arXiv subjects

Tianwei Zhang

Publications and source records attributed to Tianwei Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2026

Physical Prompt Injection Attacks on Large Vision-Language Models

Large Vision-Language Models (LVLMs) are increasingly deployed in real-world intelligent systems for perception and reasoning in open physical environments. While LVLMs are known to be vulnerable to prompt injection attacks, existing methods either require access to input channels or depend on knowledge of user queries, assumptions that rarely hold in practical deployments. We propose the first Physical Prompt Injection Attack (PPIA), a black-box, query-agnostic attack that embeds malicious typographic instructions into physical objects perceivable by the LVLM. PPIA requires no access to the model, its inputs, or internal pipeline, and operates solely through visual observation. It combines offline selection of highly recognizable and semantically effective visual prompts with strategic environment-aware placement guided by spatiotemporal attention, ensuring that the injected prompts are both perceivable and influential on model behavior. We evaluate PPIA across 10 state-of-the-art LVLMs in both simulated and real-world settings on tasks including visual question answering, planning, and navigation, PPIA achieves attack success rates up to 98%, with strong robustness under varying physical conditions such as distance, viewpoint, and illumination. Our code is publicly available at https://github.com/2023cghacker/Physical-Prompt-Injection-Attack.

cs.CV↗

arXiv2026

Towards Effective Prompt Stealing Attack against Text-to-Image Diffusion Models

Text-to-Image (T2I) models, represented by DALL$\cdot$E and Midjourney, have gained huge popularity for creating realistic images. The quality of these images relies on the carefully engineered prompts, which have become valuable intellectual property. While skilled prompters showcase their AI-generated art on markets to attract buyers, this business incidentally exposes them to \textit{prompt stealing attacks}. Existing state-of-the-art attack techniques reconstruct the prompts from a fixed set of modifiers (i.e., style descriptions) with model-specific training, which exhibit restricted adaptability and effectiveness to diverse showcases (i.e., target images) and diffusion models. To alleviate these limitations, we propose Prometheus, a training-free, proxy-in-the-loop, search-based prompt-stealing attack, which reverse-engineers the valuable prompts of the showcases by interacting with a local proxy model. It consists of three innovative designs. First, we introduce dynamic modifiers, as a supplement to static modifiers used in prior works. These dynamic modifiers provide more details specific to the showcases, and we exploit NLP analysis to generate them on the fly. Second, we design a contextual matching algorithm to sort both dynamic and static modifiers. This offline process helps reduce the search space of the subsequent step. Third, we interact with a local proxy model to invert the prompts with a greedy search algorithm. Based on the feedback guidance, we refine the prompt to achieve higher fidelity. The evaluation results show that Prometheus successfully extracts prompts from popular platforms like PromptBase and AIFrog against diverse victim models, including Midjourney, Leonardo.ai, and DALL$\cdot$E, with an ASR improvement of 25.0\%. We also validate that Prometheus is resistant to extensive potential defenses, further highlighting its severity in practice.

cs.CR↗

arXiv2026

Spectuner-D1: Spectral Line Fitting of Interstellar Molecules Using Deep Reinforcement Learning

Spectral lines from interstellar molecules provide crucial insights into the physical and chemical conditions of the interstellar medium. Traditional spectral line analysis relies heavily on manual intervention, which becomes impractical when handling the massive datasets produced by modern facilities like ALMA. To address this challenge, we introduce a novel deep reinforcement learning framework to automate spectral line fitting. Using observational data from ALMA, we train a neural network that maps both molecular spectroscopic data and observed spectra to physical parameters such as excitation temperature and column density. The neural network predictions can serve as initial estimates and be further refined using a local optimizer. Our method achieves consistent fitting results compared to global optimization with multiple runs, while reducing the number of forward modeling runs by an order of magnitude. We apply our method to pixel-level fitting for an observation of the G327.3-0.6 hot core and validate our results using XCLASS. We perform the fitting for typical complex organic molecules of hot cores, including CH$_3$OH, CH$_3$OCHO, CH$_3$OCH$_3$, C$_2$H$_5$CN, and C$_2$H$_3$CN. For a 100 $\times$ 100 region covering 5 GHz bandwidth, the fitting process requires 4.9 to 41.9 minutes using a desktop with 16 cores and one consumer-grade GPU card.

astro-ph.GA↗

arXiv2026

Beyond Retrieval: Improving Evidence Quality for LLM-based Multimodal Fact-Checking

The increasing multimodal disinformation, where deceptive claims are reinforced through coordinated text and visual content, poses significant challenges to automated fact-checking. Recent efforts leverage Large Language Models (LLMs) for this task, capitalizing on their strong reasoning and multimodal understanding capabilities. Emerging retrieval-augmented frameworks further equip LLMs with access to open-domain external information, enabling evidence-based verification beyond their internal knowledge. Despite their promising gains, our empirical study reveals notable shortcomings in the external search coverage and evidence quality evaluation. To mitigate those limitations, we propose Aletheia, an end-to-end framework for automated multimodal fact-checking. It introduces a novel evidence retrieval strategy that improves evidence coverage and filters useless information from open-domain sources, enabling the extraction of high-quality evidence for verification. Extensive experiments demonstrate that Aletheia achieves an accuracy of 88.3% on two public multimodal disinformation datasets and 90.2% on newly emerging claims. Compared with existing evidence retrieval strategies, our approach improves verification accuracy by up to 30.8%, highlighting the critical role of evidence quality in LLM-based disinformation verification.

cs.AI↗

arXiv2025

Prompt Injection attack against LLM-integrated Applications

Large Language Models (LLMs), renowned for their superior proficiency in language comprehension and generation, stimulate a vibrant ecosystem of applications around them. However, their extensive assimilation into various services introduces significant security risks. This study deconstructs the complexities and implications of prompt injection attacks on actual LLM-integrated applications. Initially, we conduct an exploratory analysis on ten commercial applications, highlighting the constraints of current attack strategies in practice. Prompted by these limitations, we subsequently formulate HouYi, a novel black-box prompt injection attack technique, which draws inspiration from traditional web injection attacks. HouYi is compartmentalized into three crucial elements: a seamlessly-incorporated pre-constructed prompt, an injection prompt inducing context partition, and a malicious payload designed to fulfill the attack objectives. Leveraging HouYi, we unveil previously unknown and severe attack outcomes, such as unrestricted arbitrary LLM usage and uncomplicated application prompt theft. We deploy HouYi on 36 actual LLM-integrated applications and discern 31 applications susceptible to prompt injection. 10 vendors have validated our discoveries, including Notion, which has the potential to impact millions of users. Our investigation illuminates both the possible risks of prompt injection attacks and the possible tactics for mitigation.

cs.CR↗

arXiv2025

Holmes: Towards Effective and Harmless Model Ownership Verification to Personalized Large Vision Models via Decoupling Common Features

Large vision models (LVMs) achieve remarkable performance in various downstream tasks, primarily by personalizing pre-trained models through fine-tuning with private and valuable local data, which makes the personalized model a valuable intellectual property. Similar to the era of traditional DNNs, model stealing attacks also pose significant risks to LVMs. However, this paper reveals that most existing defense methods (developed for traditional DNNs), typically designed for models trained from scratch, either introduce additional security risks, are prone to misjudgment, or are even ineffective for fine-tuned models. To alleviate these problems, this paper proposes a harmless model ownership verification method for personalized LVMs by decoupling similar common features. In general, our method consists of three main stages. In the first stage, we create shadow models that retain common features of the victim model while disrupting dataset-specific features. We represent the dataset-specific features of the victim model by computing the output differences between the shadow and victim models, without altering the victim model or its training process. After that, a meta-classifier is trained to identify stolen models by determining whether suspicious models contain the dataset-specific features of the victim. In the third stage, we conduct model ownership verification by hypothesis test to mitigate randomness and enhance robustness. Extensive experiments on benchmark datasets verify the effectiveness of the proposed method in detecting different types of model stealing simultaneously. Our codes are available at https://github.com/zlh-thu/Holmes.

cs.CV↗

arXiv2025

PentestEval: Benchmarking LLM-based Penetration Testing with Modular and Stage-Level Design

Penetration testing is essential for assessing and strengthening system security against real-world threats, yet traditional workflows remain highly manual, expertise-intensive, and difficult to scale. Although recent advances in Large Language Models (LLMs) offer promising opportunities for automation, existing applications rely on simplistic prompting without task decomposition or domain adaptation, resulting in unreliable black-box behavior and limited insight into model capabilities across penetration testing stages. To address this gap, we introduce PentestEval, the first comprehensive benchmark for evaluating LLMs across six decomposed penetration testing stages: Information Collection, Weakness Gathering and Filtering, Attack Decision-Making, Exploit Generation and Revision. PentestEval integrates expert-annotated ground truth with a fully automated evaluation pipeline across 346 tasks covering all stages in 12 realistic vulnerable scenarios. Our stage-level evaluation of 9 widely used LLMs reveals generally weak performance and distinct limitations across the stages of penetration-testing workflow. End-to-end pipelines reach only 31% success rate, and existing LLM-powered systems such as PentestGPT, PentestAgent, and VulnBot exhibit similar limitations, with autonomous agents failing almost entirely. These findings highlight that autonomous penetration testing demands stronger structured reasoning, where modularization enhances each individual stage and improves overall performance. PentestEval provides the foundational benchmark needed for future research on fine-grained, stage-level evaluation, paving the way toward more reliable LLM-based automation.

cs.SE↗

arXiv2025

DREAM: Scalable Red Teaming for Text-to-Image Generative Systems via Distribution Modeling

Despite the integration of safety alignment and external filters, text-to-image (T2I) generative systems are still susceptible to producing harmful content, such as sexual or violent imagery. This raises serious concerns about unintended exposure and potential misuse. Red teaming, which aims to proactively identify diverse prompts that can elicit unsafe outputs from the T2I system, is increasingly recognized as an essential method for assessing and improving safety before real-world deployment. However, existing automated red teaming approaches often treat prompt discovery as an isolated, prompt-level optimization task, which limits their scalability, diversity, and overall effectiveness. To bridge this gap, in this paper, we propose DREAM, a scalable red teaming framework to automatically uncover diverse problematic prompts from a given T2I system. Unlike prior work that optimizes prompts individually, DREAM directly models the probabilistic distribution of the target system's problematic prompts, which enables explicit optimization over both effectiveness and diversity, and allows efficient large-scale sampling after training. To achieve this without direct access to representative training samples, we draw inspiration from energy-based models and reformulate the objective into a simple and tractable form. We further introduce GC-SPSA, an efficient optimization algorithm that provides stable gradient estimates through the long and potentially non-differentiable T2I pipeline. During inference, we also propose a diversity-aware sampling strategy to enhance prompt variety. The effectiveness of DREAM is validated through extensive experiments, demonstrating state-of-the-art performance across a wide range of T2I models and safety filters in terms of both prompt success rate and diversity. Our code is available at https://github.com/AntigoneRandy/DREAM

cs.CR↗

arXiv2025

Towards Resilient Safety-driven Unlearning for Diffusion Models against Downstream Fine-tuning

Text-to-image (T2I) diffusion models have achieved impressive image generation quality and are increasingly fine-tuned for personalized applications. However, these models often inherit unsafe behaviors from toxic pretraining data, raising growing safety concerns. While recent safety-driven unlearning methods have made promising progress in suppressing model toxicity, they are found to be fragile to downstream fine-tuning, as we reveal that state-of-the-art methods largely fail to retain their effectiveness even when fine-tuned on entirely benign datasets. To mitigate this problem, in this paper, we propose ResAlign, a safety-driven unlearning framework with enhanced resilience against downstream fine-tuning. By modeling downstream fine-tuning as an implicit optimization problem with a Moreau envelope-based reformulation, ResAlign enables efficient gradient estimation to minimize the recovery of harmful behaviors. Additionally, a meta-learning strategy is proposed to simulate a diverse distribution of fine-tuning scenarios to improve generalization. Extensive experiments across a wide range of datasets, fine-tuning methods, and configurations demonstrate that ResAlign consistently outperforms prior unlearning approaches in retaining safety, while effectively preserving benign generation capability. Our code and pretrained models are publicly available at https://github.com/AntigoneRandy/ResAlign.

cs.LG↗

arXiv2025

Visible Yet Unreadable: A Systematic Blind Spot of Vision Language Models Across Writing Systems

Writing is a universal cultural technology that reuses vision for symbolic communication. Humans display striking resilience: we readily recognize words even when characters are fragmented, fused, or partially occluded. This paper investigates whether advanced vision language models (VLMs) share this resilience. We construct two psychophysics inspired benchmarks across distinct writing systems, Chinese logographs and English alphabetic words, by splicing, recombining, and overlaying glyphs to yield ''visible but unreadable'' stimuli for models while remaining legible to humans. Despite strong performance on clean text, contemporary VLMs show a severe drop under these perturbations, frequently producing unrelated or incoherent outputs. The pattern suggests a structural limitation: models heavily leverage generic visual invariances but under rely on compositional priors needed for robust literacy. We release stimuli generation code, prompts, and evaluation protocols to facilitate transparent replication and follow up work. Our findings motivate architectures and training strategies that encode symbol segmentation, composition, and binding across scripts, and they delineate concrete challenges for deploying multimodal systems in education, accessibility, cultural heritage, and security.

cs.CV↗

arXiv2025

IRCopilot: Automated Incident Response with Large Language Models

Incident response plays a pivotal role in mitigating the impact of cyber attacks. In recent years, the intensity and complexity of global cyber threats have grown significantly, making it increasingly challenging for traditional threat detection and incident response methods to operate effectively in complex network environments. While Large Language Models (LLMs) have shown great potential in early threat detection, their capabilities remain limited when it comes to automated incident response after an intrusion. To address this gap, we construct an incremental benchmark based on real-world incident response tasks to thoroughly evaluate the performance of LLMs in this domain. Our analysis reveals several key challenges that hinder the practical application of contemporary LLMs, including context loss, hallucinations, privacy protection concerns, and their limited ability to provide accurate, context-specific recommendations. In response to these challenges, we propose IRCopilot, a novel framework for automated incident response powered by LLMs. IRCopilot mimics the three dynamic phases of a real-world incident response team using four collaborative LLM-based session components. These components are designed with clear divisions of responsibility, reducing issues such as hallucinations and context loss. Our method leverages diverse prompt designs and strategic responsibility segmentation, significantly improving the system's practicality and efficiency. Experimental results demonstrate that IRCopilot outperforms baseline LLMs across key benchmarks, achieving sub-task completion rates of 150%, 138%, 136%, 119%, and 114% for various response tasks. Moreover, IRCopilot exhibits robust performance on public incident response platforms and in real-world attack scenarios, showcasing its strong applicability.

cs.CR↗

arXiv2025

ReSpec: Towards Optimizing Speculative Decoding in Reinforcement Learning Systems

Adapting large language models (LLMs) via reinforcement learning (RL) is often bottlenecked by the generation stage, which can consume over 75\% of the training time. Speculative decoding (SD) accelerates autoregressive generation in serving systems, but its behavior under RL training remains largely unexplored. We identify three critical gaps that hinder the naive integration of SD into RL systems: diminishing speedups at large batch sizes, drafter staleness under continual actor updates, and drafter-induced policy degradation. To address these gaps, we present ReSpec, a system that adapts SD to RL through three complementary mechanisms: dynamically tuning SD configurations, evolving the drafter via knowledge distillation, and weighting updates by rollout rewards. On Qwen models (3B--14B), ReSpec achieves up to 4.5x speedup while preserving reward convergence and training stability, providing a practical solution for efficient RL-based LLM adaptation.

cs.LG↗

arXiv2025

FaceTracer: Unveiling Source Identities from Swapped Face Images and Videos for Fraud Prevention

Face-swapping techniques have advanced rapidly with the evolution of deep learning, leading to widespread use and growing concerns about potential misuse, especially in cases of fraud. While many efforts have focused on detecting swapped face images or videos, these methods are insufficient for tracing the malicious users behind fraudulent activities. Intrusive watermark-based approaches also fail to trace unmarked identities, limiting their practical utility. To address these challenges, we introduce FaceTracer, the first non-intrusive framework specifically designed to trace the identity of the source person from swapped face images or videos. Specifically, FaceTracer leverages a disentanglement module that effectively suppresses identity information related to the target person while isolating the identity features of the source person. This allows us to extract robust identity information that can directly link the swapped face back to the original individual, aiding in uncovering the actors behind fraudulent activities. Extensive experiments demonstrate FaceTracer's effectiveness across various face-swapping techniques, successfully identifying the source person in swapped content and enabling the tracing of malicious actors involved in fraudulent activities. Additionally, FaceTracer shows strong transferability to unseen face-swapping methods including commercial applications and robustness against transmission distortions and adaptive attacks.Our code is available at: https://github.com/zzy224/FaceTracer.

cs.CV↗

arXiv2025

CipherGuard: Compiler-aided Mitigation against Ciphertext Side-channel Attacks

Recently, the new ciphertext side channels resulting from the deterministic memory encryption in Trusted Execution Environments (TEEs), enable ciphertexts to manifest identifiable patterns when being sequentially written to the same memory address. Attackers with read access to encrypted memory in TEEs can potentially deduce plaintexts by analyzing these changing ciphertext patterns. In this paper, we design CipherGuard, a compiler-based mitigation tool to counteract ciphertext side channels with high efficiency and security guarantees. CipherGuard is based on the LLVM ecosystem, and encompasses multiple defense strategies, including software-assisted probabilistic encryption, secret-aware register allocation, and diversion-based obfuscation. The design of CipherGuard demonstrates that compiler techniques are highly effective for fine-grained control over mitigation code generation and assisted component management. Through a comprehensive evaluation, it demonstrates that CipherGuard can strengthen the security of various cryptographic implementations more efficiently than existing state-of-the-art defense, i.e., CipherFix. In its most efficient strategy, CipherGuard incurs an average performance overhead of only 1.41X, with a maximum of 1.95X.

cs.CR↗

arXiv2025

Mask Image Watermarking

We present MaskWM, a simple, efficient, and flexible framework for image watermarking. MaskWM has two variants: (1) MaskWM-D, which supports global watermark embedding, watermark localization, and local watermark extraction for applications such as tamper detection; (2) MaskWM-ED, which focuses on local watermark embedding and extraction, offering enhanced robustness in small regions to support fine-grined image protection. MaskWM-D builds on the classical encoder-distortion layer-decoder training paradigm. In MaskWM-D, we introduce a simple masking mechanism during the decoding stage that enables both global and local watermark extraction. During training, the decoder is guided by various types of masks applied to watermarked images before extraction, helping it learn to localize watermarks and extract them from the corresponding local areas. MaskWM-ED extends this design by incorporating the mask into the encoding stage as well, guiding the encoder to embed the watermark in designated local regions, which improves robustness under regional attacks. Extensive experiments show that MaskWM achieves state-of-the-art performance in global and local watermark extraction, watermark localization, and multi-watermark embedding. It outperforms all existing baselines, including the recent leading model WAM for local watermarking, while preserving high visual quality of the watermarked images. In addition, MaskWM is highly efficient and adaptable. It requires only 20 hours of training on a single A6000 GPU, achieving 15x computational efficiency compared to WAM. By simply adjusting the distortion layer, MaskWM can be quickly fine-tuned to meet varying robustness requirements.

cs.CV↗

arXiv2025

SafeGuider: Robust and Practical Content Safety Control for Text-to-Image Models

Text-to-image models have shown remarkable capabilities in generating high-quality images from natural language descriptions. However, these models are highly vulnerable to adversarial prompts, which can bypass safety measures and produce harmful content. Despite various defensive strategies, achieving robustness against attacks while maintaining practical utility in real-world applications remains a significant challenge. To address this issue, we first conduct an empirical study of the text encoder in the Stable Diffusion (SD) model, which is a widely used and representative text-to-image model. Our findings reveal that the [EOS] token acts as a semantic aggregator, exhibiting distinct distributional patterns between benign and adversarial prompts in its embedding space. Building on this insight, we introduce SafeGuider, a two-step framework designed for robust safety control without compromising generation quality. SafeGuider combines an embedding-level recognition model with a safety-aware feature erasure beam search algorithm. This integration enables the framework to maintain high-quality image generation for benign prompts while ensuring robust defense against both in-domain and out-of-domain attacks. SafeGuider demonstrates exceptional effectiveness in minimizing attack success rates, achieving a maximum rate of only 5.48\% across various attack scenarios. Moreover, instead of refusing to generate or producing black images for unsafe prompts, SafeGuider generates safe and meaningful images, enhancing its practical utility. In addition, SafeGuider is not limited to the SD model and can be effectively applied to other text-to-image models, such as the Flux model, demonstrating its versatility and adaptability across different architectures. We hope that SafeGuider can shed some light on the practical deployment of secure text-to-image systems.

cs.CR↗

arXiv2025

Instruction Tuning for Large Language Models: A Survey

This paper surveys research works in the quickly advancing field of instruction tuning (IT), which can also be referred to as supervised fine-tuning (SFT)\footnote{In this paper, unless specified otherwise, supervised fine-tuning (SFT) and instruction tuning (IT) are used interchangeably.}, a crucial technique to enhance the capabilities and controllability of large language models (LLMs). Instruction tuning refers to the process of further training LLMs on a dataset consisting of \textsc{(instruction, output)} pairs in a supervised fashion, which bridges the gap between the next-word prediction objective of LLMs and the users' objective of having LLMs adhere to human instructions. In this work, we make a systematic review of the literature, including the general methodology of SFT, the construction of SFT datasets, the training of SFT models, and applications to different modalities, domains and application, along with analysis on aspects that influence the outcome of SFT (e.g., generation of instruction outputs, size of the instruction dataset, etc). We also review the potential pitfalls of SFT along with criticism against it, along with efforts pointing out current deficiencies of existing strategies and suggest some avenues for fruitful research. Project Page: github.com/xiaoya-li/Instruction-Tuning-Survey

cs.CL↗

arXiv2025

Work Zones challenge VLM Trajectory Planning: Toward Mitigation and Robust Autonomous Driving

Visual Language Models (VLMs), with powerful multimodal reasoning capabilities, are gradually integrated into autonomous driving by several automobile manufacturers to enhance planning capability in challenging environments. However, the trajectory planning capability of VLMs in work zones, which often include irregular layouts, temporary traffic control, and dynamically changing geometric structures, is still unexplored. To bridge this gap, we conduct the \textit{first} systematic study of VLMs for work zone trajectory planning, revealing that mainstream VLMs fail to generate correct trajectories in $68.0%$ of cases. To better understand these failures, we first identify candidate patterns via subgraph mining and clustering analysis, and then confirm the validity of $8$ common failure patterns through human verification. Building on these findings, we propose REACT-Drive, a trajectory planning framework that integrates VLMs with Retrieval-Augmented Generation (RAG). Specifically, REACT-Drive leverages VLMs to convert prior failure cases into constraint rules and executable trajectory planning code, while RAG retrieves similar patterns in new scenarios to guide trajectory generation. Experimental results on the ROADWork dataset show that REACT-Drive yields a reduction of around $3\times$ in average displacement error relative to VLM baselines under evaluation with Qwen2.5-VL. In addition, REACT-Drive yields the lowest inference time ($0.58$s) compared with other methods such as fine-tuning ($17.90$s). We further conduct experiments using a real vehicle in 15 work zone scenarios in the physical world, demonstrating the strong practicality of REACT-Drive.

cs.RO↗