Search arXiv⌕ Search

arXiv subjects

Wenyu Chen

Publications and source records attributed to Wenyu Chen.

At least 19 recordsLinked to original sources
arXiv2026

Defenses at Odds: Measuring and Explaining Defense Conflicts in Large Language Models

Large language models (LLMs) may require additional defenses after deployment as risks and governance requirements evolve. Subsequent defenses can interact with earlier defenses, raising the question of their sequential compatibility. We study this question with CONFLICTEVAL, evaluating 144 ordered compositions of six defenses spanning safety, privacy, and fairness across six models. The resulting interactions are heterogeneous across defense pairs, application orders, and models. Notably, some compositions exhibit defense conflicts: the subsequent defense improves its target objective while weakening protection established by the earlier defense. We investigate these interactions through the directional compatibility of defense-induced changes in risk-relevant representations. Across 11 selected cases, we use activation interventions to assess which defense-induced representational changes support protection and examine how subsequent defenses affect these changes. We find that, in some conflict cases, subsequent defenses counteract representational changes supporting earlier protection, providing evidence for one possible pathway to defense conflicts. Building on this analysis, we propose Conflict-Triggered Directional Retention (CTDR), which penalizes opposing shifts along directions supporting earlier protection. On six selected conflicting compositions from this analysis, CTDR reduces first-objective regression while maintaining positive gains on the subsequent defense objective.

cs.CR↗
arXiv2026

SceneJail: Exploiting Video Scenario Context to Jailbreak Multimodal LLMs

Video Multimodal Large Language Models (Video-MLLMs) support reasoning over video inputs, yet remain vulnerable to jailbreak attacks that elicit policy-violating responses. Existing video jailbreaks primarily manipulate how harmful queries are visually presented, thereby treating video merely as a carrier. Consequently, the surrounding video scenario remains unexplored as a contextual attack surface. In this paper, we show that the same harmful query can elicit different safety responses when placed in different video scenarios. To systematically exploit this vulnerability, we propose SceneJail, an adaptive black-box jailbreak framework with two coordinated components. Adaptive Scenario Construction dynamically searches for a surrounding scenario that is contextually compatible with the harmful query. Scenario-aware Prompt Search uses black-box response feedback to search for textual guidance tailored to the selected scenario. Extensive evaluations on the HADES and SafeBench datasets across eight Video-MLLMs, including two proprietary models, GPT-4.1 and Gemini3.5-Flash, demonstrate the effectiveness of SceneJail. SceneJail-F, which presents the complete query persistently, achieves average attack success rates (ASR) up to 91.5%, outperforming the strongest baselines by 29.1 percentage points. Furthermore, SceneJail-S, which distributes the query across successive frames, remains highly robust against current defenses, retaining a 72.3% ASR even under strict image filtering.

cs.CR↗
arXiv2026

One Risk Down, Another Up: Cross-Risk Interactions Induced by LLM Defenses

Large Language Models (LLMs) are increasingly deployed in high-stakes settings, where they face diverse risks. Numerous defense strategies have been proposed to mitigate these risks, but they are almost always evaluated in isolation. This isolated view leaves a critical question open: does mitigating one risk inadvertently change a model's exposure to others? Beyond the well-studied risk-utility trade-off, we present the first systematic study of cross-risk interactions induced by LLM defenses. We propose CrossRiskEval, an evaluation paradigm that situates a defended model in a multi-dimensional risk space and quantifies how a defense built for one risk shifts the others. Among 166 cross-risk evaluations covering 32 defended models, 77.1% exhibit statistically significant cross-risk interactions. Most of these interactions amplify non-target risks, with increases exceeding 100% in some cases. Beyond behavioral evaluation, we conduct neuron-level analyses in seven selected cases to investigate one possible pathway associated with these interactions. We identify conflict-entangled neurons whose activation interventions produce opposing effects on proxies for the target and non-target risks. In conflict cases, restoring these neurons to their base-model activations partially reduces the corresponding risk increases, providing evidence that defense-induced changes to these neurons may contribute to the observed interactions. Building on this evidence, we propose Conflict-Aware Freezing, a training-time strategy that prevents direct updates to the parameters associated with the identified neurons. Across five conflict cases, it offsets 35%-196% of non-target risk amplification while meeting the original defense criterion.

cs.CR↗
arXiv2026

Transferable End-to-End Optimization for Indirect Long-Term Memory Poisoning in LLM Agents

Long-term memory can turn untrusted external content into persistent influence over an LLM agent's future decisions, creating the threat of indirect memory poisoning. A successful attack must survive a multi-stage pipeline comprising memory writing, retrieval, and utilization. Existing attacks largely rely on intra-stage optimization, optimizing individual stages in isolation while overlooking inter-stage coupling. Specifically, these stages impose different requirements on the same poisoning content, and each stage operates on the transformed output of its predecessor. Consequently, optimizing one stage may undermine the effectiveness of other stages, while upstream transformations may erase improvements intended for downstream stages. Indirect memory poisoning should therefore be viewed as an end-to-end optimization problem. Based on this insight, we present \textsc{PipePoison}, which collects fine-grained stage feedback from local shadow systems, uses chain-structured losses to identify and optimize the stage bottlenecking end-to-end success, and applies stability-calibrated stage and configuration weights to improve transferability. Across three agent frameworks and four memory mechanisms, \textsc{PipePoison} improves attack utilization rate by 19.1 percentage points. Even on fully unseen victim configurations, it outperforms the strongest baseline by 16 percentage points and remains effective under eight representative defenses.

cs.CR↗
arXiv2026

Understanding Stage-Wise Utility-Risk Trade-offs in LLM Agent Memory

Long-term memory is becoming a core capability of LLM agents, enabling personalization and long-horizon interaction. However, memory mechanisms that retain, transform, or expose more information can affect both benign utility and susceptibility to memory poisoning. Existing evaluations typically measure memory utility or attack risk in isolation under fixed configurations, providing limited insight into how stage-specific design choices reshape their trade-off. We present \textsc{MemGauge}, a controllable framework that separately varies writing admission, management policy, and retrieval exposure under matched clean and poisoned conditions. Across 11 LLMs and two long-term memory benchmarks, controlled evaluations reveal three distinct profiles: a threshold-like risk transition during writing, policy-dependent local decoupling during management, and coupled growth of utility and risk during retrieval. We further apply analogous stage-level measurements to four existing memory systems and observe diagnostic associations qualitatively consistent with these profiles. These results show that targeted poisoning risk varies across memory operations and motivate stage-aware evaluation and control of LLM-agent memory.

cs.CR↗
arXiv2026

Extracting Knowledge from Tools in LLM Agents

LLM agents commonly use knowledge-based tools and access their underlying files, databases, and search indexes through tool invocation. This integration improves agents' ability to provide domain-specific services but also introduces the risk of tool-mediated knowledge extraction: source content exposed to an agent for legitimate responses may be progressively recovered from its outputs, enabling reconstruction of the knowledge source behind a target tool. This paper systematically investigates this risk and identifies two challenges introduced by tool invocation: tool-selection uncertainty, where an agent may invoke a competing tool instead of the target tool, and tool-argument compression, where fine-grained query information may be lost when the agent generates tool arguments. To tackle these challenges, we propose ToolSiphon, a query-only extraction attack that introduces two complementary signals: a target-discriminative signal, implemented through Tool Contrastive Analysis, to steer queries toward the target tool; and a response-grounded factual signal, implemented through Evidence Chained Feedback, to mitigate argument compression and progressively expand extraction coverage. Across three types of knowledge-based tools and six domain-specific datasets, ToolSiphon recovers 74.3% of source records on average when coarse-grained information about non-target tools is available, with 83.2% textual recovery and 90.2% semantic similarity. Even without such information, it recovers 66.3% of source records. ToolSiphon also remains effective against representative defenses and on three real-world agent platforms.

cs.CR↗
arXiv2026

Understanding and Exploiting Initialization Anchoring Weakness in Feedback-Based Agent Planning

Feedback-based planning improves agent reliability by incorporating tool observations and corrective feedback. However, its protection may not be distributed uniformly across planning stages. We conduct a round-wise analysis of four representative feedback mechanisms and uncover an initialization anchoring weakness: the first feedback round corrects 46\% of adversarial directions, whereas the rates fall to 13\% and 7\% among directions surviving into the next two rounds. Our analysis attributes this weakness to three interacting factors: a contextually plausible shift in the initial plan, insufficient counterevidence, and the persistence of accepted directions in the accumulated trajectory. Based on these findings, we propose \textsc{InitAnchor}, a black-box framework for exploiting this weakness through attacker-controlled external materials. It operationalizes the three factors as directional-shift, contextual-plausibility, and counterevidence-resilience signals under either limited target access or no target access. Across 112 tasks from 16 domains, six agent architectures, and five backbone LLMs, \textsc{InitAnchor} achieves average ASRs of 76.1\% and 72.0\% under the two settings while reducing first-round mitigation rates to 21.0\% and 25.0\%, respectively. It also remains effective against six defenses and across six real-world agent systems. These findings show that feedback-based agents can retain early biases even when later correction is available.

cs.CR↗
arXiv2026

Alipay-PIBench: A Realistic Payment Integration Benchmark for Coding Agents

Payment integration is a demanding repository-level software task: agents must select a suitable product, implement coordinated client-server flows, verify payment outcomes, and preserve consistency between transaction and business states. We introduce Alipay-PIBench, a benchmark for evaluating coding agents on realistic Alipay payment integration. It contains nine product-specific projects and 18 task instances, each organized into Basic functional-completion and Advanced risk-aware hardening scenarios. Scenario-specific rubrics support deterministic static, unit, integration, and end-to-end checks, supplemented by LLM-assisted assessment for semantic requirements. We evaluate six coding-agent models and report rubric pass rate (RPR). Under the with-skill condition, mean RPR ranges from 68.58% to 91.37%. Access to the alipay-payment-integration skill improves mean RPR by 10.31 percentage points on average relative to the without-skill condition, with gains varying across models, products, and scenarios. Method-level results distinguish source-level completion, executable payment behavior, and payment-domain requirements. Alipay-PIBench provides a controlled setting for diagnosing model capability and evaluating structured guidance in payment integration.

cs.AI↗
arXiv2026

Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents

LLM-based agents extend large language models with long-term memory (LTM) that persists privacy-sensitive user data across sessions. Production systems mitigate extraction risks through memory isolation, binding each user's LTM to a unique identifier. This defense has blocked known attacks on shared storage, fostering the assumption that isolated LTM is secure. We identify the tool interface as an overlooked attack surface. Agents routinely embed LTM-retrieved data in tool invocation parameters, enabling a malicious tool to exfiltrate private memory without violating user-level isolation. Naive adaptations of user-side extraction techniques fail because the adversarial command's semantics interfere with retrieval precision, and platform-imposed tool-call limits constrain the extraction budget per trigger. We present SPORE, the first extraction attack designed for this threat model. SPORE decouples the adversarial command from retrieval anchors by persisting the command in short-term memory and emitting semantically pure anchors in tool responses. The restored retrieval precision enables a geometric coverage optimization over the embedding space that systematically steers anchors toward unexplored memory regions. To sustain extraction beyond tool-call limits, SPORE persists reactivation payloads in memory that automatically resume the attack within and across sessions without additional user triggers. SPORE achieves an 80.0% record extraction rate with unlimited triggers and 47.0% with only 20 triggers. In multi-user deployments, attackers can link extracted records to user identities, enabling targeted surveillance. These results demonstrate that memory isolation alone is insufficient and call for reexamining tool-side trust boundaries in agent architectures.

cs.CR↗
arXiv2026

VLMGuard: Bootstrapping Malicious Prompt Detectors from Unlabeled Vision-Language Prompts in the Wild

Vision-language Models (VLMs) are essential for contextual understanding of both visual and textual information. However, their vulnerability to adversarially manipulated inputs presents significant risks, leading to compromised outputs and raising concerns about the reliability in VLM-integrated applications. Detecting these malicious prompts is thus crucial for maintaining trust in VLM generations. A major challenge in developing a safeguarding prompt classifier is the lack of a large amount of labeled benign and malicious data. To address the issue, we introduce VLMGuard, a novel learning framework that leverages the unlabeled user prompts in the wild for malicious prompt detection. These unlabeled prompts, which naturally arise when VLMs are deployed in the open world, consist of both benign and malicious information. To harness the unlabeled data, we present an automated maliciousness estimation score for distinguishing between benign and malicious samples within this unlabeled mixture, thereby enabling the training of a binary prompt classifier on top. Notably, our framework does not require extra human annotations and is robust to realistic prompt variations, offering strong flexibility and practicality for real-world applications. Extensive experiments show that VLMGuard achieves superior detection results, improving AUROC by 5.39% on average over the state-of-the-art method. Disclaimer: This paper may contain offensive examples; reader discretion is advised. Code is available at: https://github.com/radiolab-ntu/vlmguard.

cs.LG↗
arXiv2026

Provable Filter for Real-world Graph Clustering

Graph clustering, an important unsupervised problem, has been shown to be more resistant to advances in Graph Neural Networks (GNNs). In addition, almost all clustering methods focus on homophilic graphs and ignore heterophily. This significantly limits their applicability in practice, since real-world graphs exhibit a structural disparity and cannot simply be classified as homophily and heterophily. Thus, a principled way to handle practical graphs is urgently needed. To fill this gap, we provide a novel solution with theoretical support. Interestingly, we find that most homophilic and heterophilic edges can be correctly identified on the basis of neighbor information. Motivated by this finding, we construct two graphs that are highly homophilic and heterophilic, respectively. They are used to build low-pass and high-pass filters to capture holistic information. Important features are further enhanced by the squeeze-and-excitation block. We validate our approach through extensive experiments on both homophilic and heterophilic graphs. Empirical results demonstrate the superiority of our method compared to state-of-the-art clustering methods.

cs.LG↗
arXiv2026

Mitigating Hallucinations in Large Vision-Language Models via Causal Route Gating

Large vision-language models (LVLMs) often hallucinate content that is fluent yet unsupported by the image, limiting their reliability in real-world deployment. We show that a key failure mode arises from route competition: even when visual tokens receive attention, the final token decision can be dominated by the textual pathway, causing the decoder to follow linguistic priors over visual evidence. To mitigate this, we propose a training-free, decision-aligned intervention that decomposes each attention head into a visual route and a text route, and estimates their token-level effects using an efficient one-forward/one-gradient approximation. These estimates reveal route conflict within heads and identify prior-dominant ones, enabling selective suppression of only the text route while keeping the visual route intact. Across five benchmarks spanning discriminative and generative settings, our method consistently reduces hallucination-related errors across models with limited impact on overall multimodal performance, while incurring a modest inference-time overhead.

cs.CV↗
arXiv2026

Beyond the Singular: Revealing the Value of Multiple Generations in Benchmark Evaluation

Large language models (LLMs) have demonstrated significant utility in real-world applications, exhibiting impressive capabilities in natural language processing and understanding. Benchmark evaluations are crucial for assessing the capabilities of LLMs as they can provide a comprehensive assessment of their strengths and weaknesses. However, current evaluation methods often overlook the inherent randomness of LLMs by employing deterministic generation strategies or relying on a single random sample, resulting in unaccounted sampling variance and unreliable benchmark score estimates. In this paper, we propose a hierarchical statistical model that provides a more comprehensive representation of the benchmarking process by incorporating both benchmark characteristics and LLM randomness. We show that leveraging multiple generations improves the accuracy of estimating the benchmark score and reduces variance. Multiple generations also allow us to define $\mathbb P\left(\text{correct}\right)$, a prompt-level difficulty score based on correct ratios, providing fine-grained insights into individual prompts. Additionally, we create a data map that visualizes difficulty and semantics of prompts, enabling error detection and quality control in benchmark construction.

cs.CL↗
arXiv2026

Optimistic ε-Greedy Exploration for Cooperative Multi-Agent Reinforcement Learning

The Centralized Training with Decentralized Execution (CTDE) paradigm is widely used in cooperative multi-agent reinforcement learning. However, conventional methods based on CTDE can suffer from value underestimation and converge to suboptimal solutions. While such underestimation is typically attributed to the representational limitations of monotonic structures, we provide a novel perspective by demonstrating that the insufficient sampling of optimal joint actions during exploration is also a critical factor. To address this problem, we propose Optimistic $ε$-Greedy Exploration. Our method introduces optimistic action-value networks that serve as decoupled exploration indicators, which we theoretically prove to converge in probability to the maximum achievable returns. By sampling actions from these distributions with a probability of $ε$, we effectively increase the selection frequency of high-return joint actions. Experimental results in various environments reveal that our strategy effectively prevents the algorithm from falling into suboptimal solutions and significantly improves final returns, win rates, and convergence speeds compared to other enhanced algorithms. Our code has been open-sourced at https://github.com/qxqxtxdy/OptimisticExploration.

cs.MA↗
arXiv2026

Sparse Gaussian Graphical Models with Discrete Optimization: Computational and Statistical Perspectives

We consider the problem of learning a sparse graph underlying an undirected Gaussian graphical model, a key problem in statistical machine learning. Given $n$ samples from a multivariate Gaussian distribution with $p$ variables, the goal is to estimate the $p \times p$ inverse covariance matrix (aka precision matrix), assuming it is sparse (i.e., has a few nonzero entries). We propose GraphL0BnB, a new estimator based on an $\ell_0$-penalized version of the pseudo-likelihood function, while most earlier approaches are based on the $\ell_1$-relaxation. Our estimator can be formulated as a convex mixed integer program (MIP) which can be difficult to compute beyond $p\approx 100$ using off-the-shelf commercial solvers. To solve the MIP, we propose a custom nonlinear branch-and-bound (BnB) framework that solves node relaxations with tailored first-order methods. As a key component of our BnB framework, we propose large-scale solvers for obtaining good primal solutions that are of independent interest. We derive novel statistical guarantees (estimation and variable selection) for our estimator and discuss how our approach improves upon existing estimators. Our numerical experiments on real and synthetic datasets suggest that our BnB framework offers significant advantages over off-the-shelf commercial solvers, and our approach has favorable performance (both in terms of runtime and statistical performance) compared to the state-of-the-art approaches for learning sparse graphical models.

cs.LG↗
arXiv2026

Not Just Large: Tall Teams Dominate East Asia's Scientific Production

Purpose: This study compares the hierarchical structure of scientific teams across countries and investigates factors associated with the observed cross-national differences. Design/methodology/approach: Drawing on 150,817 publications with author contribution statements, we focus on the 15 countries with the largest volume of scientific publications, examine cross-country variations in the proportion of tall teams, and analyze how this proportion correlates with other factors. Findings: Scientific output from East Asia is dominated by tall teams, which persist after controlling for team size, indicating that this pattern cannot be fully accounted for by the prevalence of larger teams in these countries. Cultural factors, measured by Power Distance, as well as the observed funding patterns of major basic science agencies, are associated with the dominance of tall teams in East Asia. Research limitations: This study is limited by its reliance on publications with author contribution statements, which may introduce selection bias; its focus on cultural and funding factors, while leaving other institutional contexts unexamined; and its use of a leadership concentration measure that does not capture other dimensions of hierarchy. Practical implications: Understanding cross-national differences in research team structures and their associated cultural and institutional factors can inform science policy and team management. Originality/value: This study provides a systematic cross-national comparison of team hierarchy and offers a mechanistic understanding of the dominance of tall teams in East Asia, highlighting associations with cultural and funding factors.

cs.DL↗
arXiv2026

Not All Tokens Are Created Equal: Query-Efficient Jailbreak Fuzzing for LLMs

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

cs.CR↗
arXiv2026

Linear thermal noise induced by Berry curvature dipole in a four-terminal system

In this work, we numerically investigate linear thermal noise in a four-terminal system with a finite Berry curvature dipole (BCD) using the nonequilibrium Green's function formalism. By comparing with the semiclassical results for bulk systems, we establish a one-to-one correspondence between terminal-resolved linear noise in multi-terminal systems and direction-resolved noise in bulk transport. Specifically, the auto-correlation function scales as $2 k_B T$ when the driving field is perpendicular to the BCD and vanishes when they are parallel, whereas the cross-correlation scales as $k_B T$. Both the auto- and cross-correlation functions exhibit pronounced peaks near the band edges, consistent with BCD-induced features. In addition, the linear thermal noise increases approximately linearly with $T$ at low temperatures and is suppressed by dephasing effect at high temperatures. Our work bridges semiclassical bulk theory and quantum multi-terminal theory for linear thermal noise, highlighting the symmetry(geometry)-selection rule in quantum transport.

cond-mat.mes-hall↗