Papers

Web agents automate browser tasks, ranging from simple form completion to complex workflows like ordering groceries. While current benchmarks evaluate general-purpose performance~(e.g., WebArena) or safety against malicious actions~(e.g., SafeArena), no existing framework assesses an agent's ability to successfully execute user-facing website security and privacy tasks, such as managing cookie preferences, configuring privacy-sensitive account settings, or revoking inactive sessions. To address this gap, we introduce WebSP-Eval, an evaluation framework for measuring web agent performance on website security and privacy tasks. WebSP-Eval comprises 1) a manually crafted task dataset of 200 task instances across 28 websites; 2) a robust agentic system supporting account and initial state management across runs using a custom Google Chrome extension; and 3) an automated evaluator. We evaluate a total of 8 web agent instantiations using state-of-the-art multimodal large language models, conducting a fine-grained analysis across websites, task categories, and UI elements. Our evaluation reveals that current models suffer from limited autonomous exploration capabilities to reliably solve website security and privacy tasks, and struggle with specific task categories and websites. Crucially, we identify stateful UI elements are a primary reason for agent failure, with toggles causing more than 45% task failure across many models.

Blockchain and artificial intelligence (AI) are increasingly proposed together for securing intelligent networks, but the literature remains fragmented across ledger design, AI-driven detection, cyber-physical applications, and emerging agentic workflows. This paper synthesizes the area through three reusable contributions: (i) a taxonomy of blockchain-AI security for intelligent networks, (ii) integration patterns for verifiable and adaptive security workflows, and (iii) the Blockchain-AI Security Evaluation Blueprint (BASE), a reporting checklist spanning AI quality, ledger behavior, end-to-end service levels, privacy, energy, and reproducibility. The paper also maps the evidence landscape across IoT, critical infrastructure, smart grids, transportation, and healthcare, showing that the conceptual fit is strong but real-world evidence remains uneven and often prototype-heavy. The synthesis clarifies where blockchain contributes provenance, trust, and auditability, where AI contributes detection, adaptation, and orchestration, and where future work should focus on interoperable interfaces, privacy-preserving analytics, bounded agentic automation, and open cross-domain benchmarks. The paper is intended as a reference for researchers and practitioners designing secure, transparent, and resilient intelligent networks.

Given the growing reliance on private data in training Large Language Models (LLMs), Federated Learning (FL) combined with Parameter-Efficient Fine-Tuning (PEFT) has garnered significant attention for enhancing privacy and efficiency. Despite FL's privacy benefits, prior studies have shown that private data can still be extracted from shared gradients. However, these studies, mainly on full-parameter model training, are limited to reconstructing small batches, short input sequences, and specific model architectures, such as encoder-based or decoder-based models. The reconstruction quality becomes even worse when dealing with gradients from PEFT methods. To fully understand the practical attack surface of federated LLMs, this paper proposes FedSpy-LLM, a scalable and generalizable data reconstruction attack designed to reconstruct training data with larger batch sizes and longer sequences while generalizing across diverse model architectures, even when PEFT methods are deployed for training. At the core of FedSpy-LLM is a novel gradient decomposition strategy that exploits the rank deficiency and subspace structure of gradients, enabling efficient token extraction while preserving key signal components at scale. This approach further mitigates the reconstruction challenges introduced by PEFT's substantial null space, ensuring robustness across encoder-based, decoder-based, and encoder-decoder model architectures. Additionally, by iteratively aligning each token's partial-sequence gradient with the full-sequence gradient, FedSpy-LLM ensures accurate token ordering in reconstructed sequences.

In this paper, we analyze and improve the adversarial robustness of a convolutional neural network (CNN) that assists crystal-collimator alignment at CERN's Large Hadron Collider (LHC) by classifying a beam-loss monitor (BLM) time series during crystal rotation. We formalize a local robustness property for this classifier under an adversarial threat model based on real-world plausibility. Building on established parameterized input-transformation patterns used for transformation- and semantic-perturbation robustness, we instantiate a preprocessing-aware wrapper for our deployed time-series pipeline: we encode time-series normalization, padding constraints, and structured perturbations as a lightweight differentiable wrapper in front of the CNN, so that existing gradient-based robustness frameworks can operate on the deployed pipeline. For formal verification, data-dependent preprocessing such as per-window z-normalization introduces nonlinear operators that require verifier-specific abstractions. We therefore focus on attack-based robustness estimates and pipeline-checked validity by benchmarking robustness with the frameworks Foolbox and ART. Adversarial fine-tuning of the resulting CNN improves robust accuracy by up to 18.6 % without degrading clean accuracy. Finally, we extend robustness on time-series data beyond single windows to sequence-level robustness for sliding-window classification, introduce adversarial sequences as counterexamples to a temporal robustness requirement over full scans, and observe attack-induced misclassifications that persist across adjacent windows.

Vision-Language Models (VLMs) have become essential for tasks such as image synthesis, captioning, and retrieval by aligning textual and visual information in a shared embedding space. Yet, this flexibility also makes them vulnerable to malicious prompts designed to produce unsafe content, raising critical safety concerns. Existing defenses either rely on blacklist filters, which are easily circumvented, or on heavy classifier-based systems, both of which are costly and fragile under embedding-level attacks. We address these challenges with two complementary components: Hyperbolic Prompt Espial (HyPE) and Hyperbolic Prompt Sanitization (HyPS). HyPE is a lightweight anomaly detector that leverages the structured geometry of hyperbolic space to model benign prompts and detect harmful ones as outliers. HyPS builds on this detection by applying explainable attribution methods to identify and selectively modify harmful words, neutralizing unsafe intent while preserving the original semantics of user prompts. Through extensive experiments across multiple datasets and adversarial scenarios, we prove that our framework consistently outperforms prior defenses in both detection accuracy and robustness. Together, HyPE and HyPS offer an efficient, interpretable, and resilient approach to safeguarding VLMs against malicious prompt misuse.

Autonomous AI agents powered by Large Language Models can reason, plan, and execute complex tasks, but their ability to autonomously retrieve information and run code introduces significant security risks. Existing approaches attempt to regulate agent behavior through training or prompting, which does not offer fundamental security guarantees. We present ClawLess, a security framework that enforces formally verified policies on AI agents under a worst-case threat model where the agent itself may be adversarial. ClawLess formalizes a fine-grained security model over system entities, trust scopes, and permissions to express dynamic policies that adapt to agents' runtime behavior. These policies are translated into concrete security rules and enforced through a user-space kernel augmented with BPF-based syscall interception. This approach bridges the formal security model with practical enforcement, ensuring security regardless of the agent's internal design.

In recent years, the pace of development of information technology in various areas has increased drastically, forcing cybersecurity specialists to constantly review existing processes in order to prevent unauthorized access to confidential information. Using Ukraine as a primary case study, this paper explores the integration of international best practices, specifically ISO/IEC 27001 and the NIST Cybersecurity Framework, into national regulatory systems. A focus is placed on the transition from traditional compliance models to risk-based approaches, exemplified by the recent adoption of the Ukrainian normative documents. Furthermore, we propose a methodology for automating the development of target security profiles using Large Language Models (LLMs) enhanced by RetrievalAugmented Generation (RAG). By integrating a vector database of national regulations and organizational policies, the proposed RAG-based advisor reduces manual complexity, minimizes human error, and ensures alignment between technical controls and legal requirements. This study contributes to the field by providing a structured workflow for AI-assisted cybersecurity management in environments characterized by high-intensity hybrid threats.

The Zero-trust (ZT) model is an increasingly popular model that relies on the idea that no trust should be granted to any entity (network, persons, devices) by default. ZT model is gaining attention from both research and practice, with various levels of adequation between research developed and real-life applications. NIST provided a standard to fulfill requirements of ZT architecture of network core but many practical aspects remain unspecified, some of them requiring solving first research challenges in order to be implemented efficiently. An example of such an unspecified field is the integration of IoT/Smart Peripheral Devices (SPD). Various reasons explain this gap: specificities of such resources (possibly lower energy/computation power), their lifecycle, and their use, strongly depending on the use of the whole platform IoT devices are part of. Moreover, additional difficulty to have a good understanding is induced by the fact that both Zero Trust and IoT are identified as promising trends in cybersecurity: many vendors/researchers tag their solutions as IoT integration into the ZT model, with little to no effective compliance to ZT model or standard. Industry is providing many practice-oriented literature, that has to be compared to academic work and standards, in order to consolidate the current state of knowledge and solutions offered to realize this integration. In this paper, we conduct a literature review of non-academic publications, in order to consolidate current knowledge, trends, and future challenges for the industrial integration of IoT devices in ZT architecture.

Software-Defined Networking (SDN) improves network flexibility but also increases the need for reliable and interpretable intrusion detection. Large Language Models (LLMs) have recently been explored for cybersecurity tasks due to their strong representation learning capabilities; however, their lack of transparency limits their practical adoption in security-critical environments. Understanding how LLMs make decisions is therefore essential. This paper presents an attribution-driven analysis of encoder-based LLMs for network intrusion detection using flow-level traffic features. Attribution analysis demonstrates that model decisions are driven by meaningful traffic behavior patterns, improving transparency and trust in transformer-based SDN intrusion detection. These patterns align with established intrusion detection principles, indicating that LLMs learn attack behavior from traffic dynamics. This work demonstrates the value of attribution methods for validating and trusting LLM-based security analysis.

The governance of artificial intelligence has a blind spot: the machine identities that AI systems use to act. AI agents, service accounts, API tokens, and automated workflows now outnumber human identities in enterprise environments by ratios exceeding 80 to 1, yet no integrated framework exists to govern them. A single ungoverned automated agent produced $5.4-10 billion in losses in the 2024 CrowdStrike outage; nation-state actors including Silk Typhoon and Salt Typhoon have operationalized ungoverned machine credentials as primary espionage vectors against critical infrastructure. This paper makes four original contributions. First, the AI-Identity Risk Taxonomy (AIRT): a comprehensive enumeration of 37 risk sub-categories across eight domains, each grounded in documented incidents, regulatory recognition, practitioner prevalence data, and threat intelligence. Second, the Machine Identity Governance Taxonomy (MIGT): an integrated six-domain governance framework simultaneously addressing the technical governance gap, the regulatory compliance gap, and the cross-jurisdictional coordination gap that existing frameworks address only in isolation. Third, a foreign state actor threat model for enterprise identity governance, establishing that Silk Typhoon, Salt Typhoon, Volt Typhoon, and North Korean AI-enhanced identity fraud operations have already operationalized AI identity vulnerabilities as active attack vectors. Fourth, a cross-jurisdictional regulatory alignment structure mapping enterprise AI identity governance obligations under EU, US, and Chinese frameworks simultaneously, identifying irreconcilable conflicts and providing a governance mechanism for managing them. A four-phase implementation roadmap translates the MIGT into actionable enterprise programs.

In the light of the growing connectivity and sensitivity of industrial data, cyberattacks and data breaches are becoming more common in the Industrial Internet of Things (IIoT). To cope with such threats, this study presents an anomaly detection system based on a novel Federated Learning (FL) framework. This system detects anomalies such as cyberattacks and protects industrial data privacy by processing data locally and training anomaly detection models on industrial agents without sharing raw data. The proposed FL framework incorporates two key components to enhance both privacy and efficiency. The first component is Homomorphic Encryption (HE), which is integrated into the framework to further protect sensitive data transmissions such as model parameters. HE enhances privacy in FL by preventing adversaries from inferring private industrial data through attacks, such as model inversion attacks. The second component is an innovative dynamic agent selection scheme, wherein a selection threshold is calculated based on agent delays and data size. The purpose of this new scheme is to mitigate the straggler effect and the communication bottleneck that occur in traditional FL architectures, such as synchronous and asynchronous architectures. It ensures that agents are not unfairly selected by the different delays resulting from heterogeneous data in IIoT environments, while simultaneously improving model performance and convergence speed. The proposed framework exhibits superior performance over baseline approaches in terms of accuracy, precision, F1-scores, communication costs, convergence speeds, and fairness rate.

Post-quantum migration in TLS 1.3 couples signature-algorithm choice with certificate-hierarchy structure, chain exposure during the handshake, and role-dependent cryptographic cost. In certificate-based authentication, the practical effect of a signature family depends on where it appears in the certification hierarchy, how much of that hierarchy is exposed during the handshake, and how the resulting cryptographic cost is distributed across client and server roles. Post-quantum TLS migration must therefore be evaluated as cryptographic design within authenticated key establishment, with algorithm selection assessed in its deployment context. This paper presents a local experimental study of TLS 1.3 authentication strategies implemented with OpenSSL 3 and oqsprovider. Using a reproducible laboratory setting, it compares ML-DSA and SLH-DSA across multiple certificate placements, hierarchy depths, and key-exchange modes, including classical, hybrid, and pure post-quantum configurations. The analysis is organized into four complementary campaigns: a leaf-only comparison, a full hierarchy strategy matrix, a depth comparison, and a key-exchange exploration. Across the experimental matrix, the main discontinuity appears when SLH-DSA is placed in the server leaf certificate. In that configuration, handshake latency and server-side compute cost increase by orders of magnitude, whereas strategies that confine SLH-DSA to upper trust layers and preserve ML-DSA in the interactive leaf remain within a more plausible operational range. The results also show that transport size alone does not explain the heavy regime: outside leaf-SLH scenarios, transferred bytes and observed chain size track latency closely, but once SLH-DSA reaches the leaf, server-side cryptographic cost becomes dominant.

Code decompilation analysis is a fundamental yet challenging task in malware reverse engineering, particularly due to the pervasive use of sophisticated obfuscation techniques. Although recent large language models (LLMs) have shown promise in translating low-level representations into high-level source code, most existing approaches rely on generic code pretraining and lack adaptation to malicious software. We propose LLM4CodeRE, a domain-adaptive LLM framework for bidirectional code reverse engineering that supports both assembly-to-source decompilation and source-to-assembly translation within a unified model. To enable effective task adaptation, we introduce two complementary fine-tuning strategies: (i) a Multi-Adapter approach for task-specific syntactic and semantic alignment, and (ii) a Seq2Seq Unified approach using task-conditioned prefixes to enforce end-to-end generation constraints. Experimental results demonstrate that LLM4CodeRE outperforms existing decompilation tools and general-purpose code models, achieving robust bidirectional generalization.

Cryptocurrencies such as Bitcoin are defined by protocols that specify how participants record transactions and create new currency units. These protocols are not enforced by law or any central entity and instead are intended to be incentive compatible. However, the Bitcoin mining protocol proposed by Nakamoto (2008) and implemented in practice is known not to constitute an equilibrium (Eyal and Sirer, 2018). This leaves open the question of whether the decentralized outcome intended by Nakamoto can be sustained in equilibrium in the Bitcoin mining game. We propose inertial mining, a novel mining protocol that induces that outcome, i.e., a single longest chain in which each miner's asymptotic share of blocks equals its share of computational power. Our main result establishes that inertial mining constitutes an equilibrium, assuming no miner controls one half or more of the computational power. Inertial mining coincides with Nakamoto's protocol on the equilibrium path, and can be implemented in Bitcoin without any changes to its consensus mechanism or blockchain architecture. When a single miner controls more than half of the computational power, we show that no decentralized equilibrium exists.

The advancement of Large Language Models (LLMs) has raised concerns regarding their dual-use potential in cybersecurity. Existing evaluation frameworks overwhelmingly focus on Information Technology (IT) environments, failing to capture the constraints, and specialized protocols of Operational Technology (OT). To address this gap, we introduce CritBench, a novel framework designed to evaluate the cybersecurity capabilities of LLM agents within IEC 61850 Digital Substation environments. We assess five state-of-the-art models, including OpenAI's GPT-5 suite and open-weight models, across a corpus of 81 domain-specific tasks spanning static configuration analysis, network traffic reconnaissance, and live virtual machine interaction. To facilitate industrial protocol interaction, we develop a domain-specific tool scaffold. Our empirical results show that agents reliably execute static structured-file analysis and single-tool network enumeration, but their performance degrades on dynamic tasks. Despite demonstrating explicit, internalized knowledge of the IEC 61850 standards terminology, current models struggle with the persistent sequential reasoning and state tracking required to manipulate live systems without specialized tools. Equipping agents with our domain-specific tool scaffold significantly mitigates this operational bottleneck. Code and evaluation scripts are available at: https://github.com/GKeppler/CritBench

The Model Context Protocol (MCP), introduced by Anthropic in November 2024 and now governed by the Linux Foundation's Agentic AI Foundation, has rapidly become the de facto standard for connecting large language model (LLM)-based agents to external tools and data sources, with over 97 million monthly SDK downloads and more than 177000 registered tools. However, this explosive adoption has exposed a critical gap: the absence of a unified, formal security framework capable of systematically characterizing, analyzing, and mitigating the diverse threats facing MCP-based agent ecosystems. Existing security research remains fragmented across individual attack papers, isolated benchmarks, and point defense mechanisms. This paper presents MCPSHIELD, a comprehensive formal security framework for MCP-based AI agents. We make four principal contributions: (1) a hierarchical threat taxonomy comprising 7 threat categories and 23 distinct attack vectors organized across four attack surfaces, grounded in the analysis of over 177000 MCP tools; (2) a formal verification model based on labeled transition systems with trust boundary annotations that enables static and runtime analysis of MCP tool interaction chains; (3) a systematic comparative evaluation of 12 existing defense mechanisms, identifying coverage gaps across our threat taxonomy; and (4) a defense in depth reference architecture integrating capability based access control, cryptographic tool attestation, information flow tracking, and runtime policy enforcement. Our analysis reveals that no existing single defense covers more than 34 percent of the identified threat landscape, whereas MCPSHIELD's integrated architecture achieves theoretical coverage of 91 percent. We further identify seven open research challenges that must be addressed to secure the next generation of agentic AI systems.

The deployment of large language models (LLMs) in Swiss financial and regulatory contexts demands empirical evidence of both production reliability and adversarial security, dimensions not jointly operationalized in existing Swiss-focused evaluation frameworks. This paper introduces Swiss-Bench 003 (SBP-003), extending the HAAS (Helvetic AI Assessment Score) from six to eight dimensions by adding D7 (Self-Graded Reliability Proxy) and D8 (Adversarial Security). I evaluate ten frontier models across 808 Swiss-specific items in four languages (German, French, Italian, English), comprising seven Swiss-adapted benchmarks (Swiss TruthfulQA, Swiss IFEval, Swiss SimpleQA, Swiss NIAH, Swiss PII-Scope, System Prompt Leakage, and Swiss German Comprehension) targeting FINMA Guidance 08/2024, the revised Federal Act on Data Protection (nDSG), and OWASP Top 10 for LLMs. Self-graded D7 scores (73-94%) exceed externally judged D8 security scores (20-61%) by a wide margin, though these dimensions use non-comparable scoring regimes. System prompt leakage resistance ranges from 24.8% to 88.2%, while PII extraction defense remains weak (14-42%) across all models. Qwen 3.5 Plus achieves the highest self-graded D7 score (94.4%), while GPT-oss 120B achieves the highest D8 score (60.7%) despite being the lowest-cost model evaluated. All evaluations are zero-shot under provider default settings; D7 is self-graded and does not constitute independently validated accuracy. I provide conceptual mapping tables relating benchmark dimensions to FINMA model validation requirements, nDSG data protection obligations, and OWASP LLM risk categories.

This paper presents Text-Guided Backdoor (TGB), an adjustable backdoor attack against multimodal pretrained models that uses natural-word triggers, namely words that can naturally occur in ordinary textual inputs. Most existing backdoor attacks require specific trigger conditions that are typically not satisfied by ordinary inference inputs, thereby limiting their activation in real-world deployments. TGB overcomes this limitation by exploiting naturally occurring words as triggers, enabling stealthy activation without requiring explicit trigger insertion at inference time. This property avoids conspicuous trigger patterns and improves the practicality of TGB. Furthermore, we introduce visual adversarial perturbations on poisoned samples to modulate the model's learning of natural-word triggers, thereby enabling flexible adjustment of TGB attack strength without modifying the poisoned data. Extensive experiments are conducted on downstream tasks built upon multimodal pretrained models, including Composed Image Retrieval (CIR) and Visual Question Answering (VQA). Results demonstrate the effectiveness of TGB across diverse poisoning settings and its ability to flexibly adjust attack success rates, which reveal critical security vulnerabilities in multimodal pretrained models.

In LLM agent pipelines, prompt privacy risk propagates beyond a single model call: raw user content enters retrieval queries, memory writes, tool arguments, OCR-derived text, and logs, and every downstream copy inherits what the first write contained. Existing de-identification pipelines protect document boundaries but not this cross-stage surface. We present BodhiPromptShield, a policy-aware mediation layer that detects sensitive spans before they propagate, replaces each with a typed placeholder, a semantic abstraction, or a secure symbolic token under a configured policy, and defers restoration to authorized execution boundaries. We evaluate it under one protocol against Presidio, Casper-style sanitization, an LLM sanitizer, and transformer and learned detectors, on 300 AI4Privacy documents, 493 PrivacyLens trajectories, 200 PrivacyLens tasks scored by that benchmark's own judge, and AgentDojo tasks under injection. Three findings result. Identifier propagation is controllable: residual exposure falls to 7.4% on AI4Privacy and 1.8% on PrivacyLens, and exact identifiers in an agent's final action fall from 13.7% to 2.1-3.1%. Restoration timing governs what every stage upstream of the authorized boundary sees: deferring it leaves 1.6% of protected values readable in the released context against 51.0%, and 2.7% against 4.8% in what the agent emits, for 0.11 helpfulness points. Measuring factual disclosure is harder: a word-overlap metric and an LLM judge both report that mediation leaves facts intact, and both disagree with blind human annotation (kappa = 0.25 and 0.09). The human labels reverse that: inferability falls from 100% to 24-53% under mediation, so semantic-leakage measures need human validation before they are trusted. These are systems results on English text with open-weight models, not formal guarantees.

Anti-forensics includes a growing set of techniques designed to obstruct forensic analysis. While cybercriminals increasingly rely on these methods, they also help researchers identify and remedy weaknesses in forensic tools, advancing the overall robustness of digital forensics. Despite repeated efforts to define it, anti-forensics remains vague and inconsistent in its use. It also poses ethical challenges regarding the appropriateness of research practices and the legitimacy of the field itself. This article presents a systematic analysis of 123 publications on anti-forensics, combining qualitative and quantitative methods. We quantify the main techniques and attack vectors, examine their occurrence in different digital forensic subdomains, and identify typical research methods, motivations, and applications. This work also discusses what these findings mean for future research and proposes directions for building a more coherent and ethically grounded understanding of anti-forensics.

The rapid advancement of Large Language Models (LLMs) has created new opportunities for Automated Penetration Testing (AutoPT), spawning numerous frameworks aimed at achieving end-to-end autonomous attacks. However, despite the proliferation of related studies, existing research generally lacks systematic architectural analysis and large-scale empirical comparisons under a unified benchmark. Therefore, this paper presents the first Systematization of Knowledge (SoK) focusing on the architectural design and comprehensive empirical evaluation of current LLM-based AutoPT frameworks. At systematization level, we comprehensively review existing framework designs across six dimensions: agent architecture, agent plan, agent memory, agent execution, external knowledge, and benchmarks. At empirical level, we conduct large-scale experiments on 13 representative open-source AutoPT frameworks and 2 baseline frameworks utilizing a unified benchmark. The experiments consumed over 10 billion tokens in total and generated more than 1,500 execution logs, which were manually reviewed and analyzed over four months by a panel of more than 15 researchers with expertise in cybersecurity. By investigating the latest progress in this rapidly developing field, we provide researchers with a structured taxonomy to understand existing LLM-based AutoPT frameworks and a large-scale empirical benchmark, along with promising directions for future research.

Cyber-physical systems often contend with incomplete architectural documentation or outdated information resulting from legacy technologies, knowledge management gaps, and the complexity of integrating diverse subsystems over extended operational lifecycles. This architectural incompleteness impedes reliable security assessment, as inaccurate or missing architectural knowledge limits the identification of system dependencies, attack surfaces, and risk propagation pathways. To address this foundational challenge, this paper introduces ASTRAL (Architecture-Centric Security Threat Risk Assessment using LLMs), an architecture-centric security assessment technique implemented in a prototype tool powered by multimodal LLMs. The proposed approach assists practitioners in reconstructing and analysing CPS architectures when documentation is fragmented or absent. By leveraging prompt chaining, few-shot learning, and architectural reasoning, ASTRAL extracts and synthesises system representations from disparate data sources. By integrating LLM reasoning with architectural modelling, our approach supports adaptive threat identification and quantitative risk estimation for cyber-physical systems. We evaluated the approach through an ablation study across multiple CPS case studies and an expert evaluation involving 14 experienced cybersecurity practitioners. Practitioner feedback suggests that ASTRAL is useful and reliable for supporting architecture-centric security assessment. Overall, the results indicate that the approach can support more informed cyber risk management decisions.

This paper studies the creation of textual descriptions of user activities and interactions on smartphones. Our approach of referring to encrypted mobile traffic exceeds traditional smartphone activity classification methods in terms of model scalability and output readability. The paper addresses two obstacles to the realization of this idea: the semantic gap between traffic features and smartphone activity captions, and the lack of textually annotated traffic data. To overcome these challenges, we introduce a novel smartphone activity captioning system, called T2T (Traffic-to-Text). T2T consists of a flow feature encoder that converts low-level traffic characteristics into meaningful latent features and a caption decoder to yield readable transcripts of smartphone activities. In addition, T2T achieves the automatic textual annotation of mobile traffic by feeding synchronized screen capture videos into the Qwen-VL-Max vision-language model, and proposing multi-stage losses for effective cross-model training. We evaluate T2T on 40,000 traffic-description pairs collected in two real-world environments, involving 8 smartphone users and 20 mobile apps. T2T achieves a BLEU-4 score of 58.1, a METEOR score of 38.3, a ROUGE-L score of 70.5, and a CIDEr score of 108.7. The quantitative and qualitative analyses show that T2T can generate semantically accurate captions that are comparable to the vision-language model.

We present a layered and modular network architecture that combines Quantum Key Distribution (QKD) and Post-Quantum Cryptography (PQC) to provide scalable end-to-end security across long distance multi-hop, trusted-node quantum networks. To ensure interoperability and efficient practical deployment, hop-wise tunnels between physically secured nodes are protected by WireGuard with periodically rotated pre-shared keys sourced via the ETSI GS QKD 014 interface. On top, Rosenpass performs a PQC key exchange to establish an end-to-end data channel without modifying deployed QKD devices or network protocols. This dual-layer composition yields post-quantum forward secrecy and authenticity under practical assumptions. We implement the design using open-source components and validate and evaluate it in simulated and lab test-beds. Experiments show uninterrupted operation over multi-hop paths, low resource footprint and fail-safe mechanisms. We further discuss the design's compositional security, wherein the security of each individual component is preserved under their combination and outline migration paths for operators integrating QKD-aware overlays in existing infrastructures.

Agentic AI systems are increasingly deployed as personal assistants and virtual employees, and are likely to become a common object of digital investigations and incident response. Yet little is known about how their internal state and actions can be reconstructed during forensic analysis, and systematic forensic approaches remain largely unexplored. We present the first empirical forensic study of a personal AI assistant, with OpenClaw as a case study. We examine its technical design via static code analysis and apply differential forensic analysis to identify recoverable traces across stages of the agent interaction loop. We classify and correlate these traces, structuring the examination around representative investigative questions. We propose a five-plane agent artifact taxonomy that captures recurring investigative patterns; its structural completeness is assessed by cross-mapping against published artifact inventories of two related AI systems. To aid investigations, we release an open-source artifact-extraction tool with a machine-readable schema. Finally, we highlight a foundational challenge: LLM-mediated execution adds both an abstraction layer and substantial nondeterminism to trace generation. Because tool selection and state transitions are co-determined by the model, the execution environment, and the evolving context, identical inputs need not yield identical traces.

GenAI smartphones, which natively embed generative AI at the system level, are transforming mobile interactions by automating a wide range of tasks and executing UI actions on behalf of users. Their superior capabilities rely on continuous access to sensitive and context-rich data, raising privacy concerns that surpass those of traditional mobile devices. Yet, little is known about how users perceive the privacy implications of such devices or what safeguards they expect, which is especially critical at this early stage of GenAI smartphone adoption. To address this gap, we conduct 22 semi-structured interviews with everyday mobile users to explore their usage of GenAI smartphones, privacy concerns, and privacy design expectations. Our findings show that users engage with GenAI smartphones with limited understanding of how these systems operate to deliver functions, but show heightened privacy concerns once exposed to the technical details. Participants' concerns span the entire data lifecycle, including nontransparent collection, insecure storage, and weak data control. In a follow-up focus group, participants discuss a range of privacy-enhancing suggestions that call for coordinated changes across system-level controls, data management practices, and user-facing transparency. Their concerns and suggestions offer user-centered guidances for designing GenAI smartphones that balance functionality with privacy protection, offering valuable takeaways for system designers and regulators.

Protecting the intellectual property of open-weight large language models (LLMs) requires verifying whether a suspect model is derived from a victim model despite common laundering operations such as fine-tuning (including PPO/DPO), pruning/compression, and model merging. We propose AttnDiff, a data-efficient white-box framework that extracts fingerprints from models via intrinsic information-routing behavior. AttnDiff probes minimally edited prompt pairs that induce controlled semantic conflicts, captures differential attention patterns, summarizes them with compact spectral descriptors, and compares models using CKA. Across Llama-2/3 and Qwen2.5 (3B–14B) and additional open-source families, it yields high similarity for related derivatives while separating unrelated model families (e.g., $>0.98$ vs.\ $<0.22$ with $M=60$ probes). With 5–60 multi-domain probes, it supports practical provenance verification and accountability.

Vector databases serve as the retrieval backbone of modern AI applications, yet their security remains largely unexplored. We propose the Black-Hole Attack, a poisoning attack that injects a small number of malicious vectors near the geometric center of the stored vectors. These injected vectors attract queries like a black hole and frequently appear in the top-k retrieval results for most queries. This attack is enabled by a phenomenon we term centrality-driven hubness: in high-dimensional embedding spaces, vectors near the centroid become nearest neighbors of a disproportionately large number of other vectors, while this centroid region is nearly empty in practice. The attack shows that vectors in a vector database cannot be blindly trusted: geometric defects in high-dimensional embeddings make retrieval inherently vulnerable. Based on this insight, we propose four attack paths tailored to different attacker capabilities. Our experiments show that up to 94.4% of queries are successfully attacked. Additionally, we study two directions of defense: hubness mitigation and detection-based filtering. Hubness mitigation either significantly reduces retrieval accuracy or provides only limited protection, while the detection-based defense is effective against some attack paths but fails against others. A robust and adaptive defense thus remains an open problem, and our findings indicate that vector databases require more careful treatment of security.

Network Intrusion Detection Systems (NIDS) face important limitations. Signature-based methods are effective for known attack patterns, but they struggle to detect zero-day attacks and often miss modified variants of previously known attacks, while many machine learning approaches offer limited interpretability. These challenges become even more severe in IoT environments because of resource constraints and heterogeneous protocols. To address these issues, we propose MA-IDS, a Multi-Agent Intrusion Detection System that combines Large Language Models (LLMs) with Retrieval Augmented Generation (RAG) for reasoning-driven intrusion detection. The proposed framework grounds LLM reasoning through a persistent, self-building Experience Library. Two specialized agents collaborate through a FAISS-based vector database: a Traffic Classification Agent that retrieves past error rules before each inference, and an Error Analysis Agent that converts misclassifications into human-readable detection rules stored for future retrieval, enabling continual learning through external knowledge accumulation, without modifying the underlying language model. Evaluated on NF-BoT-IoT and NF-ToN-IoT benchmark datasets, MA-IDS achieves Macro F1-Scores of 89.75% and 85.22%, improving over zero-shot baselines of 17% and 4.96% by more than 72 and 80 percentage points. These results are competitive with SVM while providing rule-level explanations for every classification decision, demonstrating that retrieval-augmented reasoning offers a principled path toward explainable, self-improving intrusion detection for IoT networks.

Modern Security Operations Centers struggle with alert fatigue, fragmented tooling, and limited cross-source event correlation. Challenges that current Security Information Event Management and Extended Detection and Response systems only partially address through fragmented tools. This paper presents the LLM-assisted network Governance (LanG), an open-source, governance-aware agentic AI platform for unified security operations contributing: (i) a Unified Incident Context Record with a correlation engine (F1 = 87%), (ii) an Agentic AI Orchestrator on LangGraph with human-in-the-loop checkpoints, (iii) an LLM-based Rule Generator finetuned on four base models producing deployable Snort 2/3, Suricata, and YARA rules (average acceptance rate 96.2%), (iv) a Three-Phase Attack Reconstructor combining Louvain community detection, LLM-driven hypothesis generation, and Bayesian scoring (87.5% kill-chain accuracy), and (v) a layered Governance-MCP-Agentic AI-Security architecture where all tools are exposed via the Model Context Protocol, governed by an AI Governance Policy Engine with a two-layer guardrail pipeline (regex + Llama Prompt Guard 2 semantic classifier, achieving 98.1% F1 score with experimental zero false positives). Designed for Managed Security Service Providers, the platform supports multi-tenant isolation, role-based access, and fully local deployment. Finetuned anomaly and threat detectors achieve weighted F1 scores of 99.0% and 91.0%, respectively, in intrusion-detection benchmarks, running inferences in $\approx$21 ms with a machine-side mean time to detect of 1.58 s, and the rule generator exceeds 91% deployability on live IDS engines. A systematic comparison against eight SOC platforms confirms that LanG uniquely satisfies multiple industrial capabilities all in one open-source tool, while enforcing selected AI governance policies.