Haichang Gao

Zihui Wu, Haichang Gao, Jianping He et al.

Large language models (LLMs) have demonstrated remarkable capabilities, but their power comes with significant security considerations. While extensive research has been conducted on the safety of LLMs in chat mode, the security implications of their function calling feature have been largely overlooked. This paper uncovers a critical vulnerability in the function calling process of LLMs, introducing a novel "jailbreak function" attack method that exploits alignment discrepancies, user coercion, and the absence of rigorous safety filters. Our empirical study, conducted on six state-of-the-art LLMs including GPT-4o, Claude-3.5-Sonnet, and Gemini-1.5-pro, reveals an alarming average success rate of over 90\% for this attack. We provide a comprehensive analysis of why function calls are susceptible to such attacks and propose defensive strategies, including the use of defensive prompts. Our findings highlight the urgent need for enhanced security measures in the function calling capabilities of LLMs, contributing to the field of AI safety by identifying a previously unexplored risk, designing an effective attack method, and suggesting practical defensive measures. Our code is available at https://github.com/wooozihui/jailbreakfunction.

Yuhang Wang, Feiming Xu, Zheng Lin et al.

Although large language model (LLM)-based agents, exemplified by OpenClaw, are increasingly evolving from task-oriented systems into personalized AI assistants for solving complex real-world tasks, their practical deployment also introduces severe security risks. However, existing agent security research and evaluation frameworks primarily focus on synthetic or task-centric settings, and thus fail to accurately capture the attack surface and risk propagation mechanisms of personalized agents in real-world deployments. To address this gap, we propose Personalized Agent Security Bench (PASB), an end-to-end security evaluation framework tailored for real-world personalized agents. Building upon existing agent attack paradigms, PASB incorporates personalized usage scenarios, realistic toolchains, and long-horizon interactions, enabling black-box, end-to-end security evaluation on real systems. Using OpenClaw as a representative case study, we systematically evaluate its security across multiple personalized scenarios, tool capabilities, and attack types. Our results indicate that OpenClaw exhibits critical vulnerabilities at different execution stages, including user prompt processing, tool usage, and memory retrieval, highlighting substantial security risks in personalized agent deployments. The code for the proposed PASB framework is available at https://github.com/AstorYH/PASB.

Zheng Lin, Zhenxing Niu, Haoxuan Ji et al.

Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in solving complex problems by generating structured, step-by-step reasoning content. However, exposing a model's internal reasoning process introduces additional safety risks; for example, recent studies show that LRMs are more vulnerable to jailbreak attacks than standard LLMs. In this paper, we investigate jailbreak attacks on LRMs and reveal that the attack success rate (ASR) is closely correlated with LRMs' attention patterns. Specifically, successful jailbreaks tend to assign lower attention to harmful tokens in the input prompt, while allocating higher attention to those tokens in the reasoning content. Motivated by this finding, we propose a novel jailbreak method for LRMs that leverages reinforcement learning (RL) to enhance attack effectiveness, explicitly incorporating attention signals into the reward function design. In addition, we introduce diverse persuasion strategies to enrich the RL action space, which consistently improves the ASR. Extensive experiments on five open-source and closed-source LRMs across three benchmarks demonstrate that our method achieves substantially higher ASR, outperforming existing approaches in terms of effectiveness, efficiency, and transferability.

Shudong Zhang, Haichang Gao, Tianwei Zhang et al.

Adversarial training (AT) has proven to be one of the most effective ways to defend Deep Neural Networks (DNNs) against adversarial attacks. However, the phenomenon of robust overfitting, i.e., the robustness will drop sharply at a certain stage, always exists during AT. It is of great importance to decrease this robust generalization gap in order to obtain a robust model. In this paper, we present an in-depth study towards the robust overfitting from a new angle. We observe that consistency regularization, a popular technique in semi-supervised learning, has a similar goal as AT and can be used to alleviate robust overfitting. We empirically validate this observation, and find a majority of prior solutions have implicit connections to consistency regularization. Motivated by this, we introduce a new AT solution, which integrates the consistency regularization and Mean Teacher (MT) strategy into AT. Specifically, we introduce a teacher model, coming from the average weights of the student models over the training steps. Then we design a consistency loss function to make the prediction distribution of the student models over adversarial examples consistent with that of the teacher model over clean samples. Experiments show that our proposed method can effectively alleviate robust overfitting and improve the robustness of DNN models against common adversarial attacks.

Zihui Wu, Haichang Gao, Bingqian Zhou et al.

In this paper, we investigate on improving the adversarial robustness obtained in adversarial training (AT) via reducing the difficulty of optimization. To better study this problem, we build a novel Bregman divergence perspective for AT, in which AT can be viewed as the sliding process of the training data points on the negative entropy curve. Based on this perspective, we analyze the learning objectives of two typical AT methods, i.e., PGD-AT and TRADES, and we find that the optimization process of TRADES is easier than PGD-AT for that TRADES separates PGD-AT. In addition, we discuss the function of entropy in TRADES, and we find that models with high entropy can be better robustness learners. Inspired by the above findings, we propose two methods, i.e., FAIT and MER, which can both not only reduce the difficulty of optimization under the 10-step PGD adversaries, but also provide better robustness. Our work suggests that reducing the difficulty of optimization under the 10-step PGD adversaries is a promising approach for enhancing the adversarial robustness in AT.

Yuhang Wang, Zhenxing Niu, Haoxuan Ji et al.

The core challenge of machine unlearning is to strike a balance between target knowledge removal and non-target knowledge retention. In the context of Multimodal Large Language Models (MLLMs), this challenge becomes even more pronounced, as knowledge is further divided into visual and textual modalities that are tightly intertwined. In this paper, we introduce an MLLM unlearning approach that aims to forget target visual knowledge while preserving non-target visual knowledge and all textual knowledge. Specifically, we freeze the LLM backbone and achieve unlearning by fine-tuning the visual module. First, we propose a Contrastive Visual Forgetting (CVF) mechanism to separate target visual knowledge from retained visual knowledge, guiding the representations of target visual concepts toward appropriate regions in the feature space. Second, we identify the null space associated with retained knowledge and constrain the unlearning process within this space, thereby significantly mitigating degradation in knowledge retention. Third, beyond static unlearning scenarios, we extend our approach to continual unlearning, where forgetting requests arrive sequentially. Extensive experiments across diverse benchmarks demonstrate that our approach achieves a strong balance between effective forgetting and robust knowledge retention.

Zihui Wu, Haichang Gao, Jiacheng Luo et al.

Large Language Models (LLMs) commonly rely on explicit refusal prefixes for safety, making them vulnerable to prefix injection attacks. We introduce HumorReject, a novel data-driven approach that reimagines LLM safety by decoupling it from refusal prefixes through humor as an indirect refusal strategy. Rather than explicitly rejecting harmful instructions, HumorReject responds with contextually appropriate humor that naturally defuses potentially dangerous requests. Our approach effectively addresses common "over-defense" issues while demonstrating superior robustness against various attack vectors. Our findings suggest that improvements in training data design can be as important as the alignment algorithm itself in achieving effective LLM safety. The code and dataset are available at https://github.com/wooozihui/HumorReject.

Zihui Wu, Haichang Gao, Ping Wang et al.

Glitch tokens, inputs that trigger unpredictable or anomalous behavior in Large Language Models (LLMs), pose significant challenges to model reliability and safety. Existing detection methods primarily rely on heuristic embedding patterns or statistical anomalies within internal representations, limiting their generalizability across different model architectures and potentially missing anomalies that deviate from observed patterns. We introduce GlitchMiner, an behavior-driven framework designed to identify glitch tokens by maximizing predictive entropy. Leveraging a gradient-guided local search strategy, GlitchMiner efficiently explores the discrete token space without relying on model-specific heuristics or large-batch sampling. Extensive experiments across ten LLMs from five major model families demonstrate that GlitchMiner consistently outperforms existing approaches in detection accuracy and query efficiency, providing a generalizable and scalable solution for effective glitch token discovery. Code is available at [https://github.com/wooozihu/GlitchMiner]

Zheng Lin, Zhenxing Niu, Haoxuan Ji et al.

This paper proposes a guaranteed defense method for large language models (LLMs) to safeguard against jailbreaking attacks. Drawing inspiration from the denoised-smoothing approach in the adversarial defense domain, we propose a novel smoothing-based defense method, termed Disrupt-and-Rectify Smoothing (DR-Smoothing). Specifically, we integrate a two-stage prompt processing scheme-first disrupting the input prompt, then rectifying it-into the conventional smoothing defense framework. This disrupt-and-rectify approach improves upon previous disrupt-only approaches by restoring out-of-distribution disrupted prompts to an in-distribution form, thereby reducing the risk of unpredictable LLM behavior. In addition, this two-stage scheme offers a distinct advantage in striking a balance between harmlessness and helpfulness in jailbreaking defense. Notably, we present a theoretical analysis for generic smoothing framework, offering a tight bound for the defense success probability and the requirements on the disruption strength. Our approach can defend against both token-level and prompt-level jailbreaking attacks, under both established and adaptive attacking scenarios. Extensive experiments demonstrate that our approach surpasses current state-of-the-art defense methods in terms of both harmlessness and helpfulness.

Zheng Lin, Zhenxing Niu, Haoxuan Ji et al.

This paper proposes a jailbreaking prompt detection method for large language models (LLMs) to defend against jailbreak attacks. Although recent LLMs are equipped with built-in safeguards, it remains possible to craft jailbreaking prompts that bypass them. We argue that such jailbreaking prompts are inherently fragile, and thus introduce an embedding disruption method to re-activate the safeguards within LLMs. Unlike previous defense methods that aim to serve as standalone solutions, our approach instead cooperates with the LLM's internal defense mechanisms by re-triggering them. Moreover, through extensive analysis, we gain a comprehensive understanding of the disruption effects and develop an efficient search algorithm to identify appropriate disruptions for effective jailbreak detection. Extensive experiments demonstrate that our approach effectively defends against state-of-the-art jailbreak attacks in white-box and black-box settings, and remains robust even against adaptive attacks.

Zihui Wu, Haichang Gao, Bingqian Zhou et al.

\emph{Consistent teaching} is an effective paradigm for implementing knowledge distillation (KD), where both student and teacher models receive identical inputs, and KD is treated as a function matching task (FunMatch). However, one limitation of FunMatch is that it does not account for the transfer of adversarial robustness, a model's resistance to adversarial attacks. To tackle this problem, we propose a simple but effective strategy called Adversarial Function Matching (AdvFunMatch), which aims to match distributions for all data points within the $\ell_p$-norm ball of the training data, in accordance with consistent teaching. Formulated as a min-max optimization problem, AdvFunMatch identifies the worst-case instances that maximizes the KL-divergence between teacher and student model outputs, which we refer to as "mismatched examples," and then matches the outputs on these mismatched examples. Our experimental results show that AdvFunMatch effectively produces student models with both high clean accuracy and robustness. Furthermore, we reveal that strong data augmentations (\emph{e.g.}, AutoAugment) are beneficial in AdvFunMatch, whereas prior works have found them less effective in adversarial training. Code is available at \url{https://gitee.com/zihui998/adv-fun-match}.

Yuhang Wang, Wenjie Mei, Junkai Zhang et al.

Although Multimodal Large Language Models (MLLMs) have achieved remarkable progress across many domains, their training on large-scale multimodal datasets raises serious privacy concerns, making effective machine unlearning increasingly necessary. However, existing benchmarks mainly focus on static or short-sequence settings, offering limited support for evaluating continual privacy deletion requests in realistic deployments. To bridge this gap, we introduce ICU-Bench, a continual multimodal unlearning benchmark built on privacy-critical document data. ICU-Bench contains 1,000 privacy-sensitive profiles from two document domains, medical reports and labor contracts, with 9,500 images, 16,000 question-answer pairs, and 100 forget tasks. Additionally, new continual unlearning metrics are introduced, facilitating a comprehensive analysis of forgetting effectiveness, historical forgetting preservation, retained utility, and stability throughout the continual unlearning process. Through extensive experiments with representative unlearning methods on ICU-Bench, we show that existing methods generally struggle in continual settings and exhibit clear limitations in balancing forgetting quality, utility preservation, and scalability over long task sequences. These findings highlight the need for multimodal unlearning methods explicitly designed for continual privacy deletion.

Yuhang Wang, Zhenxing Niu, Haoxuan Ji et al.

Recently, machine unlearning approaches have been proposed to remove sensitive information from well-trained large models. However, most existing methods are tailored for LLMs, while MLLM-oriented unlearning remains at its early stage. Inspired by recent studies exploring the internal mechanisms of MLLMs, we propose to disentangle the visual and textual knowledge embedded within MLLMs and introduce a dedicated approach to selectively erase target visual knowledge while preserving textual knowledge. Unlike previous unlearning methods that rely on output-level supervision, our approach introduces a Visual Knowledge Distillation (VKD) scheme, which leverages intermediate visual representations within the MLLM as supervision signals. This design substantially enhances both unlearning effectiveness and model utility. Moreover, since our method only fine-tunes the visual components of the MLLM, it offers significant efficiency advantages. Extensive experiments demonstrate that our approach outperforms state-of-the-art unlearning methods in terms of both effectiveness and efficiency. Moreover, we are the first to evaluate the robustness of MLLM unlearning against relearning attacks.

Haoxuan Ji, Zheng Lin, Yuyao Sun et al.

Modern privacy regulations have spurred the evolution of machine unlearning, a technique that enables the removal of data from an already trained ML model without requiring retraining from scratch. Previous unlearning methods tend to induce the model to achieve lowest classification accuracy on the removal data. Nonetheless, the authentic objective of machine unlearning is to align the unlearned model with the gold model, i.e., achieving the same classification accuracy as the gold model. For this purpose, we present a Twin Machine Unlearning (TMU) approach, where a twin unlearning problem is defined corresponding to the original unlearning problem. As a results, the generalization-label predictor trained on the twin problem can be transferred to the original problem, facilitating aligned data removal. Comprehensive empirical experiments illustrate that our approach significantly enhances the alignment between the unlearned model and the gold model. Meanwhile, our method allows data removal without compromising the model accuracy.

Yuhang Wang, Haichang Gao, Zhenxing Niu et al.

Tool-augmented AI agents substantially extend the practical capabilities of large language models, but they also introduce security risks that cannot be identified through model-only evaluation. In this paper, we present a systematic security assessment of six representative OpenClaw-series agent frameworks, namely OpenClaw, AutoClaw, QClaw, KimiClaw, MaxClaw, and ArkClaw, under multiple backbone models. To support this study, we construct a benchmark of 205 test cases covering representative attack behaviors across the full agent execution lifecycle, enabling unified evaluation of risk exposure at both the framework and model levels. Our results show that all evaluated agents exhibit substantial security vulnerabilities, and that agentized systems are significantly riskier than their underlying models used in isolation. In particular, reconnaissance and discovery behaviors emerge as the most common weaknesses, while different frameworks expose distinct high-risk profiles, including credential leakage, lateral movement, privilege escalation, and resource development. These findings indicate that the security of modern agent systems is shaped not only by the safety properties of the backbone model, but also by the coupling among model capability, tool use, multi-step planning, and runtime orchestration. We further show that once an agent is granted execution capability and persistent runtime context, weaknesses arising in early stages can be amplified into concrete system-level failures. Overall, our study highlights the need to move beyond prompt-level safeguards toward lifecycle-wide security governance for intelligent agent frameworks.

Zhixuan Ma, Haichang Gao, Junxiang Huang et al.

Federated Learning enables collaborative training of a global model across multiple geographically dispersed clients without the need for data sharing. However, it is susceptible to inference attacks, particularly label inference attacks. Existing studies on label distribution inference exhibits sensitive to the specific settings of the victim client and typically underperforms under defensive strategies. In this study, we propose a novel label distribution inference attack that is stable and adaptable to various scenarios. Specifically, we estimate the size of the victim client's dataset and construct several virtual clients tailored to the victim client. We then quantify the temporal generalization of each class label for the virtual clients and utilize the variation in temporal generalization to train an inference model that predicts the label distribution proportions of the victim client. We validate our approach on multiple datasets, including MNIST, Fashion-MNIST, FER2013, and AG-News. The results demonstrate the superiority of our method compared to state-of-the-art techniques. Furthermore, our attack remains effective even under differential privacy defense mechanisms, underscoring its potential for real-world applications.

Haichang Gao, Xiuling Chang, Zhongjie Ren et al.

Graphical passwords have been proposed as an alternative to alphanumeric passwords with their advantages in usability and security. However, they still tend to follow predictable patterns that are easier for attackers to exploit, probably due to users' memory limitations. Various literatures show that baroque music has positive effects on human learning and memorizing. To alleviate users' memory burden, we investigate the novel idea of introducing baroque music to graphical password schemes (specifically DAS, PassPoints and Story) and conduct a laboratory study to see whether it is helpful. In a ten minutes short-term recall, we found that participants in all conditions had high recall success rates that were not statistically different from each other. After one week, the music group coped PassPoints passwords significantly better than the group without music. But there was no statistical difference between two groups in recalling DAS passwords or Story passwords. Further more, we found that the music group tended to set significantly more complicated PassPoints passwords but less complicated DAS passwords.

Haichang Gao, Zhongjie Ren, Xiuling Chang et al.

Graphical passwords have been demonstrated to be the possible alternatives to traditional alphanumeric passwords. However, they still tend to follow predictable patterns that are easier to attack. The crux of the problem is users' memory limitations. Users are the weakest link in password authentication mechanism. It shows that baroque music has positive effects on human memorizing and learning. We introduce baroque music to the PassPoints graphical password scheme and conduct a laboratory study in this paper. Results shown that there is no statistic difference between the music group and the control group without music in short-term recall experiments, both had high recall success rates. But in long-term recall, the music group performed significantly better. We also found that the music group tended to set significantly more complicated passwords, which are usually more resistant to dictionary and other guess attacks. But compared with the control group, the music group took more time to log in both in short-term and long-term tests. Besides, it appears that background music does not work in terms of hotspots.

Haichang Gao, Honggang Liu, Dan Yao et al.

CAPTCHAs are employed as a security measure to differentiate human users from bots. A new sound-based CAPTCHA is proposed in this paper, which exploits the gaps between human voice and synthetic voice rather than relays on the auditory perception of human. The user is required to read out a given sentence, which is selected randomly from a specified book. The generated audio file will be analyzed automatically to judge whether the user is a human or not. In this paper, the design of the new CAPTCHA, the analysis of the audio files, and the choice of the audio frame window function are described in detail. And also, some experiments are conducted to fix the critical threshold and the coefficients of three indicators to ensure the security. The proposed audio CAPTCHA is proved accessible to users. The user study has shown that the human success rate reaches approximately 97% and the pass rate of attack software using Microsoft SDK 5.1 is only 4%. The experiments also indicated that it could be solved by most human users in less than 14 seconds and the average time is only 7.8 seconds.

Haichang Gao, Zhongjie Ren, Xiuling Chang et al.

Shoulder-surfing is a known risk where an attacker can capture a password by direct observation or by recording the authentication session. Due to the visual interface, this problem has become exacerbated in graphical passwords. There have been some graphical schemes resistant or immune to shoulder-surfing, but they have significant usability drawbacks, usually in the time and effort to log in. In this paper, we propose and evaluate a new shoulder-surfing resistant scheme which has a desirable usability for PDAs. Our inspiration comes from the drawing input method in DAS and the association mnemonics in Story for sequence retrieval. The new scheme requires users to draw a curve across their password images orderly rather than click directly on them. The drawing input trick along with the complementary measures, such as erasing the drawing trace, displaying degraded images, and starting and ending with randomly designated images provide a good resistance to shouldersurfing. A preliminary user study showed that users were able to enter their passwords accurately and to remember them over time.

Liming Wang, Xiuling Chang, Zhongjie Ren et al.

Text-based password schemes have inherent security and usability problems, leading to the development of graphical password schemes. However, most of these alternate schemes are vulnerable to spyware attacks. We propose a new scheme, using CAPTCHA (Completely Automated Public Turing tests to tell Computers and Humans Apart) that retaining the advantages of graphical password schemes, while simultaneously raising the cost of adversaries by orders of magnitude. Furthermore, some primary experiments are conducted and the results indicate that the usability should be improved in the future work.

Xiyang Liu, Zhongjie Ren, Xiuling Chang et al.

The trend toward a highly mobile workforce and the ubiquity of graphical interfaces (such as the stylus and touch-screen) has enabled the emergence of graphical authentications in Personal Digital Assistants (PDAs) [1]. However, most of the current graphical password schemes are vulnerable to shoulder-surfing [2,3], a known risk where an attacker can capture a password by direct observation or by recording the authentication session. Several approaches have been developed to deal with this problem, but they have significant usability drawbacks, usually in the time and effort to log in, making them less suitable for authentication [4, 8]. For example, it is time-consuming for users to log in CHC [4] and there are complex text memory requirements in scheme proposed by Hong [5]. With respect to the scheme proposed by Weinshall [6], not only is it intricate to log in, but also the main claim of resisting shoulder-surfing is proven false [7]. In this paper, we introduce a new graphical password scheme which provides a good resistance to shouldersurfing and preserves a desirable usability.