Papers

The Industrial Internet of Things (IIoT) introduces significant security challenges as resource-constrained devices become increasingly integrated into critical industrial processes. Existing security approaches typically address threats at a single network layer, often relying on expensive hardware and remaining confined to simulation environments. In this paper, we present the research framework and contributions of our doctoral thesis, which aims to develop a lightweight, Machine Learning (ML)-based security framework for IIoT environments. We first describe our adoption of the Tm-IIoT trust model and the Hybrid IIoT (H-IIoT) architecture as foundational baselines, then introduce the Trust Convergence Acceleration (TCA) approach, our primary contribution that integrates ML to predict and mitigate the impact of degraded network conditions on trust convergence, achieving up to a 28.6% reduction in convergence time while maintaining robustness against adversarial behaviors. We then propose a real-world deployment architecture based on affordable, open-source hardware, designed to implement and extend the security framework. Finally, we outline our ongoing research toward multi-layer attack detection, including physical-layer threat identification and considerations for robustness against adversarial ML attacks.

Recently, multimodal large language models (MLLMs) have emerged as a unified paradigm for language and image generation. Compared with diffusion models, MLLMs possess a much stronger capability for semantic understanding, enabling them to process more complex textual inputs and comprehend richer contextual meanings. However, this enhanced semantic ability may also introduce new and potentially greater safety risks. Taking diffusion models as a reference point, we systematically analyze and compare the safety risks of emerging MLLMs along two dimensions: unsafe content generation and fake image synthesis. Across multiple unsafe generation benchmark datasets, we observe that MLLMs tend to generate more unsafe images than diffusion models. This difference partly arises because diffusion models often fail to interpret abstract prompts, producing corrupted outputs, whereas MLLMs can comprehend these prompts and generate unsafe content. For current advanced fake image detectors, MLLM-generated images are also notably harder to identify. Even when detectors are retrained with MLLMs-specific data, they can still be bypassed by simply providing MLLMs with longer and more descriptive inputs. Our measurements indicate that the emerging safety risks of the cutting-edge generative paradigm, MLLMs, have not been sufficiently recognized, posing new challenges to real-world safety.

Differential privacy (DP) is crucial for safeguarding sensitive client information in federated learning (FL), yet traditional DP-FL methods rely predominantly on fixed gradient clipping thresholds. Such static clipping neglects significant client heterogeneity and varying privacy sensitivities, which may lead to an unfavorable privacy-utility trade-off. In this paper, we propose PAC-DP, a Personalized Adaptive Clipping framework for federated learning under record-level local differential privacy. PAC-DP introduces a Simulation-CurveFitting approach leveraging a server-hosted public proxy dataset to learn an effective mapping between personalized privacy budgets epsilon and gradient clipping thresholds C, which is then deployed online with a lightweight round-wise schedule. This design enables budget-conditioned threshold selection while avoiding data-dependent tuning during training. We provide theoretical analyses establishing convergence guarantees under the per-example clipping and Gaussian perturbation mechanism and a reproducible privacy accounting procedure. Extensive evaluations on multiple FL benchmarks show that PAC-DP surpasses conventional fixed-threshold approaches under matched privacy budgets, improving accuracy by up to 26% and accelerating convergence by up to 45.5% in our evaluated settings.

The proliferation of local Large Language Model (LLM) runners, such as Ollama, LM Studio and llama.cpp, presents a new challenge for digital forensics investigators. These tools enable users to deploy powerful AI models in an offline manner, creating a potential evidentiary blind spot for investigators. This work presents a systematic, cross platform forensic analysis of these popular local LLM clients. Through controlled experiments on Windows and Linux operating systems, we acquired and analyzed disk and memory artifacts, documenting installation footprints, configuration files, model caches, prompt histories and network activity. Our experiments uncovered a rich set of previously undocumented artifacts for each software, revealing significant differences in evidence persistence and location based on application architecture. Key findings include the recovery of plaintext prompt histories in structured JSON files, detailed model usage logs and unique file signatures suitable for forensic detection. This research provides a foundational corpus of digital evidence for local LLMs, offering forensic investigators reproducible methodologies, practical triage commands and analyse this new class of software. The findings have critical implications for user privacy, the admissibility of AI-related evidence and the development of anti-forensic techniques.

With frequently evolving Advanced Persistent Threats (APTs) in cyberspace, traditional security solutions approaches have become inadequate for threat hunting for organizations. Moreover, SOC (Security Operation Centers) analysts are often overwhelmed and struggle to analyze the huge volume of logs received from diverse devices in organizations. To address these challenges, we propose an automated and dynamic threat hunting framework for monitoring evolving threats, adapting to changing network conditions, and performing risk-based prioritization for the mitigation of suspicious and malicious traffic. By integrating Agentic AI with Splunk, an established SIEM platform, we developed a unique threat hunting framework. The framework systematically and seamlessly integrates different threat hunting modules together, ranging from traffic ingestion to anomaly assessment using a reconstruction-based autoencoder, deep reinforcement learning (DRL) with two layers for initial triage, and a large language model (LLM) for contextual analysis. We evaluated the framework against a publicly available benchmark dataset, as well as against a simulated dataset. The experimental results show that the framework can effectively adapt to different SOC objectives autonomously and identify suspicious and malicious traffic. The framework enhances operational effectiveness by supporting SOC analysts in their decision-making to block, allow, or monitor network traffic. This study thus enhances cybersecurity and threat hunting literature by presenting the novel threat hunting framework for security decision-making, as well as promoting cumulative research efforts to develop more effective frameworks to battle continuously evolving cyber threats.

The Google Play marketplace has introduced the Data Safety section to improve transparency regarding how mobile applications (apps) collect, share, and protect user data. This mechanism requires developers to disclose privacy and security-related practices. However, the reliability of these disclosures remains dependent on developer self-reporting, raising concerns about their accuracy. This study investigates the consistency between developer-reported Data Safety disclosures and observable privacy indicators extracted from Android Application Packages (APKs). An empirical analysis was conducted on a dataset of 41 mobile gaming apps. A static analysis approach was used to extract key privacy indicators from APK files, including device IDs, data sharing, personal information access, and location access. These indicators were systematically compared with the corresponding disclosures reported in the Google Play Data Safety labels using a structured consistency evaluation framework. The results revealed varying levels of agreement across privacy categories. Device ID disclosures demonstrated relatively high consistency (87.8%), whereas other indicators exhibited substantial mismatches. Location-related disclosures showed the highest inconsistency rate (56.1%), followed by personal information and data sharing. Comparative analysis between children-oriented and general-audience apps revealed similar mismatch patterns. Also, Chi-square statistical tests indicate that these differences are not statistically significant, suggesting that disclosure inconsistencies are not associated with app category but instead reflect broader ecosystem-level challenges. These findings highlight limitations in the reliability of current marketplace transparency mechanisms and emphasize the need for improved validation and verification approaches to ensure accurate privacy reporting in mobile app ecosystems.

Financial systems have a growing reliance on computer-based and distributed systems, making FinTech systems vulnerable to advanced and quickly emerging cyber-criminal threats. Traditional security systems and fixed machine learning systems cannot identify more intricate fraud schemes whilst also addressing real-time performance and trust demands. This paper presented an Adaptive Neuro-Fuzzy Blockchain-AI Framework (ANFB-AI) to achieve security in FinTech transactions by detecting threats using intelligent and decentralized algorithms. The framework combines both an immutable, transparent and tamper resistant layer of a permissioned blockchain to maintain the immutability, transparency and resistance to tampering of transactions, and an adaptive neuro-fuzzy learning model to learn the presence of uncertainty and behavioural drift in fraud activities. An explicit mathematical model is created to explain the transaction integrity, adaptive threat classification, and unified risk based decision-making. The proposed framework uses Proof-of-Authority consensus to overcome low-latency validation of transactions and scalable real-time financial services. Massive simulations are performed in normal, moderate, and high-fraud conditions with the use of realistic financial and cryptocurrency transactions. The experimental evidence proves that ANFB-AI is always more accurate and precise than recent state-of-the-art algorithms and costs much less in terms of transaction confirmation time, propagation delay of blocks and end-to end latency. ANFB-AI performance supports the appropriateness of adaptive neuro-fuzzy intelligence to blockchain-based FinTech security.

Large language models (LLMs) are increasingly deployed on edge devices under strict computation and quantization constraints, yet their security implications remain unclear. We study query-based knowledge extraction from quantized edge-deployed LLMs under realistic query budgets and show that, although quantization introduces noise, it does not remove the underlying semantic knowledge, allowing substantial behavioral recovery through carefully designed queries. To systematically analyze this risk, we propose CLIQ (Clustered Instruction Querying), a structured query construction framework that improves semantic coverage while reducing redundancy. Experiments on quantized Qwen models (INT8/INT4) demonstrate that CLIQ consistently outperforms original queries across BERTScore, BLEU, and ROUGE, enabling more efficient extraction under limited budgets. These results indicate that quantization alone does not provide effective protection against query-based extraction, highlighting a previously underexplored security risk in edge-deployed LLMs.

AI agent protocols – including MCP, A2A, ANP, and ACP – enable autonomous agents to discover capabilities, delegate tasks, and compose services across trust boundaries. Despite massive deployment (MCP alone has 97M+ monthly SDK downloads), no systematic security framework for these protocols exists. We present three contributions. First, the Agent Protocol Stack, a 6-layer architectural model that defines what a complete agent protocol must specify at each layer – analogous to ITU-T X.800 for the OSI stack. Second, the Agent-Agnostic Security Model, 11 security principles formalized as TLA+ invariants, each tagged with a property taxonomy (spec-mandated, spec-recommended, aasm-hardening, aps-completeness) that distinguishes protocol non-conformance from framework-imposed security requirements. Third, AgentConform, a two-phase conformance checker that (i)extracts normative clauses from protocol specifications into a typed Protocol~IR with explicit Protocol/Environment/Adversary action separation, (ii)compiles the IR into TLA+ models and model-checks them against AASM invariants, then (iii)replays counterexample traces against live SDK implementations to confirm findings. We introduce the Composition Safety (CS) principle: security properties that hold for individual protocols can break when protocols are composed through shared infrastructure. We demonstrate this with formal models of five protocol composition patterns, revealing cross-protocol design gaps that individual protocol analysis cannot detect. Preliminary application to representative agent protocols reveals recurrent gaps in credential lifecycle, consent enforcement, audit completeness, and composition safety. Some findings are under coordinated disclosure; full evaluation details will be released in the complete version.

The safety of autonomous AI agents is increasingly recognized as a critical open problem. As agents transition from passive text generators to active actors capable of executing shell commands, modifying files, calling APIs, and browsing the web, the consequences of unsafe or adversarially manipulated behavior become immediate and tangible. Existing AI safety work has focused primarily on model alignment and input filtering, but these approaches do not address what happens at the moment an agent's intent becomes a real action on a real machine. This gap is especially acute in local environments, where developers run agents against their own filesystems, credentials, and infrastructure with little runtime control. This paper introduces AgentWall, a runtime safety and observability layer for local AI agents. AgentWall intercepts every proposed agent action before it reaches the host environment, evaluates it against an explicit declarative policy, requires human approval for sensitive operations, and records a complete execution trail for audit and replay. It is implemented as a policy-enforcing MCP proxy and native OpenClaw plugin, working across Claude Desktop, Cursor, Windsurf, Claude Code, and OpenClaw with a single install command. We present the design, architecture, threat model, and policy model of AgentWall, and demonstrate 92.9% policy enforcement accuracy with sub-millisecond overhead across 14 benchmark tests. AgentWall is open-source at https://github.com/agentwall/Agentwall.

This study investigates the applicability of authorship attribution based on stylistic features to support actor analysis in threat intelligence. As a foundational step toward future application to dark web forums, we conducted experiments using Japanese review data from clear web sources. We constructed datasets from Rakuten Ichiba reviews and compared four methods: TF-IDF with logistic regression (TF-IDF+LR), BERT embeddings with logistic regression (BERT-Emb+LR), BERT fine-tuning (BERT-FT), and metric learning with $k$-nearest neighbors (Metric+kNN). Results showed that BERT-FT achieved the best performance; however, training became unstable as the number of authors scaled to several hundred, where TF-IDF+LR proved superior in terms of accuracy, stability, and computational cost. Furthermore, Top-$k$ evaluation demonstrated the utility of candidate screening, and error analysis revealed that boilerplate text, topic dependency, and short text length were primary factors causing misclassification.

In a companion paper, we prove that the Burau-Lyapunov exponent LE discriminates focused from dispersed privilege escalation ratchets in cloud IAM graphs, and that no abelian statistic can replicate this discrimination. To strengthen this claim beyond its synthetic validation corpus, we apply the identical pipeline, with zero parameter retuning, to solar coronal magnetic fields: a physical system with no connection to cloud identity and access management, whose binary eruptive/confined outcome is independently established by decades of astrophysical observation.

Peer-discovery protocols within P2P networks are often vulnerable: because creating network identities is essentially free, adversaries can eclipse honest nodes or partition the overlay. This threat is especially acute for blockchains, whose security depends on resilient peer connectivity. We present AetherWeave, a stake-backed peer-discovery protocol that ties network participation to deposited stake, raising the cost of large-scale attacks. We prove that, with high probability, either the honest overlay remains connected or a $(1{-}\delta)$-fraction of nodes in every smaller component raise an attack-detection flag – even against a very powerful adversary. To our knowledge, AetherWeave is the first peer-discovery protocol to simultaneously provide Sybil resistance and privacy: nodes prove they hold valid stake without revealing which deposit they own, and gossiping does not expose peer-table contents. A cryptographic commitment scheme rate-limits discovery requests per round; exceeding the limit yields a publicly verifiable misbehavior proof that triggers on-chain slashing. Beyond deposit and slashing, the protocol requires no on-chain interaction, with per-node communication scaling as $O(s\sqrt{n})$. We validate our design through a mean-field analysis with closed-form convergence bounds, extensive adversarial simulations, and an end-to-end prototype built by forking Prysm, a leading Ethereum consensus client.

Deploying large language models (LLMs) as autonomous browser agents exposes a significant attack surface in the form of Indirect Prompt Injection (IPI). Cloud-based defenses can provide strong semantic analysis, but they introduce latency and raise privacy concerns. We present the Cognitive Firewall, a three-stage split-compute architecture that distributes security checks across the client and the cloud. The system consists of a local visual Sentinel, a cloud-based Deep Planner, and a deterministic Guard that enforces execution-time policies. Across 1,000 adversarial samples, edge-only defenses fail to detect 86.9% of semantic attacks. In contrast, the full hybrid architecture reduces the overall attack success rate (ASR) to below 1% (0.88% under static evaluation and 0.67% under adaptive evaluation), while maintaining deterministic constraints on side-effecting actions. By filtering presentation-layer attacks locally, the system avoids unnecessary cloud inference and achieves an approximately 17,000x latency advantage over cloud-only baselines. These results indicate that deterministic enforcement at the execution boundary can complement probabilistic language models, and that split-compute provides a practical foundation for securing interactive LLM agents.

The widespread adoption of web applications has made their security a critical concern and has increased the need for systematic ways to assess whether they can be considered trustworthy. However, "trust" assessment remains an open problem as existing techniques primarily focus on detecting known vulnerabilities or depend on manual evaluation, which limits their scalability; therefore, evaluating adherence to secure coding practices offers a complementary, pragmatic perspective by focusing on observable development behaviors. In practice, the identification and verification of secure coding practices are predominantly performed manually, relying on expert knowledge and code reviews, which is time-consuming, subjective, and difficult to scale. This study presents an empirical methodology to automate the trustworthiness assessment of web applications by leveraging Large Language Models (LLMs) to verify adherence to secure coding practices. We conduct a comparative analysis of prompt engineering techniques across five state-of-the-art LLMs, ranging from baseline zero-shot classification to prompts enriched with semantic definitions, structural context derived from call graphs, and explicit instructional guidance. Furthermore, we propose an extension of a hierarchical Quality Model (QM) based on the Logic Score of Preference (LSP), in which LLM outputs are used to populate the model's quality attributes and compute a holistic trustworthiness score. Experimental results indicate that excessive structural context can introduce noise, whereas rule-based instructional prompting improves assessment reliability. The resulting trustworthiness score allows discriminating between secure and vulnerable implementations, supporting the feasibility of using LLMs for scalable and context-aware trust assessment.

The emergence of decentralized satellite networks creates a pressing need for trust architectures that operate without physical access to hardware, without pre-provisioned vendor secrets, and without dependence on a single manufacturer's attestation service. Terrestrial TEEs are insufficient: hardware-based designs are susceptible to physical attacks, and most platforms root their attestation chains in secrets provisioned during manufacturing, creating a pre-launch trust window and single-vendor dependency that cannot be independently audited. We present Space Fabric, an architecture that provides the missing trust foundation for orbital computing by relocating the trusted computing stack to satellite infrastructure, exploiting post-launch physical inaccessibility as a tamper barrier unattainable by terrestrial deployments. Our Satellite Execution Assurance Protocol binds workload execution to a specific satellite via a Byzantine-tolerant endorsement quorum of distributed ground stations, certifying not only what executes inside the TEE but also where. All cryptographic secrets are generated within co-located secure elements after launch, with no signing keys accessible on Earth at any point. To reduce single-vendor dependence, Space Fabric distributes its trust anchor across two independent secure elements, an NXP SE050 and a TROPIC01, both of which must co-sign attestation evidence. We implement Space Fabric on a USB Armory Mk II with ARM TrustZone, verify attestation end-to-end using Veraison, and provide a security analysis with satisfaction arguments and impossibility bounds under a strong adaptive adversary.

Multi-chain ecosystems suffer from fragmented identity, siloed liquidity, and bridge-dependent token transfers. We present n-VM, a Layer-1 architecture that hosts n heterogeneous virtual machines as co-equal execution environments over shared consensus and shared state. The design combines three components: a dispatcher that routes transactions by opcode prefix, a unified identity layer in which one 32-byte commitment anchors VM-specifific addresses, and a unified token ledger that exposes VM-native interfaces such as ERC-20 and SPL over a common balance store. We formalize routing, identity derivation, and token transfer semantics, and prove cross-VM transfer atomicity and identity isolation under standard cryptographic assumptions. We describe a concrete instantiation with five VMs: a native runtime, EVM, SVM, Bitcoin Script, and TVM. We also present context-based sharding and a write-set scheduler for parallel execution. Under an analytical throughput model, the architecture admits a projected range of about 16,000 to 66,000 transactions per second on commodity hardware.

Federated Learning (FL) enables heterogeneous clients to collaboratively train a shared model without centralizing their raw data, offering an inherent level of privacy. However, gradients and model updates can still leak sensitive information, while malicious servers may mount adversarial attacks such as Byzantine manipulation. These vulnerabilities highlight the need to address differential privacy (DP) and Byzantine robustness within a unified framework. Existing approaches, however, often rely on unrealistic assumptions such as bounded gradients, require auxiliary server-side datasets, or fail to provide convergence guarantees. We address these limitations by proposing Byz-Clip21-SGD2M, a new algorithm that integrates robust aggregation with double momentum and carefully designed clipping. We prove high-probability convergence guarantees under standard $L$-smoothness and $\sigma$-sub-Gaussian gradient noise assumptions, thereby relaxing conditions that dominate prior work. Our analysis recovers state-of-the-art convergence rates in the absence of adversaries and improves utility guarantees under Byzantine and DP settings. Empirical evaluations on CNN and MLP models trained on MNIST further validate the effectiveness of our approach.

Cybersecurity data remains fragmented across vendors, formats, schemas, and deployment environments, forcing AI and analytics programs to spend disproportionate effort on ingestion, normalization, and brittle source-specific engineering. This paper introduces the Canonical Security Telemetry Substrate (CSTS), a canonical, AI-ready telemetry foundation designed to harmonize heterogeneous cyber data into a common representation over persistent entities, typed relations, events, temporal state, and provenance. CSTS is intended to move cybersecurity analytics beyond ad hoc record normalization toward a reusable substrate that supports anomaly detection, graph learning, forecasting, behavior-based modeling, and agentic cyber AI. We formalize the core design principles of CSTS, define its representational components, and explain how it preserves source-specific nuance through explicit mappings and extensible metadata while still enabling portable downstream inference. We further position CSTS as a cloud-agnostic and deployment-agnostic substrate suitable for on-prem, hybrid, and multi-cloud environments. The result is a unifying telemetry model that reduces the blue-collar burden of cyber data engineering and creates a clearer path to scalable, interoperable, and model-agnostic cyber AI.

The integration of machine learning (ML) algorithms into Internet of Things (IoT) applications has introduced significant advantages alongside vulnerabilities to adversarial attacks, especially within IoT-based intrusion detection systems (IDS). While theoretical adversarial attacks have been extensively studied, practical implementation constraints have often been overlooked. This research addresses this gap by evaluating the feasibility of evasion attacks on IoT network-based IDSs, employing a novel black-box adversarial attack. Our study aims to bridge theoretical vulnerabilities with real-world applicability, enhancing understanding and defense against sophisticated threats in modern IoT ecosystems. Additionally, we propose a defense scheme tailored to mitigate the impact of evasion attacks, thereby reinforcing the resilience of ML-based IDSs. Our findings demonstrate successful evasion attacks against IDSs, underscoring their susceptibility to advanced techniques. In contrast, we proposed a defense mechanism that exhibits robust performance by effectively detecting the majority of adversarial traffic, showcasing promising outcomes compared to current state-of-the-art defenses. By addressing these critical cybersecurity challenges, our research contributes to advancing IoT security and provides insights for developing more resilient IDS.

Industrial deployments increasingly rely on Open Platform Communications Unified Architecture (OPC UA) as a secure and platform-independent communication protocol, while private Fifth Generation (5G) networks provide low-latency and high-reliability connectivity for modern automation systems. However, their combination introduces new attack surfaces and traffic characteristics that remain insufficiently understood, particularly with respect to machine learning-based intrusion detection systems (ML-based IDS). This paper presents an experimental study on detecting cyberattacks against OPC UA applications operating over an operational private 5G network. Multiple attack scenarios are executed, and OPC UA traffic is captured and enriched with statistical flow-, packet-, and protocol-aware features. Several supervised ML models are trained and evaluated to distinguish benign and malicious traffic. The results demonstrate that the proposed ML-based IDS achieves high detection performance for a representative set of OPC UA-specific attack scenarios over an operational private 5G network.

Ring-mapping protocols need a canonical byte-to-residue layer before any algebraic encryption step can begin. This paper isolates that layer and presents the base-m length codec, a canonical map from byte strings of length less than 2^64 to lists of residues modulo m. The encoder builds on and adapts an rANS-based system proposed by Duda. Decoding is exact for all moduli satisfying the paper's parameter bounds. Because the encoding carries the byte length in its fixed-width header, decoding is also tolerant to appended valid suffix digits. The paper is accompanied by a Rust implementation of the described protocol, a Lean 4 formalization of the abstract codec with machine-checked proofs, and performance benchmarks. The Lean 4 formalization establishes fixed-width prefix inversion and payload-state bounds below 2^64, stream-level roundtrip correctness, and that every emitted symbol is a valid residue modulo m. We conclude with a complexity analysis and a discussion of practical considerations arising in real-world use of the codec.

The latest Wi-Fi security standard, IEEE 802.11, includes a secure authentication protocol called SAE, whose use is mandatory for WPA3-Personal networks. The protocol is specified at two separate but linked levels: a traditional cryptographic description of the communication logic between network devices, and a state machine description that realises the former in each single device. Current formal verification efforts focus mainly on communication logic. We present detailed formal models of the protocol at both levels, provide precise specifications of its security properties, and analyse machine-checked proofs in ProVerif and ASMETA. The integrated analysis of the above two models is particularly novel, enabling us to identify and address several issues in the current IEEE 802.11 specification more thoroughly than would have been possible otherwise, leading to several official revisions of the standard.

Financial institutions face increasing cyber risk while operating under strict regulatory oversight. To manage this risk, they rely heavily on Cyber Threat Intelligence (CTI) to inform detection, response, and strategic security decisions. Artificial intelligence (AI) is widely suggested as a means to strengthen CTI. However, evidence of trustworthy production use in finance remains limited. Adoption depends not only on predictive performance, but also on governance, integration into security workflows and analyst trust. Thus, we examine how AI is used for CTI in practice within financial institutions and what barriers prevent trustworthy deployment. We report a mixed-methods, user-centric study combining a CTI-finance-focused systematic literature review, semi-structured interviews, and an exploratory survey. Our review screened 330 publications (2019-2025) and retained 12 finance-relevant studies for analysis; we further conducted six interviews and collected 14 survey responses from banks and consultancies. Across research and practice, we identify four recurrent socio-technical failure modes that hinder trustworthy AI-driven CTI: (i) shadow use of public AI tools outside institutional controls, (ii) license-first enablement without operational integration, (iii) attacker-perception gaps that limit adversarial threat modeling, and (iv) missing security for the AI models themselves, including limited monitoring, robustness evaluation and audit-ready evidence. Survey results provide additional insights: 71.4% of respondents expect AI to become central within five years, 57.1% report infrequent current use due to interpretability and assurance concerns and 28.6% report direct encounters with adversarial risks. Based on these findings, we derive three security-oriented operational safeguards for AI-enabled CTI deployments.

Large Language Models(LLMs) are widely deployed, yet are vulnerable to jailbreak prompts that elicit policy-violating outputs. Although prior studies have uncovered these risks, they typically treat all tokens as equally important during prompt mutation, overlooking the varying contributions of individual tokens to triggering model refusals. Consequently, these attacks introduce substantial redundant searching under query-constrained scenarios, reducing attack efficiency and hindering comprehensive vulnerability assessment. In this work, we conduct a token-level analysis of refusal behavior and observe that token contributions are highly skewed rather than uniform. Moreover, we find strong cross-model consistency in refusal tendencies, enabling the use of a surrogate model to estimate token-level contributions to the target model's refusals. Motivated by these findings, we propose TriageFuzz, a token-aware jailbreak fuzzing framework that adapts the fuzz testing approach with a series of customized designs. TriageFuzz leverages a surrogate model to estimate the contribution of individual tokens to refusal behaviors, enabling the identification of sensitive regions within the prompt. Furthermore, it incorporates a refusal-guided evolutionary strategy that adaptively weights candidate prompts with a lightweight scorer to steer the evolution toward bypassing safety constraints. Extensive experiments on six open-source LLMs and three commercial APIs demonstrate that TriageFuzz achieves comparable attack success rates (ASR) with significantly reduced query costs. Notably, it attains a 90% ASR with over 70% fewer queries compared to baselines. Even under an extremely restrictive budget of 25 queries, TriageFuzz outperforms existing methods, improving ASR by 20-40%.

Fully Homomorphic Encryption (FHE) is rapidly emerging as a promising foundation for privacy-preserving cloud services, enabling computation directly on encrypted data. As FHE implementations mature and begin moving toward practical deployment in domains such as secure finance, biomedical analytics, and privacy-preserving AI, a critical question remains insufficiently explored: how reliable is FHE computation on real hardware? This question is especially important because, compared with plaintext computation, FHE incurs much higher computational overhead, making it more susceptible to transient hardware faults. Moreover, data corruptions are likely to remain silent: the FHE service has no access to the underlying plaintext, causing unawareness even though the corresponding decrypted result has already been corrupted. To this end, we conduct a comprehensive evaluation of SDCs in FHE ciphertext computation. Through large-scale fault-injection experiments, we characterize the vulnerability of FHE to transient faults, and through a theoretical analysis of error-propagation behaviors, we gain deeper algorithmic insight into the mechanisms underlying this vulnerability. We further assess the effectiveness of different fault-tolerance mechanisms for mitigating these faults.

Given two linear codes, the Linear Equivalence Problem (LEP) asks to find (if it exists) a linear isometry between them; as a special case, we have the Permutation Equivalence Problem (PEP), in which isometries must be permutations. LEP and PEP have recently gained renewed interest as the security foundations for several post-quantum schemes, including LESS. A recent paper has introduced the use of the Schur product to solve PEP, identifying many new easy-to-solve instances. In this paper, we extend this result to LEP. In particular, we generalize the approach and rely on the more general notion of power codes. Combining it with Frobenius automorphisms and Hermitian hulls, we identify many classes of easy LEP instances. To the best of our knowledge, this is the first work exploiting algebraic weaknesses for LEP. Finally we show an improved reduction to PEP whenever the coefficients of the monomial matrix are in a subgroup of the multiplicative group of the finite field.

Private Membership Testing (PMT) protocols enable clients to verify whether a certain data item is included in a database without revealing the item to the database operator or other external parties. This paper examines Source-assisted PMT (SPMT), in which clients leverage compact data source-provided information issued when the data item is first submitted to the database. SPMT is relevant in applications such as certificate transparency and supply-chain auditing; yet, designing an approach that is efficient, scalable, and privacy-preserving remains a challenge. This work presents Gyokuro, which takes a different approach to conventional membership testing schemes. Instead of requesting the server to produce a proof attesting that a certain data item exists in the database, we leverage Trusted Execution Environments (TEEs) to produce proofs demonstrating that the server has made enough progress to add the data item to the database. With the help of existing monitoring services, clients can infer that no items have been removed from the database. This allows Gyokuro to provide strong privacy guaranties and achieve high efficiency, as a client's membership testing query does not include any information regarding their interests, and eliminates the need for complex and inefficient protection mechanisms. Additionally, this approach enables membership testing on large-scale databases, since the communication and computation required are independent of the database size. Our evaluations show practical feasibility, achieving 7 ms membership testing latency and throughput of around 1400 requests/sec/core.

European Digital Identity (EUDI) Wallet aims to provide end users with a way to get attested credentials from issuers, and present them to different relying parties. An important property mentioned in the regulatory frameworks is the possibility to revoke a previously issued credential. While it is possible to issue a short-lived credential, in some cases it may be inconvenient, and a separate revocation service which allows to revoke a credential at any time may be necessary. In this work, we propose a full end-to-end description of a generic credential revocation system, which technically relies on a single server and secure transmission channels between parties. We prove security of the proposed revocation functionality in the universal composability model, and estimate its efficiency based on a proof-of-concept implementation.

Coordination of view coverage via privacy-aware smart cameras is key to a more socially responsible urban intelligence. Rather than maximizing view coverage at any cost or over relying on expensive cryptographic techniques, we address how cameras can coordinate to legitimately monitor public spaces while excluding privacy-sensitive regions by design. This article proposes a decentralized framework in which interactive smart cameras coordinate to autonomously select their orientation via collective learning, while eliminating privacy violations via soft and hard constraint satisfaction. The approach scales to hundreds up to thousands of cameras without any centralized control. Experimental evidence shows 18.42% higher coverage efficiency and 85.53% lower privacy violation than baselines and other state-of-the-art approaches. This significant advance further unravels practical guidelines for operators and policymakers: how the field of view, spatial placement, and budget of cameras operating by ethically-aligned artificial intelligence jointly influence coverage efficiency and privacy protection in large-scale and sensitive urban environments.