arXiv subjects

Zhibo Wang

Publications and source records attributed to Zhibo Wang.

At least 19 recordsLinked to original sources

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

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↗

arXiv2026

RerouteGuard: Understanding and Mitigating Adversarial Risks for LLM Routing

Recent advancements in multi-model AI systems have leveraged LLM routers to reduce computational cost while maintaining response quality by assigning queries to the most appropriate model. However, as classifiers, LLM routers are vulnerable to novel adversarial attacks in the form of LLM rerouting, where adversaries prepend specially crafted triggers to user queries to manipulate routing decisions. Such attacks can lead to increased computational cost, degraded response quality, and even bypass safety guardrails, yet their security implications remain largely underexplored. In this work, we bridge this gap by systematizing LLM rerouting threats based on the adversary's objectives (i.e., cost escalation, quality hijacking, and safety bypass) and knowledge. Based on the threat taxonomy, we conduct a measurement study of real-world LLM routing systems against existing LLM rerouting attacks. The results reveal that existing routing systems are vulnerable to rerouting attacks, especially in the cost escalation scenario. We then characterize existing rerouting attacks using interpretability techniques, revealing that they exploit router decision boundaries through confounder gadgets that prepend queries to force misrouting. To mitigate these risks, we introduce RerouteGuard, a flexible and scalable guardrail framework for LLM rerouting. RerouteGuard filters adversarial rerouting prompts via dynamic embedding-based detection and adaptive thresholding. Extensive evaluations in three attack settings and four benchmarks demonstrate that RerouteGuard achieves over 99% detection accuracy against state-of-the-art rerouting attacks, while maintaining negligible impact on legitimate queries. The experimental results indicate that RerouteGuard offers a principled and practical solution for safeguarding multi-model AI systems against adversarial rerouting.

cs.CR↗

arXiv2026

R$^2$BD: A Reconstruction-Based Method for Generalizable and Efficient Detection of Fake Images

Recently, reconstruction-based methods have gained attention for AIGC image detection. These methods leverage pre-trained diffusion models to reconstruct inputs and measure residuals for distinguishing real from fake images. Their key advantage lies in reducing reliance on dataset-specific artifacts and improving generalization under distribution shifts. However, they are limited by significant inefficiency due to multi-step inversion and reconstruction, and their reliance on diffusion backbones further limits generalization to other generative paradigms such as GANs. In this paper, we propose a novel fake image detection framework, called R$^2$BD, built upon two key designs: (1) G-LDM, a unified reconstruction model that simulates the generation behaviors of VAEs, GANs, and diffusion models, thereby broadening the detection scope beyond prior diffusion-only approaches; and (2) a residual bias calculation module that distinguishes real and fake images in a single inference step, which is a significant efficiency improvement over existing methods that typically require 20$+$ steps. Extensive experiments on the benchmark from 10 public datasets demonstrate that R$^2$BD is over 22$\times$ faster than existing reconstruction-based methods while achieving superior detection accuracy. In cross-dataset evaluations, it outperforms state-of-the-art methods by an average of 13.87\%, showing strong efficiency and generalization across diverse generative methods. The code and dataset used for evaluation are available at https://github.com/QingyuLiu/RRBD.

cs.CV↗

arXiv2026

Attack-Resistant Watermarking for AIGC Image Forensics via Diffusion-based Semantic Deflection

Protecting the copyright of user-generated AI images is an emerging challenge as AIGC becomes pervasive in creative workflows. Existing watermarking methods (1) remain vulnerable to real-world adversarial threats, often forced to trade off between defenses against spoofing and removal attacks; and (2) cannot support semantic-level tamper localization. We introduce PAI, a training-free inherent watermarking framework for AIGC copyright protection, plug-and-play with diffusion-based AIGC services. PAI simultaneously provides three key functionalities: robust ownership verification, attack detection, and semantic-level tampering localization. Unlike existing inherent watermark methods that only embed watermarks at noise initialization of diffusion models, we design a novel key-conditioned deflection mechanism that subtly steers the denoising trajectory according to the user key. Such trajectory-level coupling further strengthens the semantic entanglement of identity and content, thereby further enhancing robustness against real-world threats. Moreover, we also provide a theoretical analysis proving that only the valid key can pass verification. Experiments across 12 attack methods show that PAI achieves 98.43\% verification accuracy, improving over SOTA methods by 37.25\% on average, and retains strong tampering localization performance even against advanced AIGC edits. Our code is available at https://github.com/QingyuLiu/PAI.

cs.CR↗

arXiv2026

TAP-ViTs: Task-Adaptive Pruning for On-Device Deployment of Vision Transformers

Vision Transformers (ViTs) have demonstrated strong performance across a wide range of vision tasks, yet their substantial computational and memory demands hinder efficient deployment on resource-constrained mobile and edge devices. Pruning has emerged as a promising direction for reducing ViT complexity. However, existing approaches either (i) produce a single pruned model shared across all devices, ignoring device heterogeneity, or (ii) rely on fine-tuning with device-local data, which is often infeasible due to limited on-device resources and strict privacy constraints. As a result, current methods fall short of enabling task-customized ViT pruning in privacy-preserving mobile computing settings. This paper introduces TAP-ViTs, a novel task-adaptive pruning framework that generates device-specific pruned ViT models without requiring access to any raw local data. Specifically, to infer device-level task characteristics under privacy constraints, we propose a Gaussian Mixture Model (GMM)-based metric dataset construction mechanism. Each device fits a lightweight GMM to approximate its private data distribution and uploads only the GMM parameters. Using these parameters, the cloud selects distribution-consistent samples from public data to construct a task-representative metric dataset for each device. Based on this proxy dataset, we further develop a dual-granularity importance evaluation-based pruning strategy that jointly measures composite neuron importance and adaptive layer importance, enabling fine-grained, task-aware pruning tailored to each device's computational budget. Extensive experiments across multiple ViT backbones and datasets demonstrate that TAP-ViTs consistently outperforms state-of-the-art pruning methods under comparable compression ratios.

cs.CV↗

arXiv2025

Avatar4D: Synthesizing Domain-Specific 4D Humans for Real-World Pose Estimation

We present Avatar4D, a real-world transferable pipeline for generating customizable synthetic human motion datasets tailored to domain-specific applications. Unlike prior works, which focus on general, everyday motions and offer limited flexibility, our approach provides fine-grained control over body pose, appearance, camera viewpoint, and environmental context, without requiring any manual annotations. To validate the impact of Avatar4D, we focus on sports, where domain-specific human actions and movement patterns pose unique challenges for motion understanding. In this setting, we introduce Syn2Sport, a large-scale synthetic dataset spanning sports, including baseball and ice hockey. Avatar4D features high-fidelity 4D (3D geometry over time) human motion sequences with varying player appearances rendered in diverse environments. We benchmark several state-of-the-art pose estimation models on Syn2Sport and demonstrate their effectiveness for supervised learning, zero-shot transfer to real-world data, and generalization across sports. Furthermore, we evaluate how closely the generated synthetic data aligns with real-world datasets in feature space. Our results highlight the potential of such systems to generate scalable, controllable, and transferable human datasets for diverse domain-specific tasks without relying on domain-specific real data.

cs.CV↗