Searcharxiv⌕ Search

arXiv subjects

Zhibo Wang

Publications and source records attributed to Zhibo Wang.

At least 19 recordsLinked to original sources

SAVeR$^2$: Reasoning-based Safety Alignment for Large Reasoning Models via Verifiable Rewards

Large Reasoning Models (LRMs) have achieved impressive performance via explicit Chain-of-Thought (CoT) reasoning, yet this process introduces critical safety risks. We formalize LRM safety objectives as requiring non-malicious reasoning and answers, safety consistency, and explicit intent identification. We argue that existing alignment methods suffer from poor adversarial generalization, a significant ``safety tax'' on reasoning, and unfaithful coarse-grained rewards, primarily due to their neglect of the LRM's intrinsic reasoning capabilities for safety. To address these, we propose \textbf{SAVeR$^2$}, a novel \textbf{S}afety \textbf{A}lignment method via \textbf{R}ewards with \textbf{R}easoning capability. SAVeR$^2$ decomposes the safety objective into four fine-grained verifiable reward signals and utilizes a difficulty-aware data selection strategy to stabilize training. Crucially, we introduce a reasoning-preserving gradient projection mechanism that analytically resolves conflicts between safety and reasoning gradients, effectively eliminating the safety tax. Extensive experiments on models ranging from 1.5B to 671B demonstrate that SAVeR$^2$ achieves superior safety performance and significantly lower over-refusal rates across multiple benchmarks without compromising general reasoning capabilities.

cs.AI↗

UT-HMR: Uncertainty-Guided Temporal Modeling for Robust 3D Human Mesh Recovery

Monocular 3D human mesh recovery (HMR) remains unreliable under weak observations caused by long camera-to-person distances, occlusion, and motion blur. In challenging scenarios such as sports, these factors make the problem highly ambiguous, as different 3D poses may yield similar 2D projections. This ambiguity is especially severe for one-stage HMR, which recovers all meshes directly from the full image without person crops or external 2D cues. To handle this ambiguity, we present UT-HMR ( Uncertainty-guided Temporal HMR), to our knowledge the first one-stage HMR framework to capture correlated pose uncertainty across joints. Each person query predicts a Gaussian distribution over the full SMPL pose with a compact low-rank-plus-diagonal covariance, which is converted into joint-wise precision weights for heteroscedastic supervision while keeping inference deterministic. Temporal early fusion further injects neighboring-frame evidence before pose decoding, supplying the visual evidence needed to resolve the ambiguity that the uncertainty captures. With a lightweight ViT-B backbone, UT-HMR achieves state-of-the-art results on VIP-HockeyPose-v2 (ice hockey) and WorldPose (soccer), reducing MPJPE, PA-MPJPE, and PVE by up to 4.7%, 3.8%, and 5.6%, and remains competitive on 3DPW against methods with larger backbones and higher input resolutions.

cs.CV↗

AGR: Agentic Generative Retrieval with Tool-Augmented Search

Large-scale search and recommendation systems derive much of their retrieval quality from rich, heterogeneous features, yet existing generative retrieval (GR) methods struggle to exploit such feature ecosystems efficiently. Static feature injection either encodes all available signals or places them according to predetermined schemes, increasing serving cost and introducing irrelevant or conflicting evidence. More fundamentally, existing methods cannot adaptively determine when a feature is needed, which feature to select, or where it should influence hierarchical Semantic ID (SID) generation. We propose Agentic Generative Retrieval (AGR), a tool-augmented paradigm that interleaves SID generation with on-demand feature acquisition. Conditioned on the request and decoded SID prefix, AGR decides whether to invoke a tool, which feature to acquire, and where its observation should influence subsequent decoding, thereby steering generation toward feature-consistent SID branches. To supervise this behavior, we construct tool-augmented trajectories based on feature applicability and information gain along the SID hierarchy, and train the model through SID pretraining followed by tool-augmented fine-tuning. Experiments on an industrial dataset and three public benchmarks, together with an online A/B test, consistently demonstrate the effectiveness of AGR.

cs.IR↗

Some new numerical schemes for delay sub-diffusion equations via the idea of symmetry

In this work, we propose several novel numerical schemes that synergistically integrates the $L1$ method and symmetric fractional-order reduction (SFOR) method, specifically designed for fractional sub-diffusion equations with time delay. Crucially, we rigorously establish an optimal convergence order of $2 - α/2$ in temporal direction, which improves upon the classical rate of $2-α$ under identical smoothness conditions. To further enhance computational efficiency, we develop a weighted alternating direction implicit (ADI) scheme for the two-dimensional case and analyze its coupling with compact spatial discretization---thereby achieving high-order accuracy in space. Moreover, we introduce a new framework for analyzing the error of $L1$-type discretization on graded temporal meshes, which relaxes the mesh grading requirement and still guarantees optimal convergence rate at continuous time points. Some numerical experiments comprehensively validate both the theoretical accuracy and computational efficiency of the resulting schemes.

math.NA↗

Stateful Agent Backdoors: Constructing Cross-Session Attack Programs

Multi-step attacks on large language model agents may depend on opportunities distributed across sessions, such as access to target information or the availability of required tools. To combine these opportunities, an attack needs to retain its state and intermediate results, and choose actions based on current conditions. In this work, we study such attacks as cross-session attack programs. We focus on their shared control structures, including sequential execution with conditional waiting, branching and merging, looping, and condition accumulation. We represent these programs as Mealy machines and construct a sub-backdoor for each transition. We build single-session training trajectories for each sub-backdoor, combine them into a training dataset, and fine-tune the model to learn the local behaviors jointly. After a single injection of the initial trigger, the agent uses persistent memory to connect local behaviors into a complete cross-session attack program. We evaluate four instantiations of these control structures across four models in a LangChain-based agent environment. The primary instantiation achieves mean complete-program success rates of 71.7\%--94.7\% in LangChain. In the official OpenClaw runtime under a controlled configuration, it achieves a complete-program success rate of 85\% for each of two evaluated models. These results demonstrate the feasibility of end-to-end execution of cross-session attack programs with different control structures.

cs.CR↗

Privacy-Preserving LLM Embedding Transmission for End-Cloud Collaboration

Recent studies improve on-device language model (LM) inference through end-cloud collaboration, where the end device retrieves useful information from cloud databases to enhance local processing, known as Retrieval-Augmented Generation (RAG). Typically, to retrieve information from the cloud while safeguarding privacy, the end device transforms original data into embeddings with a local embedding model. However, the recently emerging Embedding Inversion Attacks (EIAs) can still recover the original data from text embeddings (e.g., training a recovery model to map embeddings back to original texts), posing a significant threat to user privacy. To address this risk, we propose EntroGuard, an entropy-driven perturbation-based embedding privacy protection method, which can protect the privacy of text embeddings while maintaining retrieval accuracy during the end-cloud collaboration. Specifically, to defeat various EIAs, we perturb the embeddings to increase the entropy of the recovered text in the common structure of transformer-based recovery models, thus steering the embeddings toward meaningless texts rather than original sensitive texts during the recovery process. To maintain retrieval performance in the cloud, we constrain the perturbations within a bound, applying the strategy of reducing them where redundant and increasing them where sparse. Moreover, EntroGuard can be directly integrated into end devices without requiring any modifications to the embedding model. Extensive experimental results demonstrate that EntroGuard effectively reduces privacy leakage metrics to near-zero levels against learning-based EIAs while also mitigating optimization-based EIAs with negligible loss of retrieval performance.

cs.CR↗

RouteScan: A Non-Intrusive Approach to Auditing MoE LLMs Safety via Expert Routing Telemetry

As Mixture-of-Experts (MoE) architectures are increasingly adopted for scaling Large Language Models (LLMs), safety auditing becomes necessary to verify whether these models produce or facilitate harmful behaviors during operation. However, existing content-based auditing methods typically require access to user prompts, model internals, or outputs, potentially exposing sensitive user information and creating a tension between LLM safety and user privacy. On the other hand, we observe that, in MoE models, different inputs induce different sparse expert-routing patterns, which produce measurable footprints in low-level GPU execution telemetry. We refer to these hardware-observable signals induced by expert-routing decisions as expert routing telemetry; they are derived from GPU execution rather than from router logits or token-level routing assignments. Inspired by this observation, we propose RouteScan, a non-intrusive auditing framework for detecting harmful behaviors through such routing-induced GPU telemetry. Specifically, RouteScan utilizes the number of active GPU threads allocated to expert modules during the prefilling phase as a discriminative micro-architectural fingerprint, and builds a lightweight detection pipeline that isolates cross-domain invariant risk indicators for the precise identification of malicious prompts. Comprehensive evaluations on four open-source MoE LLMs with distinct routing designs demonstrate that RouteScan achieves strong generalization, with an AUROC exceeding 0.91 on unseen harmful domains. Moreover, privacy stress tests show that, although aggregated execution telemetry retains input-related attribute information, full prompts and exact sensitive fields cannot be reliably recovered under the evaluated attacks.

cs.CR↗

PhyCheck: Fine-Grained Evidence-Grounded Dataset for Physical Law Understanding in Video-LLMs

Embodied intelligence and world models require video understanding systems to go beyond recognizing objects and actions and develop an understanding of physical regularities. However, despite their strong performance on general video understanding tasks, current video-language models still struggle to reliably determine whether an observed event conforms to specific physical laws. Existing benchmarks primarily assess the physical quality of generated videos, providing limited support for systematically evaluating and improving the physical-law understanding of Video Large Language Models (VideoLLMs). To address this gap, we introduce PhyCheck, a video question answering dataset organized at two complementary levels of granularity. The coarse-grained subset asks models to determine whether the phenomenon shown in a video conforms to or violates physical laws, while the fine-grained subset further examines whether models can capture physical details responsible for the violation or compliance. We use these subsets as structured supervision to improve physical understanding. In addition, the dataset contains a diagnostic subset with external causal context that reveal hidden factors affecting physical plausibility, assessing whether models can recalibrate their judgments accordingly. Experiments with Fine-tune Qwen2.5-VL show that training with the proposed data substantially improves the understanding of physical-consistency, while evaluations in the diagnostic subset reveal that current models still have difficulty incorporating additional causal conditions into their decisions. These findings highlight the gap between recognizing surface-level inconsistencies and understanding underlying physical mechanisms, and provide a foundation for evaluating and improving physical understanding in Video-LLMs.

cs.CV↗

Can Small Language Models Reliably Resist Jailbreak Attacks? A Comprehensive Evaluation

Small language models (SLMs) have emerged as promising alternatives to large language models (LLMs) due to their low computational demands, enhanced privacy guarantees, and comparable performance in specific domains. Deploying SLMs on edge devices, such as smartphones and smart vehicles, has become a growing trend. However, the security implications of SLMs have not received as much attention as those of LLMs, particularly concerning the significant jailbreak threats they face. In this paper, we conduct the first systematic empirical study of SLMs' vulnerabilities to jailbreak attacks. Through systematic evaluation on 59 SLMs from 15 mainstream SLM families against 12 state-of-the-art jailbreak methods, we demonstrate that 61.0% of evaluated SLMs show an average ASR of more than 40% under jailbreak attacks and 37.3% of them have an ASR of more than 50% on direct harmful queries. Through correlation analysis, we identify that SLM vulnerabilities are closely related to training details (e.g., training dataset and method) rather than model size scaling. We further evaluate five defenses for jailbreak attacks, revealing that prompt-level defenses remain inconsistent across SLMs and attack methods, while model-level defense improves robustness against similar attacks yet generalizes poorly to multi-turn attacks such as Crescendo, highlighting the urgent need for security-by-design approaches in SLM development.

cs.CR↗

Inverting the Hidden: Unveiling Multimodal Privacy Leakage in Collaborative LVLM Inference

Collaborative inference deploys Large Vision-Language Models (LVLMs) by partitioning computation between edge devices and the cloud. While withholding raw inputs supposedly ensures privacy, transmitting intermediate hidden states exposes a critical attack surface. However, it remains unclear whether deep-layer LVLM hidden states retain recoverable private information, given that visual content has been projected into the language embedding space. To address this concern, we theoretically analyze LVLM hidden-state recoverability and show that, under regularity assumptions and a positive semantic--nuisance margin, privacy-relevant visual semantics remain identifiable and stably recoverable. Motivated by this analysis, we propose RASR, a novel coarse-to-fine multimodal reconstruction attack. RASR obtains initial image and text reconstructions through modality-specific inverse paths that follow their respective forward processing pipelines in reverse, and then uses hidden-state consistency to refine both reconstructions. Evaluations on Qwen3-VL-8B-Instruct and LLaVA-1.5-7B across five datasets demonstrate that RASR reduces image reconstruction MSE by \(\sim\)50\% compared to the strongest baselines, while achieving up to 99\% token accuracy for text recovery. These results show that privacy-sensitive visual and textual information can be recovered even from deep-layer LVLM hidden states, exposing the privacy risks of collaborative inference.

cs.CR↗

From Role Prompt to Infinite Thinking: Exploiting Persona Conditioning for Inference Cost Attacks in LLMs

LLMs are increasingly deployed in real-world applications, making inference efficiency and service reliability critical concerns due to their substantial computational costs. However, the autoregressive generation mechanism of LLMs enables malicious prompts to manipulate generation behaviors, inducing excessive token generation that amplifies computational consumption and threatens service efficiency. Existing methods mainly rely on adversarial suffixes or explicit extension instructions, which introduce detectable behaviors and limit their applicability. In this paper, we reveal a previously unexplored vulnerability caused by persona consistency in LLMs, where models maintain assigned roles and reproduce corresponding behaviors even when they result in inefficient reasoning and excessive generation. Based on this observation, we propose RolePlay, a task-aware dynamic persona alignment framework that constructs adaptive personas to naturally induce inefficient yet semantically coherent behaviors for inference cost amplification. Extensive experiments across multiple LLMs and diverse task datasets demonstrate that RolePlay consistently outperforms existing inference extension methods, achieving an average token amplification of up to \bm{$7.64\times$} and a maximum token amplification ratio of \bm{$207.64\times$}. Our findings identify persona conditioning as a new attack surface for LLM inference efficiency and offer a new perspective on computational cost amplification.

cs.CR↗

Hiding Faces in Plain Sight: Defending DeepFakes by Disrupting Face Detection

Face-swapping DeepFakes have become an escalating societal concern, attracting increasing attention in recent years. To counter this, we investigate a new proactive defense framework to prevent individuals from being victimized in DeepFake videos. The core idea of this framework is to contaminate the inputs of DeepFake models by disrupting face detectors, based on the observation that face detectors are commonly used to automatically extract victim faces in most DeepFake techniques. Once the face detectors malfunction, the faces will not be correctly extracted, thereby impairing the training or synthesis stages of DeepFake models. To achieve this, we describe a strategy named {\em FacePoison}, which fools face detectors by adding dedicated adversarial perturbations to video frames. Building upon this, we introduce {\em VideoFacePoison}, an extended strategy that can efficiently propagate FacePoison across video frames instead of applying it individually to each frame, thus significantly reducing the computational overhead while retaining favorable attack performance. This framework is validated on five face detectors, and extensive experiments against eleven different DeepFake models demonstrate the effectiveness of disrupting face detectors to hinder DeepFake generation. The source code is publicly available at: https://github.com/OUC-VAS/FacePoison.

cs.CV↗

Attention-Free and Lightweight Token Reduction for Efficient Vision-Language Models

Vision-Language Models (VLMs) have achieved strong performance in multimodal understanding, yet remain challenging to deploy on resource-constrained edge devices due to the substantial computational overhead of processing numerous visual tokens. Token reduction is a promising direction for accelerating VLMs inference, but existing approaches either rely on attention maps that are incompatible with modern acceleration frameworks or depend on computationally intensive pairwise similarity comparisons, which undermine scalability and negate their practical benefits in deployment. In this paper, we propose an attention-free and lightweight token reduction framework as a plug-and-play module for VLMs, which preserves both important and diverse tokens to produce a compact visual representation. First, to enable attention-free importance estimation, we adopt an information-theoretic perspective and quantify token information using a novel entropy-based criterion, retaining those with more expressive and less degenerate feature representations. Second, to ensure diverse visual coverage in a lightweight manner, we introduce a transformation-induced consistency signal where similar tokens yield similar signals, such that sorting by this signal places similar tokens close to each other and enables stride-based selection to produce a diverse token set. Extensive experiments across multiple VLMs benchmarks demonstrate that our framework achieves a favorable accuracy-efficiency trade-off, maintaining competitive performance under aggressive compression.

cs.CV↗

LoRA-Key: User-Centric LoRA Watermarking for Text-to-Image Diffusion Models

Low-Rank Adaptation (LoRA) has become a widely used mechanism for customizing text-to-image diffusion models, enabling lightweight modules that are shared, reused, and commercialized as independent assets. This LoRA-centric ecosystem shifts copyright protection from foundation models to distributed LoRA modules, which are easy to copy, redistribute, or reuse without authorization. Existing watermarking methods either protect the base diffusion model or require watermark-aware retraining for each target LoRA, limiting their practicality in open community settings. To address this limitation, we propose LoRA-Key, a user-centric LoRA watermarking framework that treats copyright protection as a reusable ownership key. LoRA-Key encapsulates a recoverable secret message into a standalone user-specific Watermark LoRA, which can be attached to different target LoRAs through training-free linear superposition without per-LoRA retraining or structural modification. To train such a reusable key, we first establish a latent watermark prior in the frozen VAE latent space for robust message embedding and recovery, and then optimize the Watermark LoRA with message-conditioned watermark supervision and semantic consistency constraints. We further introduce Gradient Orthogonal Projection (GOP) to suppress watermark updates that conflict with semantic-preserving directions, reducing interference with generation fidelity and downstream style adaptation. Extensive experiments show that LoRA-Key provides lightweight plug-and-play copyright protection while preserving generation quality and style fidelity, and maintains robust ownership verification under image-level distortions, downstream fine-tuning, and multi-LoRA composition.

cs.CR↗

PromptCOS: Towards Content-only System Prompt Copyright Auditing for LLMs

System prompts are critical for shaping the behavior and output quality of large language model (LLM)-based applications, driving substantial investment in optimizing high-quality prompts beyond traditional handcrafted designs. However, as system prompts become valuable intellectual property, they are increasingly vulnerable to prompt theft and unauthorized use, highlighting the urgent need for effective copyright auditing, especially watermarking. Existing methods rely on verifying subtle logit distribution shifts triggered by a query. We observe that this logit-dependent verification framework is impractical in real-world content-only settings, primarily because (1) random sampling makes content-level generation unstable for verification, and (2) stronger instructions needed for content-level signals compromise prompt fidelity. To overcome these challenges, we propose PromptCOS, the first content-only system prompt copyright auditing method based on content-level output similarity. PromptCOS achieves watermark stability by designing a cyclic output signal as the conditional instruction's target. It preserves prompt fidelity by injecting a small set of auxiliary tokens to encode the watermark, leaving the main prompt untouched. Furthermore, to ensure robustness against malicious removal, we optimize cover tokens, i.e., critical tokens in the original prompt, to ensure that removing auxiliary tokens causes severe performance degradation. Experimental results show that promptCOS achieves high effectiveness (99.3% average watermark similarity), strong distinctiveness (60.8% higher than the best baseline), high fidelity (accuracy degradation no greater than 0.6%), robustness (resilience against four potential attack categories), and high computational efficiency (up to 98.1% cost saving).

cs.CR↗

LoopTrap: Termination Poisoning Attacks on LLM Agents

Modern LLM agents solve complex tasks by operating in iterative execution loops, where they repeatedly reason, act, and self-evaluate progress to determine when a task is complete. In this work, we show that while this self-directed loop facilitates autonomy, it also introduces a critical risk: by injecting malicious prompts into the agent's context, an adversary can distort the agent's termination judgment, making it believe the task remains incomplete and leading to unbounded computation.To understand this threat, we define and systematically characterize it as Termination Poisoning and design 10 representative attack strategies. Through a empirical study spanning 8 LLM agents and 60 tasks, we demonstrate that different LLM agents exhibit distinct behavioral signatures that determine which strategies succeed. These transferable patterns can serve as principled guidance for crafting effective attacks against previously unseen agents and tasks, enabling scalable red-teaming beyond manually designed templates. Building on these insights, we introduce LoopTrap, an automated red-teaming framework that synthesizes target-specific malicious prompts by exploiting agent behavioral tendencies. LoopTrap first constructs a behavioral profile of the target agent along four vulnerability dimensions via lightweight probing. It then performs adaptive trap synthesis, routing to the most effective strategy and selecting optimal injections via a self-scoring mechanism. Finally, successful traps are abstracted into a reusable skill library, while failed attempts are refined through self-reflection, ensuring continuous improvement. Extensive evaluation shows that LoopTrap achieves an average of 3.57$\times$ step amplification across 8 mainstream agents, with a peak of 25$\times$.

cs.CR↗

APPSI-139: A Parallel Corpus of English Application Privacy Policy Summarization and Interpretation

Privacy policies are essential for users to understand how service providers handle their personal data. However, these documents are often long and complex, as well as filled with technobabble and legalese, causing users to unknowingly accept terms that may even contradict the law. While summarizing and interpreting these privacy policies is crucial, there is a lack of high-quality English parallel corpus optimized for legal clarity and readability. To address this issue, we introduce APPSI-139, a high-quality English privacy policy corpus meticulously annotated by domain experts, specifically designed for summarization and interpretation tasks. The corpus includes 139 English privacy policies, 15,692 rewritten parallel corpora, and 36,351 fine-grained annotation labels across 11 data practice categories. Concurrently, we propose TCSI-pp-V2, a hybrid privacy policy summarization and interpretation framework that employs an alternating training strategy and coordinates multiple expert modules to effectively balance computational efficiency and accuracy. Experimental results show that the hybrid summarization system built on APPSI-139 corpus and the TCSI-pp-V2 framework outperform large language models, such as GPT-4o and LLaMA-3-70B, in terms of readability and reliability. The source code and dataset are available at https://github.com/EnlightenedAI/APPSI-139.

cs.CL↗

DualSentinel: A Lightweight Framework for Detecting Targeted Attacks in Black-box LLM via Dual Entropy Lull Pattern

Recent intelligent systems integrate powerful Large Language Models (LLMs) through APIs, but their trustworthiness may be critically undermined by targeted attacks like backdoor and prompt injection attacks, which secretly force LLMs to generate specific malicious sequences. Existing defensive approaches for such threats typically rely on high access rights, impose prohibitive costs, and hinder normal inference, rendering them impractical for real-world scenarios. To solve these limitations, we introduce DualSentinel, a lightweight and unified defense framework that can accurately and promptly detect the activation of targeted attacks alongside the LLM generation process. We first identify a characteristic of compromised LLMs, termed Entropy Lull: when a targeted attack successfully hijacks the generation process, the LLM exhibits a distinct period of abnormally low and stable token probability entropy, indicating it is following a fixed path rather than making creative choices. DualSentinel leverages this pattern by developing an innovative dual-check approach. It first employs a magnitude and trend-aware monitoring method to proactively and sensitively flag an entropy lull pattern at runtime. Upon such flagging, it triggers a lightweight yet powerful secondary verification based on task-flipping. An attack is confirmed only if the entropy lull pattern persists across both the original and the flipped task, proving that the LLM's output is coercively controlled. Extensive evaluations show that DualSentinel is both highly effective (superior detection accuracy with near-zero false positives) and remarkably efficient (negligible additional cost), offering a truly practical path toward securing deployed LLMs. The source code can be accessed at https://doi.org/10.5281/zenodo.18479273.

cs.CR↗