and exclusion decisions were guided by predefined criteria related to relevance, clarity of assumptions, and contribution to quantum-resilient security understanding. Studies that relied primarily on unverifiable performance claims, incomplete threat models, or narrowly scoped experimental setups were deprioritized in favor of work offering transferable conceptual insights. Through this methodology, the review aims to provide a coherent and defensible synthesis of the current research landscape, forming a reliable foundation for the integrated frameworks and research directions discussed in subsequent sections.

4.1 Search strategy and database selection

To ensure transparency and reproducibility, this review adopted a structured search strategy drawing on four major academic databases: IEEE Xplore, Scopus, Web of Science, and the ACM Digital Library. These databases were selected for comprehensive coverage of peer-reviewed publications in cryptography, cybersecurity, and artificial intelligence. Google Scholar was additionally consulted for conference proceedings and grey literature. The search covered publications from 2019 to February 2026.

The following Boolean queries were applied across all databases:

• Q1: (“post-quantum cryptography” OR “quantum-resistant cryptography” OR “lattice-based cryptography”) AND (“cybersecurity” OR “cyber defense” OR “security framework”).
• Q2: (“AI-driven threat hunting” OR “machine learning” OR “anomaly detection”) AND (“post-quantum” OR “quantum resilient” OR “quantum-safe”).
• Q3: (“ML-KEM” OR “ML-DSA” OR “CRYSTALS-Kyber” OR “CRYSTALS-Dilithium” OR “SPHINCS+”) AND (“security” OR “deployment” OR “performance”).
• Q4: (“quantum-resilient” OR “quantum-aware”) AND (“threat intelligence” OR “security architecture” OR “zero trust” OR “federated learning”).

4.2 Inclusion and exclusion criteria

Studies were assessed against predefined criteria applied independently by both authors, with disagreements resolved through discussion. Inclusion criteria: (i) peer-reviewed publication in IEEE, ACM, Springer, Elsevier, or equivalent Q1/Q2 venues; (ii) published 2019–2026; (iii) addresses PQC, AI-based threat detection, or their integration; (iv) includes explicit adversarial or security assumptions. Exclusion criteria: (i) editorials, opinion pieces, or non-peer-reviewed preprints without a companion publication; (ii) publications before 2019 unless foundational; (iii) publications lacking a stated threat model; (iv) anonymous, self-published, or predatory journal venues.

4.3 Literature selection workflow

The selection process followed a four-stage workflow. Stage 1 (Identification): Boolean queries returned a combined pool of 847 records across all databases after deduplication. Stage 2 (Screening): title and abstract review against inclusion criteria reduced the pool to 312 candidate studies. Stage 3 (Eligibility): full-text review of 312 candidates, excluding studies relying solely on unverifiable benchmarks or narrowly scoped experiments. Stage 4 (Inclusion): 84 studies were retained and thematically classified across five categories: (i) PQC foundations and standardization, (ii) AI-driven threat detection, (iii) integrated PQC-AI architectures, (iv) trust and risk modeling, and (v) deployment challenges and open problems. This structured workflow ensures a balanced, representative, and critically evaluated literature base.

5 AI techniques in quantum-resilient threat hunting

As cybersecurity systems transition toward quantum-resilient cryptographic foundations, threat hunting mechanisms must also evolve to operate effectively under revised cryptographic and adversarial assumptions. AI-driven threat hunting provides the analytical capabilities required to reason about complex, encrypted, and adaptive behaviors in environments protected by post-quantum cryptography (Khan et al., 2024). This section examines how AI techniques can be systematically applied to quantum-resilient settings, focusing on feature representation, learning paradigms, and formal detection models.

5.1 Feature representation in post-quantum environments

Feature representation plays a critical role in enabling AI-based threat hunting, particularly in environments where cryptographic mechanisms limit direct visibility into payload content. In post-quantum environments, this challenge is further compounded by the structural differences introduced by quantum-resistant cryptographic primitives, such as larger key sizes, altered handshake behaviors, and modified protocol flows (Turnip et al., 2025). Rather than relying on raw cryptographic material, effective feature representations in post-quantum settings emphasize behavioral, structural, and temporal attributes of cryptographic events. Examples include key exchange patterns, protocol state transitions, message timing characteristics, and frequency of cryptographic operations (Ishaq and Al-Anesi, 2025). These features abstract cryptographic processes into representations suitable for learning-based analysis while preserving security and privacy guarantees. Formally, let e t denote a cryptographic event observed at time t, and let ϕ · be a feature extraction function that maps observed events into a feature space F:

x t = ϕ e t , x t ∈ F

This abstraction decouples threat analysis from specific cryptographic implementations, allowing AI models to operate across heterogeneous post-quantum schemes. Importantly, feature representations must remain stable under cryptographic transitions to ensure continuity of threat inference as systems migrate toward quantum-resistant protocols.

5.2 Learning paradigms for threat hunting

AI-driven threat hunting leverages multiple learning paradigms to address different threat detection objectives in quantum-resilient systems. Supervised learning approaches are applicable when labeled security data is available, enabling classification of known malicious behaviors based on historical observations (Rakhshanda and Iqra, 2025). However, the scarcity of labeled data for emerging and quantum-era threats limits the effectiveness of purely supervised techniques (Bedi et al., 2025). Unsupervised and semi-supervised learning paradigms play a central role in quantum-resilient threat hunting by modeling normal system behavior and identifying deviations that may indicate malicious activity (Choudhury et al., 2025). These approaches are particularly well-suited for detecting stealthy, previously unseen threats that evade signature-based defenses (Rustamova et al., 2025). By learning latent representations of system behavior, unsupervised models support anomaly detection without relying on predefined attack templates (Rishiwal et al., 2025). Reinforcement learning further extends threat hunting capabilities by enabling adaptive decision-making under uncertainty. In this paradigm, defensive agents learn policies that balance detection sensitivity, response timing, and operational constraints through interaction with the environment. Such adaptability is essential in post-quantum contexts, where both cryptographic mechanisms and adversarial strategies may evolve over time. Collectively, these learning paradigms support a layered threat hunting strategy that combines recognition of known patterns with the discovery of novel behaviors and the continuous refinement of defensive actions (Alzahrani, 2025). Table 2 summarizes the relationship between feature abstractions, learning paradigms, and threat hunting objectives in quantum-resilient environments, providing a conceptual bridge between behavioral representation and formal threat inference.

TABLE 2. Conceptual mapping of AI techniques for threat hunting in post-quantum environments.

5.3 Mathematical formulation of AI-based detection

To provide a formal foundation for AI-driven threat hunting in quantum-resilient environments, threat detection can be modeled as a probabilistic inference and decision optimization problem (Ajax et al., 2025). Let x ∈ F denote an observed feature vector derived from cryptographic and behavioral events, and let h ∈ H represent a threat hypothesis, where H includes both benign and malicious states. The objective of threat inference is to estimate the most plausible hypothesis given observed behavior:

h * = arg ⁡ max ⁡ h ∈ H P h ∣ x

This formulation captures the core reasoning process underlying AI-driven threat hunting without prescribing a specific model architecture. It accommodates probabilistic classifiers, anomaly detection models, and hybrid inference mechanisms.

Defensive decision-making can be further expressed as an optimization problem, where the system selects a response action a ∈ A to minimize expected risk under uncertainty:

a * = arg ⁡ min ⁡ a ∈ A E h ∼ P h ∣ x R a , h

Here, R a , h denotes a risk function capturing the potential impact of action a under hypothesis h. This abstraction supports adaptive response strategies while remaining agnostic to specific cryptographic implementations. Crucially, the integrity of these inference and optimization processes depends on the robustness of the cryptographic foundations that protect observation channels and system state. Post-quantum cryptography ensures that cryptographic protections remain effective under future adversarial models, thereby preserving the reliability of AI-based threat detection and decision-making. This interdependence reinforces the necessity of integrating AI-driven threat hunting with quantum-resilient cryptographic mechanisms.

6 Quantum-resilient defense frameworks

The transition toward quantum-resilient cybersecurity requires defense frameworks that extend beyond cryptographic robustness to encompass architectural adaptability, formal security guarantees, and dynamic trust and risk reasoning (Zaheer et al., 2025). While post-quantum cryptography addresses the mathematical resilience of security mechanisms, effective defense frameworks must integrate cryptographic assurance with architectural design choices and intelligent monitoring capabilities. This section reviews key architectural models, formal security properties, and trust and risk abstractions that collectively support quantum-resilient defense strategies (Saini et al., 2025). Figure 3 Layered architectural view of quantum-resilient defense frameworks, illustrating how security architecture choices integrate with post-quantum cryptography, AI-assisted monitoring, core security property enforcement, and trust and risk reasoning within a unified defense structure (Kodete et al., 2024).

6.1 Architectural models

Architectural design plays a foundational role in determining how quantum-resilient security mechanisms are deployed, scaled, and adapted over time (Kotangale and Kumar, 2025). Centralized architectures provide unified control and simplified policy enforcement, but they may introduce single points of failure and scalability constraints as systems grow in size and complexity (Darzi and Yavuz, 2024c). In quantum-resilient contexts, such architectures require careful integration of cryptographic mechanisms and monitoring components to ensure that centralized decision points remain trustworthy under evolving adversarial models (Tiwo, 2025). Distributed architectures address many of these limitations by decentralizing control and security functions across multiple nodes or domains. This distribution enhances resilience and fault tolerance, while also enabling localized threat detection and response (Roodsari et al., 2025). However, distributed designs introduce coordination and consistency challenges, particularly when cryptographic protocols and security policies must remain coherent across heterogeneous environments transitioning toward post-quantum primitives (Shukur and Dileep, 2023). Federated architectural models offer a complementary approach by enabling collaborative security analysis across organizational or administrative boundaries without requiring full data centralization. In quantum-resilient defense settings, federated models support information sharing and collective threat awareness while preserving autonomy and minimizing exposure of sensitive data (Sankalp et al., 2025). Such architectures are particularly relevant for large-scale ecosystems where cryptographic transitions and threat intelligence must be coordinated without compromising trust boundaries (Owolabi, 2025). Zero-trust-aligned architectures further refine quantum-resilient defense by rejecting implicit trust assumptions and enforcing continuous verification of entities, devices, and processes. In these models, cryptographic assurance and AI-driven monitoring operate in tandem to validate access decisions and behavioral consistency throughout the system lifecycle (Singh et al., 2025). Architectural choices aligned with zero-trust principles enhance adaptability by allowing security policies to evolve dynamically in response to observed risk and threat conditions.

6.2 Formal security properties

Quantum-resilient defense frameworks must preserve core security properties confidentiality, integrity, and availability under revised cryptographic and adversarial assumptions (Kaleem et al., 2025). In post-quantum contexts, confidentiality is maintained through cryptographic schemes whose security arguments remain valid against quantum-capable adversaries (Abrar et al., 2025). However, confidentiality must also be interpreted in conjunction with traffic analysis resistance and behavioral inference risks, which are increasingly addressed through AI-assisted monitoring (Mamun et al., 2024). Integrity assurance in quantum-resilient systems extends beyond cryptographic correctness to include protection against subtle manipulation of system behavior, configuration, or data flows (Kuznetsov et al., 2024). AI-driven monitoring complements cryptographic mechanisms by enabling continuous verification of operational integrity and by identifying deviations that may indicate compromise, misuse, or adversarial interference. Availability, traditionally treated as a system-level property, becomes more complex in quantum-resilient environments due to increased computational overheads, protocol transitions, and adaptive adversarial strategies (Anny, 2025). Defense frameworks must therefore incorporate mechanisms that balance cryptographic robustness with operational continuity. AI-assisted analysis supports this balance by enabling adaptive resource allocation, anomaly-aware mitigation, and early detection of conditions that threaten service availability (Saeed and Alqahtani, 2026). Collectively, these security properties are not static guarantees but evolving objectives that must be continuously assessed and reinforced as cryptographic assumptions and threat models change (Darzi and Yavuz, 2024a). Quantum-resilient defense frameworks must therefore treat confidentiality, integrity, and availability as interdependent and dynamically maintained properties (Acharya et al., 2025).

6.3 Trust and risk modeling

Trust and risk modeling constitute essential components of quantum-resilient defense frameworks, particularly in environments characterized by intelligent and adaptive adversaries (Younis and Al-Hassani, 2026). Traditional static trust models, which rely on fixed assumptions about system components or actors, are insufficient in settings where adversarial capabilities and behaviors evolve over time. Quantum-resilient environments require trust models that can be updated continuously based on observed behavior, cryptographic assurance, and contextual information (Singamaneni et al., 2025). Dynamic risk assessment enables defense frameworks to evaluate the likelihood and potential impact of threats as system conditions change. Rather than assigning fixed risk values, quantum-resilient models incorporate adaptive reasoning that reflects evolving attack surfaces, cryptographic transitions, and adversarial learning (Purohit and Vyas, 2025). AI-driven analysis supports this process by correlating behavioral signals, environmental context, and historical patterns to infer changes in risk posture (Badar et al., 2025). Adaptive trust evaluation further refines this approach by allowing trust relationships to be strengthened, weakened, or revoked in response to observed evidence (Gupta et al., 2026). In quantum-resilient settings, trust decisions are informed jointly by cryptographic guarantees and behavioral consistency, ensuring that neither dimension is relied upon in isolation (Cherbal et al., 2024). This integration is particularly important in distributed and federated architectures, where trust boundaries are fluid and adversarial incentives may vary across domains. By integrating trust and risk modeling into architectural and cryptographic design, quantum-resilient defense frameworks enable proactive and context-aware security management (Hu et al., 2025). These capabilities provide the foundation for the integrated AI–PQC frameworks introduced in the following section, where cryptographic assurance, intelligent threat inference, and adaptive decision-making are combined into a unified defense model.

6.3.1 Formal definitions of trust and risk

To provide formal grounding for trust and risk reasoning in quantum-resilient environments, this section introduces explicit mathematical definitions. Let S = {s1, s2, …, sn} denote the set of system entities in a distributed environment. Trust T (si, t) is defined as a time-varying scalar in [0, 1] representing the degree of confidence assigned to entity si at time t, computed as a weighted combination of three evidence streams: T (si, t) = alpha * C (si) + beta * B (si, t) + gamma * H (si), where C (si) is the cryptographic attestation score, B (si, t) is behavioral consistency at time t, H (si) is historical compliance, and the coefficients alpha, beta, gamma sum to 1 and are calibrated to reflect the relative reliability of each evidence source. Risk R (si, t) is defined as R (si, t) = L (si, t) * I (si), where L (si, t) is the threat likelihood inferred from the AI detection layer and I (si) is the estimated impact of a successful attack on entity si. This formulation makes explicit the dependency between the cryptographic assurance layer, which determines C (si), and the AI inference layer, which determines L (si, t) through the threat hypothesis h. A high cryptographic attestation score combined with anomalous behavioral signals B (si, t) should yield a reduced overall trust score T (si, t), triggering elevated risk R (si, t) and prompting adaptive response action a from the coordination layer. These definitions connect naturally to established trust reasoning frameworks. Subjective Logic (Gupta et al., 2026) provides a principled basis for representing uncertainty in trust evaluations through opinion triplets (belief, disbelief, uncertainty), which is particularly important in quantum-resilient contexts where both cryptographic assumptions and AI model outputs carry inherent epistemic limitations. Dynamic Bayesian Networks (Singamaneni et al., 2025; Purohit and Vyas, 2025) offer a complementary approach for modeling temporal trust evolution, capturing how trust updates propagate across interdependent entities as new behavioral and cryptographic evidence accumulates over time.

7 Quantum-resilient defense frameworks

This section presents an integrated defense framework that unifies post-quantum cryptographic assurance with AI-driven threat hunting to support quantum-resilient cybersecurity operations (De Macêdo et al., 2025). The framework is conceptual by design, focusing on architectural composition, information flow, and formal interaction between cryptographic guarantees and intelligent threat inference rather than implementation-specific details. Its primary objective is to illustrate how cryptographic resilience and adaptive security intelligence can be co-designed to address emerging quantum-enabled and AI-assisted threat models (Bera et al., 2025b).

7.1 Framework overview and components

The integrated AI–PQC defense framework is structured around four tightly coupled components: cryptographic assurance, observation and feature abstraction, intelligent threat inference, and adaptive response coordination (Ahmad et al., 2025). Post-quantum cryptography establishes a trusted substrate that protects communication channels, identities, and system states against quantum-capable adversaries (Reddy et al., 2025). On top of this substrate, AI-driven threat hunting operates to infer malicious intent, detect anomalous behavior, and support informed defensive decision-making. Unlike isolated deployments of cryptographic or AI-based mechanisms, the framework treats cryptographic assurance as a prerequisite for reliable observation and learning (Mamun et al., 2026). Cryptographic mechanisms ensure the integrity and authenticity of security-relevant signals, while AI-driven analysis provides the contextual reasoning required to interpret those signals under uncertainty (SaberiKamarposhti et al., 2024). This interdependence enables the framework to maintain security guarantees while adapting dynamically to evolving threat conditions.

7.1.1 Literature grounding of framework components

Each of the four framework components is grounded in converging evidence from the reviewed literature. The cryptographic assurance layer draws on the NIST-standardized schemes (ML-KEM, ML-DSA, SLH-DSA) and the body of research on lattice-based and hash-based hardness assumptions (Saeed, 2025; Mujlid and Alshahrani, 2025; Darzi and Yavuz, 2024b). Its role as a trusted substrate is consistent with findings by Darzi and Yavuz (2024a), who demonstrate that without quantum-resilient authentication, AI-based monitoring pipelines remain vulnerable to cryptanalytic interception and data manipulation at the observation layer. The observation and feature abstraction layer reflects insights from AI-based intrusion detection research (Ahmad et al., 2025) confirm that meaningful feature extraction from network telemetry is a prerequisite for reliable threat inference, and that the integrity of these features depends directly on the trustworthiness of the channels through which they are collected. This bidirectional dependency between cryptographic protection and feature quality is a recurring theme across the reviewed literature (Şahin, 2026; Alyas et al., 2025; Divyajyothi et al., 2025). The intelligent threat inference layer is grounded in learning-based security analytics research. Badar et al. (Badar et al., 2025) demonstrate that AI-enhanced anomaly detection complements cryptographic access control in smart grid environments. Choudhury et al. (Choudhury et al.) show that ensemble deep learning models maintain detection performance under resource constraints typical of IoT deployments. Nayak et al. (Nayak et al., 2025) confirm that federated learning extends AI-based threat inference to distributed environments while preserving data privacy, a property central to multi-domain quantum-resilient deployments. The adaptive response coordination layer reflects findings from autonomous cyber defense research. Yavuz et al. (Yavuz et al., 2022) establish that distributed post-quantum infrastructures require adaptive coordination mechanisms that adjust cryptographic policies in response to inferred threat states. Chen et al. (Chen et al., 2024) show that privacy-preserving threat intelligence can feed adaptive response decisions without exposing sensitive cryptographic state. Together, these studies confirm that adaptive response is not merely a desirable feature but a structural requirement for operating securely in environments where adversarial capabilities evolve faster than static configurations can accommodate.

7.2 Data flow and observation model

Within the framework, system activity is continuously monitored through cryptographically protected observation channels. Let e t denote a sequence of observed events at time t, including protocol interactions, authentication attempts, and system state transitions (Yavuz et al., 2022). These events are protected by post-quantum cryptographic mechanisms to preserve confidentiality and integrity. Observed events are transformed into abstract feature representations suitable for learning-based analysis through a feature extraction function ϕ:

x t = ϕ e t , x t ∈ F

Here, F denotes a feature space that captures behavioral and structural characteristics without exposing sensitive cryptographic material. This abstraction allows AI-based components to operate independently of specific cryptographic implementations while remaining aligned with post-quantum security requirements.

7.3 Threat inference and decision loop

Threat inference within the framework is modeled as a probabilistic reasoning process over abstracted observations. Let h ∈ H represent a threat hypothesis, where H includes benign, suspicious, and malicious states. Given an observed feature vector x t, the framework infers the most plausible hypothesis according to:

h t * = arg ⁡ max ⁡ h ∈ H P h ∣ x t

This inference step does not assume a specific learning algorithm and may be realized through probabilistic classifiers, anomaly detection models, or hybrid inference mechanisms. The inferred hypothesis feeds into a decision loop that determines appropriate defensive actions based on current risk posture and operational constraints. Let a ∈ A denote a set of admissible defensive actions, such as access restriction, increased monitoring, or cryptographic re-keying. The framework selects actions by minimizing expected risk:where R a , h represents a risk function capturing the potential impact of action a under hypothesis h. This formulation enables adaptive and context-aware response without relying on fixed thresholds or static rules.

a t * = arg ⁡ min ⁡ a ∈ A E h ∼ P h ∣ x t R a , h

7.4 Adaptive feedback and co-evolution

A key characteristic of the integrated AI–PQC framework is its adaptive feedback mechanism. Defensive actions and inferred threat states influence subsequent observations, feature representations, and trust assessments (Xu et al., 2023). This feedback loop allows the framework to refine its threat models and response strategies over time, supporting continuous learning in the presence of intelligent adversaries. Crucially, post-quantum cryptographic mechanisms remain an active component of this adaptive process (Patel et al., 2024). Cryptographic policies, key management strategies, and protocol configurations may be adjusted in response to inferred risk, ensuring that cryptographic assurance evolves alongside threat intelligence (Chen et al., 2024). This co-evolution of cryptographic and AI-driven components distinguishes the framework from static defense models and supports long-term resilience under changing adversarial assumptions.

7.5 Framework scope and limitations

The proposed framework is intentionally abstract and technology-agnostic. It does not prescribe specific cryptographic schemes, learning algorithms, or deployment configurations. Instead, it provides a formal and conceptual foundation that can be instantiated across diverse environments as post-quantum standards and AI techniques mature. By avoiding implementation-specific claims, the framework remains robust to future developments in both quantum computing and artificial intelligence.

7.6 Application to smart grid communication infrastructures

Smart grid communication infrastructures represent one of the most operationally demanding environments for the deployment of integrated AI-PQC defense frameworks. These environments combine real-time control requirements, legacy operational technology (OT) components, and highly constrained communication channels, creating a set of security challenges that neither post-quantum cryptography nor AI-driven threat hunting can address in isolation. This subsection elaborates on how the proposed framework applies specifically to smart grid settings, with attention to the communication protocols, latency constraints, and adversarial models that distinguish this domain from general enterprise cybersecurity.

Smart grid communications rely on a heterogeneous protocol stack that includes IEC 61850 for substation automation and inter-device messaging, DNP3 for supervisory control and data acquisition (SCADA) communications, and GOOSE (Generic Object-Oriented Substation Event) messaging for time-critical protection signaling. Each of these protocols was designed under classical security assumptions and without built-in support for quantum-resilient cryptographic mechanisms. The integration of post-quantum primitives such as ML-KEM or ML-DSA into these protocol stacks introduces non-trivial overhead in terms of key size, handshake latency, and signature verification time. For example, ML-DSA signatures are substantially larger than their ECDSA equivalents, which may exceed the payload constraints of DNP3 frames or conflict with the strict timing requirements of GOOSE messages, where delivery windows are typically measured in single-digit milliseconds (Badar et al., 2025). These constraints mean that PQC integration in smart grid environments cannot follow the same migration path as enterprise IT systems and requires purpose-designed lightweight adaptations or hybrid schemes that preserve timing guarantees while achieving quantum-resilient authentication.

The adversarial threat model applicable to smart grid environments differs significantly from general enterprise settings. Attackers targeting grid infrastructure may combine classical intrusion techniques with quantum-enabled cryptanalytic capabilities to compromise substation authentication, intercept control commands, or manipulate metering data in ways that evade detection by rule-based SCADA security systems. In this context, the harvest-now-decrypt-later threat is particularly acute: adversaries with long-term strategic interests may already be collecting encrypted grid telemetry and control traffic with the intent of decrypting it once sufficiently capable quantum computers become available. Beyond cryptanalysis, smart grid adversaries also pursue operational disruption objectives, including coordinated false data injection attacks against advanced metering infrastructure (AMI), manipulation of demand-response signals, and disruption of protection relay coordination. These multi-layered threat objectives require a defense strategy that simultaneously addresses the cryptographic channel and the behavioral anomaly layers, which is precisely the integration the proposed AI-PQC framework is designed to provide (Badar et al., 2025; Darzi and Yavuz, 2024a).

Within the proposed framework, the cryptographic assurance layer addresses the quantum threat to grid communication channels by enabling the migration of IEC 61850 and DNP3 authentication mechanisms toward NIST-standardized post-quantum schemes. The observation and feature abstraction layer extracts behavioral signals from grid telemetry that are not dependent on payload decryption, including message timing patterns, protocol state transition frequencies, device authentication rates, and deviation from normal load profiles. These features remain informative even under end-to-end encryption and provide the AI inference layer with the input necessary to detect anomalous behaviors consistent with false data injection, relay spoofing, or coordinated denial-of-service campaigns against grid communication channels. The adaptive response coordination layer closes the loop by enabling automated responses proportionate to the inferred threat level, ranging from increased monitoring and alert generation at lower confidence levels to cryptographic re-keying, session isolation, or operator notification at higher threat levels. This graduated response model is essential in OT environments where disruptive automated actions may themselves cause operational harm if triggered incorrectly. Bera et al. (Bera et al., 2025a) provide supporting evidence from IEEE Transactions on Smart Grid, demonstrating that post-quantum communication protocols can be integrated into smart grid networks without prohibitive latency penalties when appropriately scoped to the specific protocol and device class.

Reliability and availability constraints further distinguish smart grid deployments from standard IT security contexts. Grid communication systems are required to maintain five-nines or higher availability, and security mechanisms that introduce queuing delays, connection overhead, or computational bottlenecks at critical relay or RTU nodes are operationally unacceptable. This places a direct constraint on the computational budget available for both PQC operations and AI inference. Future instantiations of the proposed framework in smart grid environments will therefore need to explore hardware-accelerated implementations of post-quantum primitives and lightweight, edge-deployable AI inference models that operate within the resource envelope of embedded grid devices. These requirements represent an open implementation challenge that the conceptual framework identifies but defers to domain-specific engineering work, consistent with the scope of a review-level contribution.

7.6.1 Critical comparative analysis of existing approaches

To address the requirement for systematic critical synthesis raised by the reviewers, the following table maps representative studies from the reviewed literature across six analytical dimensions: PQC family, AI method, threat model, evaluation setting, and key limitation. This analysis reveals cross-cutting patterns and gaps that directly motivate the proposed framework. Table 3 presents a critical comparative analysis of representative PQC–AI integration studies across PQC family, AI method, threat model, evaluation, and key limitation.

TABLE 3. Critical comparative analysis of representative PQC-AI integration studies.

The comparative analysis reveals four cross-cutting patterns. First, the majority of studies evaluate their approaches on synthetic or narrowly scoped datasets, limiting generalizability to real quantum-era threat environments. Second, AI methods range from anomaly detection and deep learning to federated and reservoir computing, yet few explicitly model adversarial robustness of the AI component against poisoning or evasion. Third, PQC integration is predominantly confined to key encapsulation and authentication layers, leaving AI data pipeline integrity largely unaddressed. Fourth, explainability remains a consistent gap: ensemble and deep learning models rarely address how security analysts would interpret outputs in operational settings. These observations directly motivate the open challenges in Section 8 and the research directions in Section 9.

The proposed AI-PQC framework advances beyond each of these individual contributions in several respects that are worth stating explicitly. Unlike Badar et al. (Badar et al., 2025), whose Q-BLAISE protocol is confined to a simulated smart grid IoT environment with no real quantum adversary modelled, the proposed framework is family-agnostic across all four NIST-standardized PQC families and explicitly addresses adversarial robustness at the AI inference layer. Unlike Hanafi et al. (Hanafi et al., 2024), whose adversarial neural cryptography approach is limited to a narrow key-exchange scope with no deployment validation, the proposed framework encompasses a full four-layer architecture integrating cryptographic assurance, feature abstraction, probabilistic threat inference, and adaptive response coordination. Unlike Choudhury et al. (Choudhury et al.), who leave the explainability of ensemble decisions unaddressed, the proposed framework identifies explainability as a first-class design requirement and connects it to post hoc explanation methods such as SHAP and LIME within the threat inference layer. Unlike Nayak et al. (Nayak et al., 2025), whose federated learning approach does not address Byzantine robustness, the proposed framework treats adversarial manipulation of the AI monitoring pipeline as an explicit threat vector within its adaptive co-evolution mechanism. Unlike Darzi and Yavuz (2024d), whose scope is limited to denial-of-service defense in a narrowly scoped experimental setting, the proposed framework provides a unified conceptual architecture applicable across distributed, federated, and zero-trust deployment contexts. In this sense, the framework does not merely catalogue existing approaches but synthesizes their partial contributions into a coherent and formally grounded model that systematically addresses the gaps each individual study leaves open. Figure 4 presents the analytical synthesis of the integrated AI–PQC defense framework, including the literature coverage map and key open challenges.

8 Open challenges and research gaps

Despite the growing interest in post-quantum cryptography and AI-driven threat hunting, their effective convergence remains constrained by several unresolved challenges and open research gaps. These challenges span technical, architectural, and methodological dimensions, and must be addressed to enable robust and deployable quantum-resilient defense frameworks. One significant challenge relates to scalability. Post-quantum cryptographic mechanisms often introduce increased computational and communication complexity at the protocol level, while AI-driven threat hunting systems require continuous data processing and model adaptation. Integrating these components at scale particularly in large, distributed, or resource-constrained environments raises questions about architectural efficiency, coordination overhead, and long-term operational sustainability. Current research lacks systematic analysis of how scalable defense architectures can jointly accommodate post-quantum cryptographic transitions and AI-based monitoring without compromising resilience.

Another open challenge concerns the explainability and transparency of AI-driven security decisions. As threat hunting increasingly relies on learning-based inference, the ability to interpret and justify AI-generated alerts and response actions becomes critical, especially in high-assurance and regulated environments. While cryptographic mechanisms provide mathematically grounded guarantees, AI-based decisions often remain opaque. The absence of explainable threat inference models complicates trust establishment, incident response validation, and human oversight, highlighting a gap between cryptographic assurance and operational accountability. Cryptographic performance overhead represents a further challenge in the convergence of PQC and AI-driven defense. Although post-quantum cryptography aims to preserve long-term security guarantees, its integration into real-world systems may affect protocol behavior, system responsiveness, and monitoring fidelity. Existing studies often examine cryptographic or AI components in isolation, leaving a gap in understanding how performance trade-offs propagate across integrated defense frameworks and influence threat detection and response effectiveness. The limited availability of representative datasets poses an additional research gap. AI-driven threat hunting depends on behavioral data that accurately reflects both benign activity and adversarial behavior. In quantum-resilient contexts, such datasets are inherently scarce, as large-scale quantum-enabled attacks and fully deployed post-quantum infrastructures remain largely prospective. This scarcity limits empirical validation and encourages over-reliance on synthetic or narrowly scoped data, which may not capture the complexity of future threat environments. Finally, validation against future quantum adversaries remains an open methodological challenge. While post-quantum cryptographic schemes are designed based on current theoretical models of quantum computation, the practical capabilities of future adversaries remain uncertain. Similarly, AI-driven threat hunting systems are trained and evaluated under present-day assumptions about attack strategies and system behavior. The lack of established validation frameworks for assessing integrated defenses under evolving and uncertain adversarial models represents a critical gap in current research. Collectively, these challenges highlight that the convergence of post-quantum cryptography and AI-driven threat hunting is not merely a matter of combining technologies, but requires deeper investigation into architectural design, trust reasoning, explainability, and long-term validation. Addressing these gaps is essential for advancing from conceptual integration toward deployable and resilient quantum-era cybersecurity solutions.

8.1 Deeper analysis of persistent challenges

Scalability: This challenge persists for structural rather than incidental reasons. Post-quantum cryptographic operations, particularly lattice-based key encapsulation under ML-KEM and signature verification under ML-DSA, impose computational overheads that scale with the number of authenticated sessions. In large enterprise or critical infrastructure environments, thousands of concurrent authentication events may occur per second. Simultaneously, AI-driven threat hunting requires continuous ingestion and processing of behavioral telemetry at comparable rates. The compound effect of PQC overhead on telemetry collection latency degrades the timeliness of threat inference, creating a tension between cryptographic robustness and detection responsiveness that current research has not systematically resolved. Resource-constrained environments such as industrial control systems and IoT networks face this tension most acutely, since neither PQC nor AI components can be selectively simplified without compromising overall security guarantees.

Explainability: The explainability gap is not merely a usability concern; it is a fundamental barrier to operational trust and regulatory compliance. In finance, healthcare, and critical infrastructure, security decisions must be auditable and defensible. When an AI model flags an event as anomalous and triggers a cryptographic re-keying or access restriction, security analysts must understand what behavioral signals drove the decision and how confident the model is in its assessment. Current deep learning-based threat detection systems rarely provide this transparency. Post-hoc explanation approaches such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations) offer promising directions, but their integration into real-time security operations pipelines remains largely unexplored. This gap is particularly acute in quantum-resilient contexts because PQC transitions introduce new protocol behaviors that may appear anomalous to models trained on pre-migration traffic, generating systematic false positives that only explainable models can reliably diagnose and correct. Adversarial Robustness of AI Components: An underappreciated challenge is the vulnerability of AI-based threat hunting systems to adversarial manipulation. A sophisticated adversary operating in a post-quantum environment may not only attempt to break cryptographic protections but also to poison the behavioral training data that the AI system learns from, inject misleading telemetry to suppress true alerts, or exploit model inversion to infer cryptographic policy configurations from observed model behavior. These adversarial machine learning attack vectors apply directly to integrated AI-PQC systems and represent a threat model that most reviewed studies do not explicitly address. Developing defenses, including robust training procedures, PQC-protected input validation pipelines, and anomaly detection on the AI model’s own inference patterns, constitutes a critical and currently open research priority. Dataset Scarcity and Benchmark Absence: The absence of representative benchmarks for quantum-era cybersecurity evaluation is a structural problem with no near-term solution, because sufficiently capable quantum adversaries do not yet exist at scale. Any AI model trained and evaluated today is calibrated to a classical threat landscape. When quantum-enabled attacks emerge, the behavioral signatures they produce in authentication traffic, key exchange flows, and protocol anomaly patterns may differ qualitatively from anything in existing training corpora. Closing this gap requires investment in high-fidelity simulation environments that model quantum-enabled adversarial capabilities, development of synthetic dataset generation frameworks grounded in realistic quantum attack models, and community agreement on benchmark evaluation protocols analogous to those established for classical intrusion detection. Migration-Period Security: An overlooked challenge concerns security properties during the transition from classical to post-quantum cryptography. Migration is not instantaneous but a multi-year process during which hybrid classical-PQC configurations must operate safely. During this period, AI-based monitoring systems trained on pre-migration traffic patterns must adapt to new cryptographic protocol behaviors, larger key sizes, and different authentication flows, without generating excessive false positives or missing genuine threats. Simultaneously, adversaries aware of an organization’s migration timeline may exploit transitional windows of reduced cryptographic coverage. Designing AI-driven monitoring that remains well-calibrated throughout migration, rather than only in stable pre- or post-migration states, represents a practically significant open challenge for any organization undertaking a structured PQC transition.

9 Future research directions

The convergence of post-quantum cryptography and AI-driven threat hunting opens a range of promising research directions that extend beyond incremental improvements to existing security mechanisms. These directions reflect the need for co-designed, adaptive, and future-aware cybersecurity frameworks capable of operating under evolving computational and adversarial paradigms. One important research direction lies in AI-assisted optimization of post-quantum cryptographic schemes. While post-quantum cryptography provides long-term resilience against quantum-enabled attacks, its integration into operational systems introduces new design and configuration challenges. AI techniques offer opportunities to support cryptographic parameter selection, protocol adaptation, and policy tuning in response to changing system conditions and risk profiles. Future research can explore how learning-based approaches may assist in managing cryptographic agility without compromising formal security guarantees. Another promising direction concerns quantum-aware threat intelligence. As adversarial capabilities evolve, threat intelligence systems must move beyond classical assumptions and incorporate awareness of quantum-related risks and attack vectors. Research in this area may focus on extending existing threat modeling and intelligence frameworks to account for quantum-enabled reconnaissance, cryptanalytic preparation, and hybrid attack strategies. Integrating such intelligence with AI-driven analysis can enhance situational awareness and support proactive defense planning in quantum-resilient environments. The development of autonomous cyber defense systems represents a further research frontier. By combining AI-driven threat inference with adaptive decision-making and cryptographic assurance, future defense systems may be capable of responding to threats with minimal human intervention while remaining aligned with predefined security objectives and trust constraints. Research challenges in this area include balancing autonomy with oversight, ensuring explainability of automated actions, and maintaining robust cryptographic foundations under continuous adaptation. Finally, hybrid classical–quantum security architectures offer a long-term research direction as quantum technologies mature. Such architectures may integrate classical security mechanisms, post-quantum cryptographic primitives, and emerging quantum communication or computation capabilities within unified defense frameworks. Investigating how AI-driven monitoring and reasoning can operate across hybrid security layers will be essential for supporting gradual transitions toward quantum-era infrastructures without disrupting existing systems. Collectively, these research directions highlight that the future of quantum-resilient cybersecurity lies not in isolated technological advances, but in the coordinated evolution of cryptographic assurance, intelligent threat analysis, and adaptive system architecture. Advancing these areas will require interdisciplinary collaboration and a continued emphasis on formal reasoning, architectural coherence, and long-term resilience.

10 Conclusion

This review has examined the emerging convergence between post-quantum cryptography and AI-driven threat hunting as a foundational requirement for future-ready cybersecurity. As quantum computing challenges long-standing cryptographic assumptions and adversaries increasingly leverage intelligent and adaptive attack strategies, security mechanisms designed in isolation are no longer sufficient. Addressing these parallel developments requires defense approaches that integrate cryptographic resilience with intelligent, context-aware threat analysis. By systematically reviewing the technical foundations of quantum-resistant cryptography and AI-based threat hunting, this paper has highlighted their complementary roles and inherent limitations when deployed independently. The analysis has shown that post-quantum cryptography provides essential mathematical assurance against quantum-capable adversaries, while AI-driven threat hunting enables proactive detection, behavioral reasoning, and adaptive response in complex and evolving environments. Their integration forms the basis for defense frameworks capable of preserving trust, visibility, and adaptability under advanced adversarial models. The proposed integrated AI–PQC defense framework consolidates these insights into a coherent conceptual structure, illustrating how cryptographic assurance, feature abstraction, intelligent threat inference, and adaptive decision-making can be co-designed within quantum-resilient architectures. Rather than offering implementation-specific prescriptions, the framework provides a formal and extensible foundation that can guide future research and system design as post-quantum standards and AI techniques continue to mature. In identifying open challenges and future research directions, the review underscores that the transition to quantum-resilient cybersecurity is not a purely cryptographic problem, nor solely an AI challenge. It is an architectural and methodological endeavor that demands interdisciplinary collaboration, careful reasoning about trust and risk, and sustained attention to explainability, scalability, and long-term validation. Overall, this work contributes a structured and analytically grounded perspective on the convergence of post-quantum cryptography and AI-driven threat hunting. By framing this convergence as a necessary evolution in cybersecurity design philosophy, the paper aims to support the development of resilient, adaptive, and future-aware defense frameworks capable of withstanding the uncertainties of the quantum era.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Cristian Randieri, University of eCampus, Italy

Reviewed by: Belinda Mutunhu Ndlovu, National University of Science and Technology, Zimbabwe

Elyson De La Cruz, Imagine Believe Realize, United States

Padmavathi U, Shiv Nadar University Chennai, India