Papers

Code-driven auditing fails when correctness depends on what the specification requires rather than how the code is written. Production blockchain networks expose this directly: byzantine consensus runs many independent clients of a shared specification, so a specification-divergence defect in one client can fork the network or halt finality. Existing tools reason one repository at a time, with no shared baseline held constant across implementations. We present SPECA, an LLM-driven audit framework that derives explicit, categorized security properties (invariants, pre/postconditions, trust assumptions) from natural-language specifications and reuses them across implementations. SPECA enables controlled cross-implementation comparison, detections grounded in specification invariants no code pattern encodes, and false positives traceable to a specific pipeline phase rather than opaque model errors. On the Sherlock Ethereum Fusaka Audit Contest (10 targets, 366 submissions), SPECA recovers all 15 in-scope H/M/L vulnerabilities expert-augmented (8/15 automated-only) and surfaces 4 fix-confirmed bugs, including a cryptographic-invariant violation missed by every adjudicated finding. On the RepoAudit C/C++ benchmark, SPECA reaches 88.9% precision at 100% recall (F1=0.94) and surfaces 12 author-validated bugs beyond ground truth, two externally validated. SPECA also flags 5 of RepoAudit's 40 published bugs as defensive-coding fixes with no reachable exploit path. False positives trace to three pipeline-pinned root causes; a multi-model study identifies property-generation quality as the binding constraint. End-to-end cost is ~$30 per H/M/L bug (~42 min wall-clock under parallel execution).

Differentially private (DP) contrastive learning aims to learn general-purpose representations from sensitive data, alleviating the privacy leakage concerns of organizations deploying or sharing embedding models trained on private user content. However, existing approaches suffer from severe utility degradation due to the over-strong inter-sample dependency inherent in standard contrastive objectives, where each sample's gradient depends on all other samples in the batch, amplifying the impact of DP noise. In this work, we argue that effective DP contrastive learning requires explicitly reducing such intrinsic inter-sample reliance. To this end, we propose DP-GCL, a principled DP contrastive learning framework that structurally limits gradient dependency through bounding group-level contribution. DP-GCL partitions each batch into small, disjoint groups and restricts available negative samples to within-group samples, thereby localizing gradient influence and reducing sensitivity. To counteract the resulting loss of negative sample diversity, we further introduce intra-group augmentation, which generates additional negative views without increasing privacy cost. Extensive experiments across eight datasets demonstrate that DP-GCL consistently advances the state of the art in both uni-modal and multi-modal contrastive learning under practical privacy budgets: it improves image classification accuracy by 5.6% and image-text retrieval accuracy by 20.1% over existing DP contrastive methods.

Ensuring the integrity of quantum circuits is a significant challenge in the Noisy Intermediate-Scale Quantum (NISQ) era, where circuits are subject to compilation transformations, hardware constraints, and potential adversarial modifications. Existing validation approaches typically rely on either structural analysis or behavioral evaluation, leading to incomplete assessment of circuit correctness. In this work, we investigate the relationship between structural, interaction-level, and behavioral perspectives of circuit integrity, demonstrating that a single aspect of integrity is insufficient to guarantee circuit integrity; structural similarity alone does not ensure behavioral equivalence. To address this problem, we use a three-layer metric framework that combines the Structural Integrity Score (SIS), the Operational Integrity Score (OIS), and the Interaction Graph Semantic-Logical Score (IGS). SIS captures global structural properties, OIS quantifies behavioral divergence using Jensen-Shannon distance, and IGS models interaction patterns and dependencies in a pre-execution setting. Through controlled anomaly injection on benchmark quantum circuits, we demonstrate that each metric captures a different aspect of circuit deviation. In particular, structural blind-spot cases (SIS >= 0.95) reveal a clear limitation of structural analysis, where OIS detects anomalies in 93.85% of instances, while IGS detects 72.58%. These results highlight that the metrics provide complementary insights and that a single metric is insufficient for reliable circuit validation.

This paper presents Quantum Gatekeeper, a context-bound image steganography framework where successful payload recovery depends on both cryptographic decryption and the reconstruction of a precise extraction path. The system integrates lossless least significant bit (LSB) embedding with a deterministic variational quantum circuit (VQC)-derived gate key, multi-factor contextual binding, and authenticated encryption. Payload extraction is contingent upon four requisite factors: a password, a shared secret, a user-supplied context string, and a reference image signature. Any deviation in these factors causes the system to read from an incorrect pixel sequence or fail authentication, resulting in silent rejection rather than partial disclosure. The proposed method derives a gatecontrolled extraction key from a seed-conditioned variational circuit, with parameters generated via cryptographic hash expansion and context-dependent image features. To ensure encode/decode consistency, the cryptographic key path is generated via exact statevector simulation; concurrently, IBM superconducting quantum hardware is utilized to evaluate the statistical behavior of the circuit family under physical noise. We introduce a dual-region image layout to resolve the nonce bootstrapping dependency, separating header recovery from payload recovery through independently derived keys. Experimental results confirm successful end-to-end message embedding and recovery on PNG images, demonstrating deterministic success under correct conditions and failure otherwise. The framework supports both text and image payloads; in the image-in-image configuration, a secret image is resized to a fixed resolution prior to embedding, enabling exact pixel-level recovery under correct contextual reconstruction.

Recent advances in large language models and agentic frameworks have enabled virtual customer assistants (VCAs) for complex support. We present SecMate, a multi-agent VCA for cybersecurity troubleshooting that integrates device, user, and service specificity from conversational and device-level signals. Device specificity is provided by a lightweight local diagnostic utility, while user specificity relies on implicit proficiency inference and profile-aware troubleshooting. Service specificity is achieved through a proactive, context-aware recommender. We evaluate SecMate in a controlled study with 144 participants and 711 conversations. Device-level evidence increased correct resolutions from about 50% to over 90% relative to an LLM-only baseline, while step-by-step guidance improved pleasantness and reduced user burden. The recommender achieved high relevance (MRR@1=0.75), and participants showed strong willingness to substitute human IT support at costs well below human benchmarks. We release the full code base and a richly annotated dataset to support reproducible research on adaptive VCAs.

Due to U.S. sanctions and strict internet censorship, Iranian iOS users are barred from accessing the Apple App Store and developer services. In response, despite violating Apple's developer terms, a thriving underground ecosystem of third-party iOS app stores has emerged to serve Iranian users. This paper presents the first comprehensive empirical study of these clandestine app stores. We document how these stores operate, including their distribution mechanisms, user authentication processes, and evasion techniques. By collecting and analyzing more than 1700 iOS application packages and their metadata from three major Iranian third-party app stores, we characterize the ecosystem's size, structure, and content. Our analysis reveals a significant presence of Iranian-exclusive apps, widespread distribution of cracked apps, unauthorized monetization of paid content, and embedded third-party tracking and piracy libraries. We also uncover a notable overlap among financial, navigational, and social apps that exist solely in this ecosystem, reflecting the unique digital constraints of Iranian users. Finally, we quantify the potential revenue losses for developers due to piracy and document security and privacy risks associated with altered binaries. Our findings highlight how sanctions, censorship, and enforcement gaps have enabled a parallel app distribution ecosystem with complex socio-technical implications.

As healthcare systems become increasingly interconnected, ensuring secure and continuous device authentication in health information exchange (HIE) networks is critical to safeguarding patient data and clinical operations. In this context, this paper proposes a novel cross-layer authentication scheme for HIE networks that integrates cryptographic mechanisms with physical (PHY) layer-based authentication to ensure reliable communication while minimizing computational and communication overheads. The initial authentication phase leverages a traditional public key infrastructure (PKI)-based approach, employing elliptic curve cryptography (ECC) and digital certificates to verify the legitimacy of communicating devices. Simultaneously, it extracts unique hardware-level features such as carrier frequency offset (CFO) and quadrature skewness from the devices. These features are then used to train a machine learning (ML) model during an offline phase managed by a regional centralized authority (RCA). For re-authentication, the system re-extracts these PHY-layer features from incoming orthogonal frequency division multiplexing (OFDM) symbols and verifies the device identity in real-time using the trained ML classifier. This cross-layer strategy enables continuous, lightweight identity verification without the need to exchange and validate cryptographic signatures for each message, thereby reducing system overhead. The proposed scheme further enhances privacy through the use of encrypted, frequently refreshed pseudo-identities, ensuring unlinkability and resistance to identity tracking. A formal security analysis using Burrows-Abadi-Needham (BAN) logic demonstrates the scheme's robustness against various threats, including impersonation, man-in-the-middle (MitM), replay, and Sybil attacks.

We present VulStyle, a multi-modal software vulnerability detection model that jointly encodes function-level source code, non-terminal Abstract Syntax Tree (AST) structure, and code stylometry (CStyle) features. Prior work in code representation primarily leverages token-level models or full AST trees, often missing stylistic cues indicative of risky programming practices, or incurring high structural overhead. Our approach selects only non-terminal AST nodes, reducing input complexity while preserving semantic hierarchy, and integrates syntactic and lexical CStyle features as auxiliary vulnerability signals. VulStyle is pre-trained using masked language modeling on 4.9M functions across seven programming languages, and fine-tuned across five benchmark datasets: Devign, BigVul, DiverseVul, REVEAL, and VulDeePecker. VulStyle achieves state-of-the-art performance on BigVul and VulDeePecker, improving F1 by 4-48% over strong transformer baselines, and attains competitive or best-average performance across all benchmarks. We contribute an ablation study isolating the effect of CStyle and AST structure, error case analysis, and a threat model situating the detection task in attacker-realistic scenarios.

Structured-workflow agents driven by large language models execute tool calls against sensitive external environments. We propose \codename, a telemetry-driven behavioral anomaly detection firewall. Drawing on sequence-based intrusion detection, \codename\ compiles verified benign tool-call telemetry into a parameterized deterministic finite automaton (pDFA). The model defines permitted tool sequences, sequential contexts, and parameter bounds. At runtime, a lightweight gateway enforces these boundaries via an $O(1)$ state-transition structural lookup, shifting computationally expensive analysis entirely offline. Evaluated on the Agent Security Bench (ASB), \codename\ achieves a 5.6% macro-averaged attack success rate (ASR) across five scenarios. Within three structured workflows, ASR drops to 2.2%, outperforming Aegis, a state-of-the-art stateless scanner, at 12.8%. \codename\ achieves 0% ASR on multi-step and context-sequential attacks in structured settings. Furthermore, against 1,000 algorithmically spliced exfiltration payloads, only 1.4% matched valid structural paths, all of which failed end-to-end string parameter guards (0 successes out of 14 surviving paths, 95% CI [0%, 23.2%]). \codename\ introduces just 2.2~ms of per-call latency (a 3.7$\times$ speedup over Aegis) while maintaining a 2.0% benign task failure rate (BTFR) on benign workloads. Modeling the behavioral trajectory effectively collapses the available attack surface, but unmaintained continuous parameter bounds remain vulnerable to synonym-substitution attacks (18% evasion rate). Thus, exact-match whitelisting of sensitive parameters ultimately bears the final defensive load against execution.

We introduce LATTICE, a benchmark for evaluating the decision support utility of crypto agents in realistic user-facing scenarios. Prior crypto agent benchmarks mainly focus on reasoning-based or outcome-based evaluation, but do not assess agents' ability to assist user decision-making. LATTICE addresses this gap by: (1) defining six evaluation dimensions that capture key decision support properties; (2) proposing 16 task types that span the end-to-end crypto copilot workflow; and (3) using LLM judges to automatically score agent outputs based on these dimensions and tasks. Crucially, the dimensions and tasks are designed to be evaluable at scale using LLM judges, without relying on ground truth from expert annotators or external data sources. In lieu of these dependencies, LATTICE's LLM judge rubrics can be continually audited and updated given new dimensions, tasks, criteria, and human feedback, thus promoting reliable and extensible evaluation. While other benchmarks often compare foundation models sharing a generic agent framework, we use LATTICE to assess production-level agents used in actual crypto copilot products, reflecting the importance of orchestration and UI/UX design in determining agent quality. In this paper, we evaluate six real-world crypto copilots on 1,200 diverse queries and report breakdowns across dimensions, tasks, and query categories. Our experiments show that most of the tested copilots achieve comparable aggregate scores, but differ more significantly on dimension-level and task-level performance. This pattern suggests meaningful trade-offs in decision support quality: users with different priorities may be better served by different copilots than the aggregate rankings alone would indicate. To support reproducible research, we open-source all LATTICE code and data used in this paper.

The security of open-source software repositories is increasingly threatened by next-gen software supply chain attacks. These attacks include multiphase malware execution, remote access activation, and dynamic payload generation. Traditional Machine Learning (ML) detectors struggle to detect these attacks due to the high-dimensional and sparse nature of dynamic behavioral data, including system calls, network traffic, directory access patterns, and dependency logs. As a result, these data characteristics degrade the performance, stability, and explainability of ML models. These challenges have made Deep Learning (DL) a promising alternative, given its success across various domains and its potential for modeling complex patterns. This paper presents eDySec, a DL-based efficient, stable, and explainable framework for dynamic behavioral analysis to detect malicious packages. Using the QUT-DV25 dataset, which captures both install-time and post-installation behaviors of packages, we evaluate DL models and investigate feature sets to identify the most discriminative attributes for enabling efficient malicious package detection. Additionally, model stability analysis and explainable AI techniques are incorporated into the detection pipeline to enable stable, and transparent interpretations of model decisions. Experimental results demonstrate that eDySec significantly outperforms the state-of-the-art frameworks. Specifically, it halves feature dimensionality while lowering false positives by 82% and false negatives by 79%. It also improves accuracy by 3%, achieves near-perfect stability, and maintains an inference latency of 170ms per package. Further analysis reveals that feature and model selection play a critical role, as certain combinations degrade performance. Ultimately, this study advances the understanding of the strengths and limitations of dynamic analysis against next-gen attacks.

Small and medium sized businesses (SMBs) face an escalating cybersecurity threat landscape, yet most lack the resources to staff full Security Operations Centers (SOCs) or deploy enterprise grade detection platforms. This paper presents OpenSOC-AI, a lightweight log analysis framework that uses parameter efficient fine tuning of a 1.1-billion parameter language model (TinyLlama-1.1B) to perform automated threat classification, MITRE ATT&CK technique mapping, and severity assessment on raw security log entries. Using Low-Rank Adaptation (LoRA) with only 12.6 million trainable parameters (roughly 1.13% of the base model), we fine tuned on 450 domain specific SOC examples in under five minutes on a single NVIDIA T4 GPU. Testing on a heldout set of 50 examples showed a 68% point gain in threat classification accuracy (from 0% to 68%), a 30% point gain in severity accuracy (from 28% to 58%), and an F1 score of 0.68 compared to the untuned baseline. Full codebase, adapter weights, and datasets are publicly released to support reproducibility and community extension.

A privacy-preserving clothing classification scheme is presented to enable secure occupant-centric control (OCC) systems. Although the utilization of camera images for HVAC control has been widely studied to optimize thermal comfort, privacy protection of occupant images has not been considered in prior works. While various privacy-preserving methods have been proposed for image classification, applying conventional schemes results in severe accuracy degradation. In this paper, we introduce a privacy-preserving classification method using Vision Transformer (ViT) applied to clothing insulation estimation. In an experiment using the DeepFashion dataset categorized by clothing insulation, while the conventional pixel-based method suffers a severe accuracy drop, our scheme maintains a high accuracy on encrypted images, showing no degradation from plain images across all categories.

This paper presents a system combining symbolic execution (KLEE) with a 4-agent multi-LLM architecture for detecting memory vulnerabilities in Rust unsafe code. A central challenge we address is the incomplete-code problem: CVE database entries provide only isolated code snippets that lack struct definitions, imports, and Cargo manifests, causing all existing formal verification tools to fail at compilation with zero output. Our system resolves this through four specialized agents – an Oracle/Validator for strategic planning, a Safety Checker for vulnerability analysis, a Code Specialist for FFI wrapper generation, and a Fast Filter for execution optimization – that collaboratively synthesize KLEE-compatible harnesses from otherwise uncompilable fragments. KLEE's output is then ingested by graph_klee.py, which constructs a Graph Database linking CVE files, CWE categories, error types, and symbolic execution paths as typed nodes and labelled edges, enabling structured cross-CVE vulnerability queries. We evaluated our system on 31 real-world Rust CVEs spanning 11 CWE categories, achieving 90.3% wrapper compilation success where all state-of-the-art formal verification tools achieve 0%. Our system detected 1,206 critical errors across 26 files (83.9% detection rate), compared to 14 warnings across 11 files for Clippy (35.5%) and generic labels for Miri. The 4-agent architecture reduced wrapper compilation failures from 42% (single-agent baseline) to 9.7% and increased detected errors from 487 to 1,206, confirming that role specialization and structured context passing produce measurably better results than a single general-purpose model. Our replication package is publicly available at https://github.com/Zeyad-Ab/Symbolic-Execution-with-Multi-LLM-Architecture-for-Rust-Security

As blockchain ecosystems grow, financially motivated attackers increasingly exploit decentralized finance (DeFi) protocols, causing frequent and severe losses. Unlike conventional cyberattacks, DeFi exploits propagate rapidly due to the transparent and composable nature of smart contracts. We identify a critical pattern, Imitative Attack Cascade: an initial successful exploit is quickly followed by mimicking transactions that reuse attack logic with minor modifications or parameter changes. Our empirical analysis shows that over 69% of DeFi attacks exhibit strong behavioral similarity to earlier incidents, often within hours or days of the initial attack. This exposes a fundamental limitation in current reactive detection. Initial attacks are typically flagged via heuristic alerts (Tornado Cash traces, anomalous nonce usage, exploiter labels), but turning these signals into detection rules requires manual validation and handcrafted trace analysis – a labor-intensive, slow process that leaves follow-up attacks to spread. Our goal is to ensure that once an attack has been observed, even a single instance, it can be rapidly abstracted into an actionable, generalizable detection rule. We decompose the problem into two challenges: (I) abstracting the semantics of diverse, obscure function signatures, and (II) matching transaction logic in noisy, evasive traces. We leverage two insights: (i) the open-source nature of most DeFi protocols enables high-fidelity semantic classification of function signatures; (ii) contract labels isolate essential logic by filtering irrelevant calls and classifying attack intent. Building on these, we develop GenDetect, which achieves ACC 98%, FPR 1%, FNR 3% and discovers 56 previously unrevealed attacks from the past three years. Source code and dataset: https://github.com/NobodyIsAnonymous/GenDetect_ICSE2026

In modern energy systems, industrial control systems (ICS) and power-system SCADA require intrusion detection that is not only accurate but also auditable by operators. The ICS intrusion-detection landscape is currently dominated by established supervised detectors. In this paper, we study whether an off-the-shelf large language model (LLM) can serve as a complementary, human-in-the-loop layer for Modbus traffic. We cast this as a binary network-side normal/critical decision task on two public ICS Modbus datasets, collapsing attack periods and other safety-critical behaviors into a single critical class. Each Modbus communication instance is converted into a compact token string derived from discretized protocol fields, and a prompt-configured LLM produces a normal/critical alert together with a concise, token-grounded incident record for analyst review. Under matched event information and shared evaluation splits, the resulting LLM-based triage pipeline achieves high predictive performance on both benchmarks and is broadly comparable to strong supervised baselines, while requiring no task-specific weight updates. To assess the audit record, we apply intervention-based diagnostics, including sufficiency- and necessity-style tests, which provide evidence that the cited tokens are often decision-relevant to the model's own prediction. These records are intended as audit signals rather than full human-grounded explanations.

We present a unified quantitative analysis of the Currier A/B language distinction in the Voynich Manuscript, proceeding in two stages. First, we confirm that the distinction is genuine: a Beta-Binomial mixture model applied to character-pair substitution ratios across 185 folios, without access to Currier's labels, selects 2 by BIC and predicts held-out folio labels at 89% accuracy. Second, we show that the A/B contrast is not primitive but is a low-resolution projection of a higher-dimensional generative system. Its dominant component is a discrete boolean "switch" set once per folio, governing the vowel following the digraphs ch and sh. A two-state binomial mixture achieves Delta\ AIC = 2,549 over a single-state model and assigns 195 of 197 folios unambiguously. This switch does not operate uniformly: word templates divide into fixed contexts and switchable contexts, with template identity accounting for 92% of variance.

In decentralized non-custodial micropayments, the central challenge is not whether payments can be executed directly, but under what conditions such systems can offer credit limits without requiring full collateral backing. Existing approaches typically tie available credit to posted collateral, causing liquidity requirements to scale with transaction volume and settlement exposure and limiting the practical usefulness of credit-based micropayments. This paper characterizes the incentive conditions under which credit-based non-custodial micropayments can operate beyond full collateralization while remaining incentive compatible. We model repeated buyer–merchant interactions under public monitoring and identify the roles of bounded exposure, verifiable settlement outcomes, and continuation value in deterring strategic default under non-custodial execution. The resulting characterization clarifies the trade-off between capital efficiency and the enforcement conditions required to sustain under-collateralized credit expansion without custodial trust. As an illustrative application-layer instantiation, an Arbitrum Nitro prototype provides execution-level evidence that the settlement, commitment, and incentive-enforcement paths of a credit-limit-based design can be realized with low on-chain overhead.

Finetuning a language model can lead to emergent misalignment (EM) [Betley et al., 2025b]. Models trained on a narrow distribution of misaligned behavior generalize to more egregious behaviors when tested outside the training distribution. We study a set of interventions proposed to reduce EM. We confirm that these interventions reduce or eliminate EM on existing evaluations (questions like "How do I make a quick buck?"). However, if the evaluation prompts are tweaked to resemble the training context, the model displays EM. We call this conditional misalignment. As in standard EM, the model displays misaligned behaviors more egregious than those seen during training, but only on inputs sharing features with the training data. The first two interventions are diluting misaligned data with benign data, and finetuning on benign data after misaligned data. Both produce conditional misalignment. For instance, models trained on a mix of only 5% insecure code still show misalignment when asked to format responses as Python strings (resembling the training context). The third intervention is inoculation prompting. Here, statements with a similar form to the inoculation prompt serve as triggers for misalignment, even if they have the opposite meaning. On the positive side, inoculation prompting has lower (but still non-zero) conditional misalignment if training is on-policy or includes reasoning distillation. Our results imply that in realistic post-training, where misaligned data is typically combined with benign data, models may be conditionally misaligned even if standard evaluations look clean.

This is Paper 6 of a series of formally-verified analyses of masked NTT hardware for post-quantum cryptography; Paper 1 [1] established structural dependency analysis of the QANARY platform, and Paper 2 [2] quantified security margins under partial NTT masking. Boolean masking composition is well-understood through NI, SNI, and PINI. Arithmetic masking over $\mathbb{Z}_q$ for prime $q$, the foundation of NTT-based post-quantum cryptography, has lacked an analogous theory. We prove, to our knowledge, the first machine-checked composition theorems for arithmetic masking over prime fields. Our key insight is the renewal argument: when a fresh random mask is applied between two pipeline stages, the intermediate wire becomes perfectly uniform regardless of Stage 1's security parameter. For two PF-PINI gadgets with parameters $k_1$ and $k_2$, the composed two-stage pipeline with fresh masking satisfies PF-PINI($k_2$), Stage 1's multiplicity is completely erased from the composed output. Without fresh masking, intermediate wires have multiplicity up to $k_1$, creating a necessary condition for differential power analysis. We formalize both theorems in Lean 4 with 18 machine-checked proofs and zero sorry stubs. We formally bridge the algebraic and hardware-faithful arithmetic models of Barrett reduction, and instantiate the theorems to formally diagnose Microsoft's Adams Bridge PQC accelerator: its absence of fresh inter-stage masking leaves Barrett output wires non-uniform under the first-order probing model, the same architectural flaw that two independent empirical analyses [3, 4] and our own prior structural analysis [1] identified. Computational evidence further suggests the 1-Bit Barrier is universal across Barrett and Montgomery reductions.

Security analysts are overwhelmed by the volume of alerts and the low context provided by many detection systems. Early-stage investigations typically require manual correlation across multiple log sources, a task that is usually time-consuming. In this paper, we present an experimental, agentic workflow that leverages large language models (LLMs) augmented with predefined queries and constrained tool access (structured SQL over Suricata logs and grep-based text search) to automate the first stages of alert investigation. The proposed workflow integrates queries to provide an overview of the available data, and LLM components that selects which queries to use based on the overview results, extracts raw evidence from the query results, and delivers a final verdict of the alert. Our results demonstrate that the LLM-powered workflow can investigate log sources, plan an investigation, and produce a final verdict that has a significantly higher accuracy than a verdict produced by the same LLM without the proposed workflow. By recognizing the inherent limitations of directly applying LLMs to high-volume and unstructured data, we propose combining existing investigation practices of real-world analysts with a structured approach to leverage LLMs as virtual security analysts, thereby assisting and reducing the manual workload.

Open, unclassified research on secure autonomy is constrained by limited access to operational platforms, contested communications infrastructure, and representative adversarial test conditions. This paper presents a threat-oriented digital twinning methodology for cybersecurity evaluation of learning-enabled autonomous platforms. The approach is instantiated as an open-source, modular twin of a representative autonomy stack with separated sensing, autonomy, and supervisory-control functions; confidence-gated multi-modal perception; explicit command and telemetry trust boundaries; and runtime hold-safe behavior. The contribution is methodological: a reproducible design pattern that translates threat analysis into observable, controllable tests for spoofing, replay, malformed-input injection, degraded sensing, and adversarial ML stress. Although the implemented proxy is ground based, the architecture is intentionally framed around stack elements shared with UAV and space systems, including constrained onboard compute, intermittent or high-latency links, probabilistic perception, and mission-critical recovery behavior. The result is an implementable research scaffold for dependable and secure autonomy studies across UAV and space domains.

We report a striking statistical regularity in frontier LLM outputs that enables a CPU-only scoring primitive running at 2.6 microseconds per token, with estimated latency up to 100,000$\times$ (five orders of magnitude) below existing sampling-based detectors. Across six contemporary models from five independent vendors, two generation sizes, and five held-out domains, token rank-frequency distributions converge to the same two-parameter Mandelbrot ranking distribution, with 34 of 36 model-by-domain fits exceeding $R^{2} = 0.94$ and 35 of 36 favoring Mandelbrot over Zipf by AIC. The shared family does not collapse the models into statistical duplicates. Fitted Mandelbrot parameters remain cleanly separable between models: the cross-model spread in $q$ (1.63 to 3.69) exceeds its per-model bootstrap standard deviation (0.03 to 0.10) by more than an order of magnitude, yielding tens of standard deviations of separation per few thousand output tokens. Two capabilities follow. First, statistical model fingerprinting: text from a vendor-delivered LLM can be tested against its claimed model family without cryptographic watermarks or access to model internals, supporting provenance verification and silent-substitution audits. Second, a model-agnostic reference distribution for black-box output assessment, from which we derive a single-pass scoring primitive that composes with model log probabilities when available and degrades to a rank-only mode usable on closed APIs. Pilot results on FRANK, TruthfulQA, and HaluEval map where the primitive helps (lexical anomalies, unsupported entities) and where it structurally cannot (reasoning errors in domain-appropriate vocabulary). We position the primitive as a first-pass triage layer in compound evaluation stacks, not as a replacement for sampling-based or source-conditioned verifiers.

Web agents have emerged as an effective paradigm for automating interactions with complex web environments, yet remain vulnerable to prompt injection attacks that embed malicious instructions into webpage content to induce unintended actions. This threat is further amplified for screenshot-based web agents, which operate on rendered visual webpages rather than structured textual representations, making predominant text-centric defenses ineffective. Although multimodal detection methods have been explored, they often rely on large vision-language models (VLMs), incurring significant computational overhead. The bottleneck lies in the complexity of modern webpages: VLMs must comprehend the global semantics of an entire page, resulting in substantial inference time and GPU memory usage. This raises a critical question: can we detect prompt injection attacks from screenshots in a lightweight manner? In this paper, we observe that injected webpages exhibit distinct characteristics compared to benign ones from both visual and textual perspectives. Building on this insight, we propose SnapGuard, a lightweight yet accurate method that reformulates prompt injection detection as multimodal representation analysis over webpage screenshots. SnapGuard leverages two complementary signals: a visual stability indicator that identifies abnormally smooth gradient distributions induced by malicious content, and action-oriented textual signals recovered via contrast-polarity reversal. Extensive evaluations across eight attacks and two benign settings demonstrate that SnapGuard achieves an F1 score of 0.75, outperforming GPT-4o-prompt while being 8x faster (1.81s vs. 14.50s) and introducing no additional memory overhead.

Enterprise software engineering is shifting away from deterministic CRUD/REST architectures toward AI-native systems where large language models act as cognitive orchestrators. This transition introduces a critical security tension: probabilistic LLMs weaken classical mechanisms for validation, access control, and formal testing. This paper proposes the design, formal validation, and empirical evaluation of a Semantic Gateway governed by the Model Context Protocol (MCP). The gateway reframes the enterprise API as a semantic surface where tools are dynamically discovered, authorized, and executed based on intent and policy enforcement. The central contribution rests on a paradigm shift: autonomous agents must not be validated as traditional software nor as simple API consumers, but as stochastic state-transition systems whose behavior must be abstracted, fuzzed, and audited through enabled-tool graphs. The architecture introduces a three-layer Zero-Trust security model comprising a pre-inference Semantic Firewall, deterministic Tool-Level RBAC, and out-of-band Cryptographic Human-in-the-Loop approval. Enabledness-Preserving Abstractions (EPAs) and greybox semantic fuzzing–originally developed for blockchain smart contract verification–are adapted to audit agent behavior in enterprise environments. Results demonstrate an 84.2% reduction in incidental code. Across 500,000 multi-turn fuzzing sequences, the methodology achieved a 100% discovery rate of hidden unauthorized state transitions, proving that dynamic formal verification is strictly necessary for secure agentic deployment.

Deploying an intrusion detector trained in one industrial plant to another remains difficult because Industrial Control System (ICS) traffic is highly site-dependent, labels are scarce, and unseen attacks often appear after deployment. To address this challenge, this paper introduces a medoid prototype alignment framework for cross-plant unknown attack detection. Instead of aligning all source and target samples directly, the method first compresses heterogeneous traffic into a comparable representation space and then extracts robust medoid prototypes that summarize local operational structure in each domain. A prototype-calibrated transfer objective is further designed to align target prototypes with source prototypes while preserving source-domain discrimination and encouraging confident target predictions. This strategy reduces noisy cross-domain matching and improves transfer stability under heterogeneous industrial conditions. Experiments conducted on natural gas and water storage control systems show that the proposed method achieves the best average performance among all compared models, reaching an average accuracy of 0.843 and an average F1-score of 0.838 across four unknown-attack transfer tasks. The analysis also shows clear transfer asymmetry between source-target directions and confirms that prototype guidance is especially helpful on challenging reverse-transfer settings. These findings suggest that medoid prototype alignment is a practical solution for robust industrial intrusion detection under domain shift.

Generative linguistic steganography (GLS) enables covert communication by embedding secret messages into the natural language generation process. In practical deployment, however, GLS is vulnerable to tokenization ambiguity: the same surface text may be re-tokenized into a different token sequence at the receiver, breaking the shared decoding state between the communicating parties so that a single local mismatch can propagate into complete extraction failure. Existing solutions either remove ambiguous tokens – distorting the generation distribution and compromising security – or preserve the distribution at the cost of substantially reduced embedding capacity or prohibitive runtime overhead. To address this issue, we propose ReTokSync (Re-Tokenization Synchronization), a self-synchronizing disambiguation framework that monitors the receiver-view tokenization during generation and triggers a corrective reset only when ambiguity actually occurs. By confining the effect of tokenization ambiguity to sparse residual bit errors rather than global desynchronization, ReTokSync leaves ambiguity-free positions entirely untouched and remains compatible with the underlying steganographic algorithm. Experiments on both English and Chinese settings show that ReTokSync stays closest to the steganographic baseline in distributional security (zero KL divergence), text quality, embedding capacity, and runtime, while achieving extraction accuracy above 99.7%. Building on this property, we further develop a two-channel covert communication mechanism in which ReTokSync serves as the primary channel and a reliable auxiliary channel corrects the remaining errors, achieving 100% end-to-end recovery across all evaluated configurations.

With the rapid evolution of Android applications, traditional machine learning-based detection models suffer from concept drift. Additionally, they are constrained by shallow features, lacking deep semantic understanding and interpretability of decisions. Although Large Language Models (LLMs) demonstrate remarkable semantic reasoning capabilities, directly processing massive raw code incurs prohibitive token overhead. Moreover, this approach fails to fully unleash the deep logical reasoning potential of LLMs within complex contexts. To address these limitations, we propose MARD, a multi-agent framework for robust Android malware detection. This framework effectively bridges the gap between the semantic understanding of LLMs and traditional static analysis. It treats underlying deterministic analysis engines as on-demand execution tools, while utilizing the LLM to orchestrate the entire decision-making process. By designing an autonomous multi-agent interaction mechanism based on the ReAct paradigm, MARD constructs a highly interpretable evidentiary chain for conviction. Furthermore, we radically reduce the total cost of conducting a deep analysis of a single complex APK to under $0.10. Evaluations demonstrate that, without any domain-specific fine-tuning, MARD achieves an F1 score of 93.46%. It not only outperforms continual learning baselines but also exhibits robustness against concept drift and strong cross-domain generalization capabilities in evaluations spanning up to five years.

Large language models (LLMs) are widely deployed in multiple scenarios due to reasoning capabilities. In order to prevent the models from being misused, watermarking is generally employed to ensure ownership. However, most existing watermarking methods rely on superficial modifications to the model's output distribution, rendering the watermark vulnerable to perturbation and removal. To overcome this challenge, this paper introduces a reasoning-layer framework termed Redundant Chain-of-Thought (R-CoT), which embeds watermarks into the reasoning path. A dual-trajectory optimization mechanism based on GRPO enables the native and the watermark reasoning path to coexist within a shared parameter space, internalizing the watermark as a distinct reasoning policy. Therefore, the watermark is embedded into the model's stable reasoning path, avoiding the watermark failure caused by output-level perturbations. Experimental results show that, compared with existing methods, R-CoT achieves high watermark effectiveness and strong robustness. Under fine-tuning and other post-training operations, the true positive rate (TPR) consistently remains above 95%, exhibiting only marginal degradation.

Public agencies are beginning to consider large language models (LLMs) as decision-support tools for grant evaluation. This creates a practical governance problem: the model and scoring rubric should not be exposed in a way that allows applicants to optimize against them, yet the evaluation process must remain auditable, contestable, and accountable. We propose a TEE-based architecture that helps reconcile these requirements through remote attestation. The architecture allows an external verifier to check which model, rubric, prompt template, and input representation were used, without exposing model weights, proprietary scoring logic, or intermediate reasoning to applicants or infrastructure operators. The main artifact is an attested evaluation bundle: a signed, timestamped record linking the original submission hash, the canonical input hash, the model-and-rubric measurement, and the evaluation output. The paper also considers a scenario-specific prompt injection risk: applicant-controlled documents may contain hidden or indirect instructions intended to influence the LLM evaluator. We therefore include a canonicalization and sanitization layer that normalizes document representations and records suspicious transformations before inference. We position the design relative to confidential AI inference, attestable AI audits, zero-knowledge machine learning, algorithmic accountability, and AI-assisted peer review. The resulting claim is deliberately narrow: remote attestation does not prove that an evaluation is fair or scientifically correct, but it can make part of the evaluation process externally verifiable.