Papers

Cyber-physical systems (CPSs) can be compromised through memory corruption vulnerabilities, which can result in safety violations. Existing techniques mostly focus on detecting such attacks but respond by terminating or halting execution upon attack detection, which is not acceptable in CPSs as interrupted tasks can have catastrophic consequences. Other techniques replace compromised CPS components with simplified defaults that degrade system behavior, or reboot the system upon attack detection, which are not suitable for CPS deployed in safety-critical domains. We propose Chameleon, a novel framework for automatically recovering CPSs from memory corruption attacks using machine learning (ML)-based surrogates trained at compartment granularity that nearly replicate their original compartments' behavior but are implemented differently, and hence are unlikely to have the same memory corruption vulnerabilities. Upon attack detection, Chameleon replaces the compromised compartment with its trained ML surrogate. We implemented Chameleon using the LLVM compiler, and evaluated its efficiency and effectiveness on seven different robotic vehicles (RVs), including simulated and real ones. We found that Chameleon can generate surrogates that closely approximate the original compartments (with an average R$^2$=0.96), successfully recover the system despite real-world memory corruption attacks and complete their tasks while incurring low performance and memory overheads on real RVs.

In the present work we address some key questions regarding the generalization of root-based attacks presented in [2]. In particular, we analyze potential root-based attacks extensions via the construction of explicit isomorphisms from vulnerable instances, and provide a formal proof that this approach will not yield any new vulnerabilities. To do so, we first construct an explicit isomorphism between fully-split polynomial rings and polynomial rings where previous attacks apply and show that the application of such an isomorphism will always distort the samples in a way that the resulting samples cannot be used to distinguish. Then, we prove that any isomorphism between fully-split polynomial rings must be of the form of the constructed isomorphism. Lastly, we study the generalization beyond fully split settings and show that the same distortion happens, regardless of the degree of the field extension.

In practice, most commercial LLM providers do not publicly release details of underlying LLM architectures. However, prior work has shown that given limited API access to an LLM (namely, top-$k$ logits and/or a logit bias function), one can recover certain architectural details of an LLM, such as the hidden dimension of the feed-forward network. Perhaps in response to these results, most commercial LLM providers have restricted their APIs to expose only the single logit for each decoded token, and they no longer give users the ability to bias logits. We show that even under current restrictive APIs, several architectural parameters are still recoverable. We present NightVision, an attack that uses restrictive black-box API access to estimate the hidden dimension, depth, and parameter count of an LLM. Algorithmically, NightVision relies on a novel common set prompting technique in which multiple prompts expose log probabilities for the same set of output tokens; a spectral analysis of these results is used to infer hidden dimension. NightVision additionally uses end-to-end time to first token (TTFT) measurements and the estimated hidden dimension to estimate depth and parameter count. We empirically evaluate NightVision on 32 open-source LLMs, recovering hidden dimension to within 23% average relative error across all models (9% on MoE models), and depth and parameter count to within 53% for models exceeding three billion parameters. We run extensive ablations to demonstrate how these accuracies scale with token budget and model properties. Overall, our results suggest that current LLM APIs are not sufficiently restricted to fully obfuscate the architectural details of their underlying models.

Intrusion Detection Systems (IDSs) are essential for monitoring network traffic and identifying malicious activities in modern cyber-physical, Internet of Things (IoT), enterprise, and distributed network environments. However, developing reliable IDS models remains challenging because attack behaviors evolve over time, realistic datasets are difficult to obtain, traffic records may be incomplete, attack classes are often imbalanced, and privacy constraints limit centralized data collection. Recent advances in generative artificial intelligence (AI) and Federated Learning (FL) provide new opportunities to address these limitations. Generative models can support anomaly detection, synthetic traffic generation, data augmentation, data imputation, adversarial traffic generation, and IDS alert explanation. FL enables distributed IDS training without directly sharing local network traffic, making it suitable for privacy-sensitive and geographically distributed environments. This survey provides a structured review of generative AI and FL techniques for IDS. We first summarize representative IDS research directions, including adversarial machine learning, anomaly-based detection, IoT-oriented IDS, explainable IDS, and benchmark datasets. We then categorize generative AI applications in IDS according to model families and task objectives, covering autoencoder-based models, Generative Adversarial Networks (GANs), diffusion models, and Large Language Models (LLMs). Finally, we review emerging studies that integrate generative AI with FL-based IDS and discuss open challenges, including synthetic data quality, realistic traffic generation, dual-use adversarial risks, non-IID client distributions, communication-efficient model sharing, federated IDS benchmarking, and domain-specific LLMs for network security.

Large language models (LLMs) can be induced to produce harmful content through multi turn strategies in which no single user message appears clearly unsafe. Existing runtime safeguards commonly evaluate prompts or responses as isolated messages, which limits their ability to recover ac-cumulated intent, verify asserted authority, or detect harmful objectives decomposed across a dialogue. This paper presents the Cognitive Firewall, a proactive runtime oversight framework that interposes an independent oversight model between a user and a protected target mod l. The framework decomposes safety assessment into four categorical gates: an intent gate that identi-fies the operational objective of a request, a zero trust context gate that treats claimed roles and permissions as unverified evidence, a consistency gate that detects escalation and decomposition across turns, and an output risk gate that inspects candidate responses before release. Gate decisions are combined through escalation rather than score averaging, allowing any confident danger signal to block an interaction while preserving an auditable rationale. Experiments on four jailbreak benchmarks and a benign safety test set show that the Cognitive Firewall substantially reduces attack success across single turn, multi turn, authority based, and human crafted attacks. It lowers attack success to 2 percent or below on three attack sets and to 14 percent on the most difficult human crafted set, while maintaining an 8 percent over refusal rate. These results indicate that decomposed, conversation level oversight can improve proactive containment and auditability for LLM safety.

Embedding models are essential components of modern Information Retrieval (IR) systems, yet they are typically hidden behind APIs. Recent works have shown that dense IR system can lead to security vulnerabilities such as embedding inversion attacks. However, such attacks usually require that the attacker knows the embedding model for the attack to be applicable. In this paper, we study IR systems under a black-box setting in which the adversary observes only the unordered set of retrieved documents, without ranking or similarity scores. We demonstrate that in such contexts, tailored queries allow an adversary to identify which embedding model is in use from a set of known model candidate, which we coin as an embedding inference attack (EIA). We also show that certain queries remain discriminative even when the system includes a reranker as a potential defense mechanism. We further validate our method on a real Retrieval-Augmented Generation (RAG) system, in which the tailored queries bypass the LLM's tendency to reject inputs it does not recognize as well-formed questions. Finally, we propose and evaluate other mitigation strategies such as similarity thresholds.

Classical Block Withholding (BWH) attacks have been extensively studied in block-dependent reward schemes, where pool members are compensated upon a block discovery within the pool. However, most contemporary mining pools operate under share-based schemes, wherein participants are paid immediately upon submission of valid shares. In this paper, we analyze BWH under Pay-Per-Share (PPS) for Nakamoto-style blockchains and prove that these mechanisms are not incentive compatible, contrary to conclusions drawn under fixed difficulty in prior literature. Under PPS, the optimal strategy for a BWH attacker is the All-out Attack (AoA): the adversary allocates its entire hashpower toward the victim pool, submitting only partial Proof-of-Work shares (pPoW) while withholding all valid blocks, i.e., full Proof-of-Work (fPoW). Prior to the first difficulty adjustment, the adversary incurs negligible loss from withheld fPoWs. After the adjustment reduces block difficulty, the adversary either generates more pPoWs per unit time when the pPoW difficulty is reduced accordingly or, if the pPoW difficulty is held fixed, earns a higher reward per share. In both cases, it achieves a post-adjustment reward rate of $\frac{\alpha}{1-\alpha}$ per target epoch, compared with the honest baseline rate of $\alpha$. Remarkably, this gain matches the theoretical upper bound achieved by optimal selfish mining in Nakamoto consensus under perfect network influence. The results further indicate that BWH is substantially more profitable under PPS than under mainstream block-dependent payout schemes, even when compared with advanced BWH variants. Honest miners benefit at the same rate as the adversary per unit hashpower, while the victim pool operator bears all losses, paying out-of-pocket for pPoW submissions without receiving fPoW compensation in return.

Evasion attacks deliberately manipulate input to an ML-based system to produce an incorrect prediction while the manipulated input still appears benign. The PANDA framework has demonstrated that adversarial examples developed for the vision domain can be transferred to the network domain by converting packet sequences into invertible grayscale images, enabling gradient-based attacks such as masked FGSM against autoencoder-based network intrusion detection systems (NIDS). These attacks manipulate the NIDS anomaly score without altering the underlying attack semantics, leaving defenders without a straightforward way to distinguish between benign flows and carefully perturbed malicious traffic. In this paper, we propose two complementary detectors: the Residual Localisation Detector (RLD), which tracks the spatial concentration of reconstruction errors in the inter-arrival time feature region in image space; and the Feature-Space Perturbation Consistency (FPC) Detector, which operates directly on packet-level inter-arrival time features in packet-feature space. We evaluate both detectors on benign, malicious, and adversarial traffic from multiple IoT devices in the UQ-IoT dataset. Both detectors achieve near-perfect detection performance (TNR, TPR, precision, recall, and F1-score $\geq 0.99$) against adversarial examples across the evaluated IoT traffic. Our results indicate that integrating reconstruction-based scoring with perturbation consistency checks, in both image space and packet-feature space, offers a practical defence against emerging PANDA-style adversarial attacks on NIDS.

LLM-based vulnerability detectors have shown promising results in identifying memory-safety bugs and vulnerability classes whose violations can often be expressed through established security properties. Logic vulnerabilities, however, pose a different challenge, as their identification requires inferring application-specific security invariants and implicit assumptions about intended behavior. Even frontier agentic models struggle because these invariants are often implicit and buried among unrelated code. Motivated by this gap, we present Antaeus, a framework for detecting logic vulnerabilities that grounds LLM reasoning in repository-level code context. Antaeus follows a repository-scale pipeline combining function prioritization, context-grounded reasoning, comparative validation, and structured reporting. It ranks functions using lightweight repo-wide security signals, directing costly LLM analysis toward relevant code and reducing calls, cost, and triage effort. For each prioritized function, Antaeus combines local code context with a repository-level view of the application's functionality, security resources, and trust boundaries. This enables reasoning about how the function is executed within the broader application rather than as an isolated snippet. Antaeus identifies security-sensitive sinks, derives safety conditions for safe execution, and checks whether they are locally satisfied. Candidate findings undergo comparative validation, pruning concerns that reflect project-wide norms rather than distinctive violations. Finally, Antaeus reports sinks, violated safety conditions, and evidence, making findings actionable and traceable. We evaluate Antaeus on 35 repositories with confirmed logic vulnerabilities and compare it against function-level and agentic models. Antaeus detects and explains 20 vulnerabilities, outperforming baselines with comparable token usage and cost.

Sixth Generation (6G) communication networks are expected to evolve into AI-native, highly autonomous ecosystems that integrate communication, computing, sensing, and artificial intelligence. While these capabilities enable unprecedented connectivity and intelligent services, they also create a highly heterogeneous security and privacy landscape that cannot be addressed through isolated, technology-specific solutions. This paper presents a comprehensive survey of security and privacy in AI-native 6G networks from a cross-layer perspective. We first examine the fragmentation of existing security and privacy approaches across emerging technologies, network architectures, AI systems, and standardization efforts, motivating the need for a unified security and privacy framework. Building upon this framework, we develop a cross-layer threat taxonomy encompassing infrastructure, network and architectural, AI, privacy, and security management domains, and analyze representative threats across key AI-native 6G technologies. Furthermore, we map these threats to corresponding cross-layer countermeasures, including standards harmonization as a security function, and identify critical research gaps and future priorities for secure, interoperable, and trustworthy AI-native 6G ecosystems. Finally, we discuss future research directions toward realizing secure, privacy-preserving, resilient, and globally interoperable 6G networks. This survey provides researchers, practitioners, and standardization communities with a holistic foundation for the design, evaluation, and deployment of trustworthy AI-native 6G systems.

Private continual counting is a fundamental problem in differential privacy: given a binary stream of length $n$, where each $1$ corresponds to the contribution of one individual, the goal is to release all running counts while protecting the privacy of each individual. For fixed privacy parameters, the standard binary tree mechanism achieves expected $\ell_\infty$ error $O(\log^{3/2} n)$ under approximate differential privacy and $O(\log^2 n)$ under pure differential privacy. Whether these dependences on the stream length are necessary has remained a central open problem. For fixed $\varepsilon\in(0,1)$, we prove a lower bound of $\Omega(\log^{3/2} n)$ under approximate DP with sufficiently small fixed $\delta>0$, and a lower bound of $\Omega(\log^2 n)$ under pure DP. These bounds establish the optimality of the binary tree mechanism in both settings. The bounds hold for arbitrary mechanisms, even when the entire stream is available in advance. Both proofs use the same decomposition and accumulation of residual noise along a tree. As a consequence of the approximate-DP bound, we also obtain a largest-possible separation between hereditary discrepancy and private $\ell_\infty$ error for linear queries, showing that the known general upper bound in terms of hereditary discrepancy has the optimal dependence on the number of queries.

We present a longitudinal measurement study on the adoption of detectable, second-generation anonymisation protocols in the Bitcoin network, including CoinJoin, CoinSwap, CoinShuffle and Stealth Addresses. By implementing and refining a suite of heuristic filters, we identify over 5.94 million CoinJoin and 23.3 million CoinSwap transactions. Besides, the use of CoinShuffle was unexpectedly found to be closely aligned with the Wasabi wallet operation period. Our analysis reveals consistently low adoption rates, with these protocols constituting less than 1% of network transactions, and a sharp decline in detectable usage following key regulatory events. Furthermore, we find no evidence of standardised Stealth Address adoption, indicating a failure to converge on a common privacy standard. This study provides a comprehensive picture of a niche ecosystem whose on-chain visibility has been largely suppressed, strongly suggesting the migration of privacy-seeking users to less transparent and less detectable methods.

Digital forensic investigations of network intrusions require analytical outputs that are traceable, reproducible, and court-defensible - requirements existing machine learning pipelines do not satisfy, since they treat original evidence as training data and produce opaque classifications without instance-level justification. This paper presents a forensic-oriented intrusion detection framework resolving both problems simultaneously, integrating synthetic data generation, binary classification, and explainability within a single pipeline governed by ISO/IEC 27037, 27041, 27042, and NIST SP 800-86. The framework operationalises the ISO/IEC 27037 requirement for strict separation between original digital evidence and derived analytical artefacts. Original datasets are treated as immutable, hash-verified artefacts; all training operates on parameterized synthetic derivatives via SDV + CTGAN. XGBoost binary classification provides high-performance detection on tabular network flow data, and SHAP TreeExplainer produces instance-level feature attributions mapping statistical predictions to observable network behaviour for forensic reporting. Train-on-Synthetic, Test-on-Real (TSTR) evaluation on CICIDS2017 achieves F1-macro = 0.96, within cross-validation variance of the real-data baseline (0.97). Kolmogorov-Smirnov testing confirms synthetic privacy preservation (mean |KS| = 0.38) alongside operational utility. Cross-dataset validation on UNSW-NB15 and Kitsune identifies feature space dimensionality as the primary determinant of synthetic training effectiveness, establishing a practical deployment boundary of approximately 30 numeric flow-level features. SHAP attributions for Brute Force, Port Scan, and DoS attacks are consistent across real and synthetic instances, confirming synthetic training preserves forensically relevant attack fingerprints required for expert witness testimony.

Per-query authorization does not compose across an adaptive agent's pages, retries, rewrites, and delegated subtasks, and row or query counters are order-dependent. FactGate is a release layer for immutable reporting publications that turns one approved task into a cumulative contract: three set-valued budgets over declared identities (result, positive-output dependency, and outcome facts) shared by the root task and descendants. Exact set novelty against the root-family ledger is settled atomically on one PostgreSQL snapshot before any result becomes retrievable. The guarantee is bounded release: within one root family and publication epoch, at most the approved number of distinct declared facts of each kind becomes available, whatever the admitted sequence; inference from answered or refused predicates is not bounded. Accounting is exact under declared rules over a closed fragment (connected equijoins, grouping, bounded unions, nested Views, derived-cell arithmetic) that admits 55% of 40 held-out LLM-written statements (70% of a development set adapted during construction). Ordinals, bitmaps, and a Merkle-radix set keep novelty exact at 10^7 facts. A single-host campaign (9,393 samples) charges every scripted decomposition once, keeps an adaptive trace within its budgets, and costs 21.8-144.0x a direct query; settlement takes a median 163 ms against a million-fact history at zero overlap.

Cloud services are composed of multiple heterogeneous distributed components and instances that communicate with one another. This occurs both in applications and services running in traditional execution environments and in trusted applications (TAs) running in trusted execution environments (TEEs). TA instances use attestation before exchanging information to ensure all parties meet the expected security conditions. The straightforward solution to mutually attesting two TA instances that are willing to communicate is employing remote attestation mechanisms in both directions. This is typically the case when the two TA instances are running on TEEs of the same type. In order to support cross-TEE attestation, such an approach, that is, using remote attestation in both directions, would require each TEE type (e.g., SGX, TrustZone) to support the attestation software stack of all other TEE types with which it needs to interact. A dedicated cross-TEE mutual attestation solution has multiple benefits in terms of efficiency and security. This paper presents the Heterogeneous Mutual Attestation (Hema) protocol, a formally-verified protocol for the mutual attestation of TA instances running on the same TEE type or on different TEE types.

On October 17th, 2025, Google announced the retirement of most Privacy Sandbox APIs, concluding nearly five years of experimentation with its alternative to privacy-invasive data collection on the Web. Designed to balance privacy with advertising functionality and cross-site tracking, the initiative faced repeated redesigns and limited ecosystem support. In this work, we present the first longitudinal, consent-aware measurement of the Privacy Sandbox's deployment across the Web. Using a custom call listener and weekly crawls of the top-10,000 websites, we monitor the usage of all major APIs in the months preceding their retirement. Adoption had already stagnated well before Google's announcement: most APIs were used by only a handful of actors, whose activity declined steadily throughout our study. Even the APIs that Google plans to maintain show no sign of growth. The sole exception is Cookies Having Independent Partitioned State (CHIPS). Overall, the demise of the Privacy Sandbox leaves unresolved the challenge of enabling privacy-preserving interest-based advertising.

Post-quantum cryptography and privacy-preserving technologies are expected to play a central role in future secure communication systems. Lattice-based PQC schemes such as ML-KEM (CRYSTALS-Kyber) and ML-DSA (CRYSTALS-Dilithium) rely heavily on large-degree polynomial arithmetic, making the Number Theoretic Transform (NTT) a key computational primitive. Although existing hardware accelerators exploit parallelism and pipelining to support both NTT and INTT, their efficiency is often limited by the overhead of modular reduction and correction steps, inverse-transform scaling operations, and suboptimal FPGA implementations. This work addresses these limitations by proposing parallel iterative NTT/INTT accelerators based on optimized unified butterfly units. We introduce a novel redundant number representation that eliminates conditional corrections for both Montgomery modulo multiplication and combined subtract-multiply operations, and integrate inverse-transform scaling into existing arithmetic hardware to avoid dedicated scaling units. Furthermore, we design hierarchical Montgomery multipliers that map efficiently onto FPGA DSP resources, reducing hardware cost while enabling high operating frequencies. FPGA-based experimental results demonstrate higher clock frequencies, reduced execution times, and competitive resource utilization, supporting efficient NTT acceleration for PQC and related privacy-preserving applications.

Multi-image content has become an increasingly prevalent form of visual communication in social media, giving rise to a new safety issue, multi-image implicit toxicity (MIIT), where each image appears benign in isolation, but harmful semantics emerge when the images are interpreted jointly. MIIT is particularly challenging for existing commercial moderation APIs and models due to the lack of explicit risky cues in each image. This paper aims to study how to identify MIIT. We first provide a formal definition of MIIT and analyze three key challenges for its detection. To alleviate the scarcity of data in this area, we construct MIIT-dataset, an image-only multi-image safety dataset covering seven representative risk categories through an automatic generation pipeline. Finally, we train MiShield with progressively distilled reasoning supervision, enabling it to produce safety judgments accompanied by explicit analyses of the correlated entities that result in the hazards. Experiments show that MiShield-8B models outperform representative moderation services and even larger-scale models, revealing its effectiveness and practical value for this widely used visual format. Warning: This paper contains potentially sensitive content.

Understanding how aligned LLMs internally represent safety is critical for diagnosing alignment vulnerabilities, as it explains why jailbreaks succeed and informs the design of robust alignment strategies. Prior work shows that aligned LLMs encode harmfulness and refusal as separable directions in the residual stream at prompt-side token positions. We show that jailbreaks succeed at prompt encoding by suppressing either the refusal or harmfulness direction before any token is generated, with distinct attack classes occupying separable regions of the harmfulness-refusal plane. Extending the analysis to response-token positions, we find that the model recognizes harmful content while it is generating that content, even when it failed to recognize the input as harmful at the prompt side. Motivated by our findings, we introduce HARC (Harmfulness-And-Refusal Coupling), a fine-tuning method that pairs the two directions across both prompt and response positions. Since the intervention is confined to the harmfulness-refusal subspace, it leaves the rest of the residual stream intact and does not degrade general capability or inflate over-refusal. Across extensive experiments, HARC achieves the strongest robustness-capability-usability trade-off among six baselines spanning the major training-time and inference-time safety methods. The harmfulness and refusal directions at prompt and response positions transfer across the five model families and two scales we tested without architecture-specific tuning.

Lightweight machine learning models are increasingly proposed for intrusion detection in Industrial Internet of Things (IIoT) networks due to their suitability for resource-constrained edge deployment. Most reported results evaluate these models only within their training network, leaving behavior on unseen networks unverified. This study trains four lightweight architectures on one IIoT dataset and evaluates them, without retraining, on two structurally distinct IIoT datasets using a feature representation restricted to attributes available across all three sources. Explainability analysis across two top-performing models shows both rely overwhelmingly on coarse port-category features; the most influential category occurs in source-domain attack traffic at 96 to 435 times the rate in the two target domains, indicating that coarsening port resolution relocates rather than removes a documented shortcut. Evaluation under naturally imbalanced class distributions reveals a further effect: the evaluation protocol used can reverse which target network appears to pose the greater generalization challenge. Adversarial robustness and recovery through limited target-domain exposure are also assessed; robustness to adversarial perturbation is unrelated to cross-network generalization, and recovery through adaptation varies considerably by architecture. These findings suggest deployment readiness should be assessed using cross-network evaluation under realistic class distributions, rather than within-domain accuracy alone.

Jailbreak attacks remain a critical threat to the safe deployment of large language models (LLMs). While prior work has primarily studied attacks and defenses at the prompt level, we show that this prompt-centric paradigm overlooks a structural vulnerability in stateful, function-calling environments. In such applications, developer-defined schemas, structured arguments, and untrusted tool outputs are interleaved into a single shared model context. This architecture expands the attack surface by blurring the boundary between trusted control logic and untrusted data, allowing adversarial intent to be distributed across a multi-turn execution path. We exploit this architectural flaw through SMT, a black-box attack framework based on Simulated Moderation Traces. Departing from purely prompt-based interactions, SMT constructs a multi-turn trajectory that simulates a legitimate moderation-auditing workflow. Within this trajectory, a fabricated moderation frame leverages red-team testing as a pretext to elicit harmful generations. The subsequent validation feedback treats safety refusals as execution failures, prompting refinements that gradually weaken the model's safety constraints and ultimately trigger harmful outputs. Extensive empirical evaluations on prominent commercial LLMs from five different providers across two standardized safety benchmarks show that SMT consistently achieves the highest average attack success rate and HarmScore while requiring a near-minimal number of queries, substantially outperforming existing baselines. These findings demonstrate that prompt-level sanitization alone is fundamentally insufficient for defending tool-enabled LLM systems and highlight the urgent need for context-aware validation across schemas, arguments, tool outputs, and accumulated conversation state. The code is available at https://github.com/liujlong27/SMT.

Sophisticated cyber attacks, particularly Advanced Persistent Threats (APTs), require effective post-intrusion forensic analysis. Provenance-based backward tracking reconstructs attack scenarios by tracing causality from security alerts, but existing methods rely on low-level statistical features and rigid traversal strategies, limiting their ability to capture high-level adversarial intent and suffering from dependency explosion. We present Minos, a multi-agent framework that formulates backward tracking as an LLM-driven reasoning process. Minos adopts a two-tiered architecture: for event-level analysis, it combines hierarchical context management, retrieval-augmented reasoning with citation verification, and adversarial deliberation to improve reasoning quality; for graph exploration, it coordinates four specialized agents under a finite state machine (FSM), replacing exhaustive traversal with hypothesis-guided reasoning and count-first query protocols to efficiently prune the search space. Experiments on 14 attack scenarios across five public datasets show that Minos achieves an average recall of 0.92 and precision of 0.64, significantly outperforming state-of-the-art baselines while producing attack subgraphs that are 49% more compact. Moreover, Minos generates interpretable reasoning throughout the tracking process, facilitating forensic auditing and system refinement. These results demonstrate the effectiveness of LLM-driven reasoning for automated provenance-based backward tracking.

Retrieval-Augmented Generation (RAG) systems are vulnerable to poisoning attacks that inject malicious documents into the retrieval process to manipulate model outputs. Recent Agentic RAG systems are more robust to such attacks because they iteratively perform retrieval and reasoning, allowing them to ignore weakly relevant poisoned documents and preserve the reasoning chain induced by the user query. However, existing attacks on Agentic RAG systems often assume white-box access to system prompts, reasoning traces, retrievers, or model parameters, limiting their applicability in realistic settings. In this paper, we study black-box poisoning attacks against Agentic RAG systems, where the attacker can only publish externally retrievable poisoned documents. We propose KidnapRAG, a sequential poisoning attack that hijacks the agent's multi-step reasoning chain using three role-specific documents: Bait, Chain-Link, and Mal-Ins, which attract initial retrieval, induce query reformulation, and provide attacker-controlled evidence, respectively. Experiments across multiple Agentic RAG frameworks, LLM backbones, and benchmarks show that KidnapRAG consistently outperforms existing poisoning baselines under black-box conditions. Further analyses show that KidnapRAG progressively weakens the original retrieval intent, redirects retrieval behavior, and increases reliance on attacker-controlled evidence. Our code is publicly available at https://github.com/chanwoochoi316/KidnapRAG.

Large language models are increasingly used by participants on crowdsourcing platforms when responding to surveys, potentially undermining the validity of collected data. Our study aims to quantify the prevalence of this behavior and investigate methods to detect and prevent it. In a series of surveys (N = 250), we examined conditions such as platform choice, survey length, requests not to use AI, and disabling copy-paste functionality. We were able to identify distinct characteristics of LLM-assisted responses and found that their frequency varied widely, from under 10% on Prolific to over 80% on Mechanical Turk. Mitigation measures reduced LLM usage but did not necessarily improve data quality. No participants employed browser-use agents at the time of our survey, but we report on our own detection experiments. We recommend that researchers actively screen survey responses for LLM usage by recording and analyzing keystroke data and crafting instructions and questions aimed at AI.

Mobile on-device AI (MoAI) systems that integrate locally deployed AI models with conventional mobile software components are emerging as a key paradigm for delivering intelligent functionality directly on end-user devices. By moving inference from remote cloud services to the local mobile environment, such systems enable privacy-preserving, low-latency, and offline-capable AI functionality, yet introduce new security risks arising from the local storage of AI models. This paper presents the first comprehensive systematization of knowledge on MoAI security, covering security pillars, attack landscape, and defense landscape of MoAI systems. We further identify unresolved gaps in current attack and defense research and point to promising directions for future research in this emerging area. Our work establishes the first systematic framework for understanding the attack and defense landscapes of MoAI systems, serving as a foundation for building secure MoAI systems and advancing research in this critical domain. Companion resources are available at https://github.com/Jinxhy/Awesome-MoAI-Security.

Vision–Language Models (VLMs) are increasingly deployed in safety-critical applications, yet remain vulnerable to backdoor attacks. Existing methods primarily manipulate final outputs, often producing reasoning traces that are inconsistent or easily detectable. In this paper, we propose ReShift, the novel aha-moment-driven reasoning-level backdoor framework that explicitly redirects the internal chain-of-thought (CoT) trajectory while preserving surface-level coherence. ReShift introduces a Poisoned Reasoning-Aware Data Construction (PRDC) pipeline and a Supervised–Reinforcement Joint Optimization (SRJO) strategy to induce stable trigger-conditioned reasoning shifts. We further formalize Entropy Rebound as a principled signal for characterizing reasoning redirection and provide theoretical guaranties linking entropy gaps to trajectory-level divergence. Extensive experiments demonstrate that ReShift achieves high attack success rates while maintaining clean-task performance and realistic reasoning traces, substantially improving stealthiness against existing defenses.

Third-party mobile agents powered by Vision-Language Models (VLMs) have emerged as a promising paradigm for automating smartphone interactions. These agents act as high-privilege decision-makers, perceiving device states through screenshots and executing actions via VLM reasoning, transforming how an agent app interacts with the environment (i.e., other apps or the OS). Correspondingly, this transformation introduces new attack surfaces or transforms benign/harmless interfaces into exploitable ones for mobile devices. In this paper, we summarize key differences between third-party mobile agent apps and general apps when interacting with the environment, analyze the security posture of agents, and identify two unique attack surfaces compared to general mobile apps: the Screen Perception Attack Surface, which exploits the gap between human and machine vision, and the Misused Channel Attack Surface, which intercepts or manipulates the agent's execution pipeline. We design and implement seven concrete attacks, from subliminal text injection and invisible pixel zone exploitation to screenshot tampering and host PC command injection. Our evaluation of five popular mobile agent frameworks demonstrates that a malicious app can hijack agent actions and achieve arbitrary command execution even without any privilege permissions, while remaining visually indistinguishable to users. These findings reveal a fundamental trust mismatch in autonomous agent design and highlight the urgent need for perception-aware security models on multi-tenant platforms.

This paper introduces the Federated Sovereign Transport Protocol (FSTP), a synchronization boundary and transport layer for federated networks in which nodes have heterogeneous privacy requirements. Existing federation protocols leave data confinement to operator policy: they define message formats and delivery semantics but impose no structural constraint on what a conforming server may emit. FSTP addresses this gap by making data confinement a property of the protocol itself. The central mechanism is a synchronization agent whose output type set is formally closed. Raw internal data cannot appear in any federation message because the constraint is enforced by the Rust type system at compile time, not by a runtime check. A received artifact carries a usage scope: a conforming agent refuses to re-emit it toward a third node unless that scope, or a fresh governance grant, authorizes the recipient. A contextual identity model derives a separate, unlinkable identifier for each federation relationship, preventing cross-context correlation structurally. A Blocklace-based event substrate provides tamper-evident, partially ordered logging with synchronization cost proportional to the symmetric difference between node states, and supports data erasure without breaking the hash chain. The result is proof without exposure: a federation participant can verify that a process occurred, that a credential is authentic, and that an outcome is uncorrupted without accessing the internal data that produced these artifacts. FSTP is developed as the inter-node transport layer of Velyzor, a governance platform for institutions with demanding confidentiality requirements. The specification and reference implementation are released as open-source infrastructure under Apache 2.0; source code and figures accompany this paper.

Additive Manufacturing (AM) has revolutionized major sectors, including aerospace, automotive, and healthcare, by enabling adjustable production. As the usage of AM increases, so does the risk of Intellectual Property (IP) leakage during the printing process due to unintended side-channel emissions. Current studies and attack scenarios on 3D printers face three challenges: low success and accuracy rates in final G-code reconstruction, limited distance range for attacking the 3D printer's IP, and reliance on specialized, overt data-collection tools. This paper presents a side-channel attack that addresses the noted limitations by using two smartphones' internal sensors. We position the smartphones 60 cm away in a non-line-of-sight setup to collect the 3D printer's acoustic and magnetic emissions. Our attack successfully reconstructs the G-code commands of the final objects at a rate of 98.89% on command-level reconstruction accuracy. Additionally, we evaluate the transferability of our attack strategy by applying it to another 3D printer in a different environment. Our proven unauthorized access to the reconstructed G-code and thus to the IP of the AM system indicates the security weaknesses in 3D printing, highlighting the need for mitigating side-channel attacks.

Semantic communication (SemCom) has emerged as a promising paradigm in which the transmission of task-relevant information is prioritized over raw data, enabling efficient and robust communication under resource and channel constraints. In this paper, the privacy implications of relay-assisted SemCom systems are studied, where the intermediate relay node operates directly on learned latent representations. It is shown that the relay, even without access to source data, can reliably infer semantic meaning and reconstruct signals with performance comparable to that of the legitimate receiver, revealing a fundamental privacy vulnerability of semantic representations. To address this issue, an iterative adversarial training framework is proposed in which a strong, adaptively trained eavesdropper at the relay is explicitly accounted for. The proposed approach alternates between optimizing the relay's eavesdropping function and the legitimate system, resulting in representations that preserve semantic decoding performance at the intended receiver while degrading semantic inference at the relay. The semantic accuracy gap between the legitimate receiver and the eavesdropper is significantly enlarged across channel conditions. Importantly, this protection is achieved in a stealthy manner, with high reconstruction fidelity maintained while semantic leakage is selectively suppressed.