Prompt injection attacks pose a critical threat to large language models (LLMs), enabling goal hijacking and data leakage.Prompt guard models, though effective in defense, suffer from over-defense-falsely flagging benign inputs as malicious due to trigger word bias.To address this issue, we introduce NotInject, an evaluation dataset that systematically measures over-defense across various prompt guard models.NotInject contains 339 benign samples enriched with trigger words common in prompt injection attacks, enabling fine-grained evaluation.Our results show that state-of-the-art models suffer from over-defense issues, with accuracy dropping close to random guessing levels (60%).To mitigate this, we propose PIGuard, a novel prompt guard model that incorporates a new training strategy, Mitigating Over-defense for Free (MOF), which significantly reduces the bias on trigger words.PIGuard demonstrates state-of-the-art performance on diverse benchmarks including Not-Inject, surpassing the existing best model by 30.4%, offering a robust and open-source solution for detecting prompt injection attacks.The code and datasets are released at https: //github.com/leolee99/PIGuard.

Weidi Luo, Shenghong Dai, Xiaogeng Liu, Suman Banerjee, Huan Sun, Muhao Chen, Chaowei Xiao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Jailbreak attacks serve as essential red-teaming tools, proactively assessing whether LLMs can behave responsibly and safely in adversarial environments. Despite diverse strategies (e.g., cipher, low-resource language, persuasions, and so on) that have been proposed and shown success, these strategies are still manually designed, limiting their scope and effectiveness as a red-teaming tool. In this paper, we propose AutoDAN-Turbo, a black-box jailbreak method that can automatically discover as many jailbreak strategies as possible from scratch, without any human intervention or predefined scopes (e.g., specified candidate strategies), and use them for red-teaming. As a result, AutoDAN-Turbo can significantly outperform baseline methods, achieving a 74.3% higher average attack success rate on public benchmarks. Notably, AutoDAN-Turbo achieves an 88.5 attack success rate on GPT-4-1106-turbo. In addition, AutoDAN-Turbo is a unified framework that can incorporate existing human-designed jailbreak strategies in a plug-and-play manner. By integrating human-designed strategies, AutoDAN-Turbo can even achieve a higher attack success rate of 93.4 on GPT-4-1106-turbo.

We introduce MuirBench, a comprehensive benchmark that focuses on robust multi-image understanding capabilities of multimodal LLMs. MuirBench consists of 12 diverse multi-image tasks (e.g., scene understanding, ordering) that involve 10 categories of multi-image relations (e.g., multiview, temporal relations). Comprising 11,264 images and 2,600 multiple-choice questions, MuirBench is created in a pairwise manner, where each standard instance is paired with an unanswerable variant that has minimal semantic differences, in order for a reliable assessment. Evaluated upon 20 recent multi-modal LLMs, our results reveal that even the best-performing models like GPT-4o and Gemini Pro find it challenging to solve MuirBench, achieving 68.0% and 49.3% in accuracy. Open-source multimodal LLMs trained on single images can hardly generalize to multi-image questions, hovering below 33.3% in accuracy. These results highlight the importance of MuirBench in encouraging the community to develop multimodal LLMs that can look beyond a single image, suggesting potential pathways for future improvements.

The powerful capabilities of LLMs stem from their rich training data and high-quality labeled datasets, making the training of strong LLMs a resource-intensive process, which elevates the importance of IP protection for such LLMs. Compared to gathering high-quality labeled data, directly sampling outputs from these fully trained LLMs as training data presents a more cost-effective approach. This practice—where a suspect model is fine-tuned using high-quality data derived from these LLMs, thereby gaining capabilities similar to the target model—can be seen as a form of IP infringement against the original LLM. In recent years, LLM watermarks have been proposed and used to detect whether a text is AI-generated. Intuitively, if data sampled from a watermarked LLM is used for training, the resulting model would also be influenced by this watermark. This raises the question: can we directly use such watermarks to detect IP infringement of LLMs? In this paper, we explore the potential of LLM watermarks for detecting model infringement. We find that there are two issues with direct detection: (1) The queries used to sample output from the suspect LLM have a significant impact on detectability. (2) The watermark that is easily learned by LLMs exhibits instability regarding the watermark's hash key during detection. To address these issues, we propose LIDet, a detection method that leverages available anchor LLMs to select suitable queries for sampling from the suspect LLM. Additionally, it adapts the detection threshold to mitigate detection failures caused by different hash keys. To demonstrate the effectiveness of this approach, we construct a challenging model set containing multiple suspect LLMs on which direct detection methods struggle to yield effective results. Our method achieves over 90\% accuracy in distinguishing between infringing and clean models, demonstrating the feasibility of using LLM watermarks to detect LLM IP infringement.

Large Language Models (LLMs) have demonstrated the ability to solve a wide range of practical tasks within multi-agent systems. However, existing human-designed multi-agent frameworks are typically limited to a small set of pre-defined scenarios, while current automated design methods suffer from several limitations, such as the lack of tool integration, dependence on external training data, and rigid communication structures. In this paper, we propose MetaAgent, a finite state machine based framework that can automatically generate a multi-agent system. Given a task description, MetaAgent will design a multi-agent system and polish it through an optimization algorithm. When the multi-agent system is deployed, the finite state machine will control the agent’s actions and the state transitions. To evaluate our framework, we conduct experiments on both text-based tasks and practical tasks. The results indicate that the generated multi-agent system surpasses other auto-designed methods and can achieve a comparable performance with the human-designed multi-agent system, which is optimized for those specific tasks.

Peiran Wang, Xiaogeng Liu, Chaowei Xiao. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Large Language Models (LLMs) are increasingly central to agentic systems due to their strong reasoning and planning capabilities. By interacting with external environments through predefined tools, these agents can carry out complex user tasks. Nonetheless, this interaction also introduces the risk of prompt injection attacks, where malicious inputs from external sources can mislead the agent’s behavior, potentially resulting in economic loss, privacy leakage, or system compromise. System-level defenses have recently shown promise by enforcing static or predefined policies, but they still face two key challenges: the ability to dynamically update security rules and the need for memory stream isolation. To address these challenges, we propose DRIFT, a Dynamic Rule-based Isolation Framework for Trustworthy agentic systems, which enforces both control- and data-level constraints. A Secure Planner first constructs a minimal function trajectory and a JSON-schema-style parameter checklist for each function node based on the user query. A Dynamic Validator then monitors deviations from the original plan, assessing whether changes comply with privilege limitations and the user's intent. Finally, an \textit{Injection Isolator} detects and masks any instructions that may conflict with the user query from the memory stream to mitigate long-term risks. We empirically validate the effectiveness of DRIFT on the AgentDojo and ASB benchmark, demonstrating its strong security performance while maintaining high utility across diverse models—showcasing both its robustness and adaptability. The code is released at https://github.com/SaFoLab-WISC/DRIFT.

The aligned Large Language Models (LLMs) are powerful language understanding and decision-making tools that are created through extensive alignment with human feedback. However, these large models remain susceptible to jailbreak attacks, where adversaries manipulate prompts to elicit malicious outputs that should not be given by aligned LLMs. Investigating jailbreak prompts can lead us to delve into the limitations of LLMs and further guide us to secure them. Unfortunately, existing jailbreak techniques suffer from either (1) scalability issues, where attacks heavily rely on manual crafting of prompts, or (2) stealthiness problems, as attacks depend on token-based algorithms to generate prompts that are often semantically meaningless, making them susceptible to detection through basic perplexity testing. In light of these challenges, we intend to answer this question: Can we develop an approach that can automatically generate stealthy jailbreak prompts? In this paper, we introduce AutoDAN, a novel jailbreak attack against aligned LLMs. AutoDAN can automatically generate stealthy jailbreak prompts by the carefully designed hierarchical genetic algorithm. Extensive evaluations demonstrate that AutoDAN not only automates the process while preserving semantic meaningfulness, but also demonstrates superior attack strength in cross-model transferability, and cross-sample universality compared with the baseline. Moreover, we also compare AutoDAN with perplexity-based defense methods and show that AutoDAN can bypass them effectively. Code is available at https://github.com/SheltonLiu-N/AutoDAN.

Adversarial examples for deep neural networks (DNNs) are transferable: examples that successfully fool one white-box surrogate model can also deceive other black-box models with different architectures. Although a bunch of empirical studies have provided guidance on generating highly transferable adversarial examples, many of these findings fail to be well explained and even lead to confusing or inconsistent advice for practical use.In this paper, we take a further step towards understanding adversarial transferability, with a particular focus on surrogate aspects. Starting from the intriguing "little robustness" phenomenon, where models adversarially trained with mildly perturbed adversarial samples can serve as better surrogates for transfer attacks, we attribute it to a trade-off between two dominant factors: model smoothness and gradient similarity. Our research focuses on their joint effects on transferability, rather than demonstrating the separate relationships alone. Through a combination of theoretical and empirical analyses, we hypothesize that the data distribution shift induced by off-manifold samples in adversarial training is the reason that impairs gradient similarity.Building on these insights, we further explore the impacts of prevalent data augmentation and gradient regularization on transferability and analyze how the trade-off manifests in various training methods, thus building a comprehensive blueprint for the regulation mechanisms behind transferability. Finally, we provide a general route for constructing superior surrogates to boost transferability, which optimizes both model smoothness and gradient similarity simultaneously, e.g., the combination of input gradient regularization and sharpness-aware minimization (SAM), validated by extensive experiments. In summary, we call for attention to the united impacts of these two factors for launching effective transfer attacks, rather than optimizing one while ignoring the other, and emphasize the crucial role of manipulating surrogate models.

Deep neural networks are proven to be vulnerable to backdoor attacks. Detecting the trigger samples during the inference stage, i.e., the test-time trigger sample detection, can prevent the backdoor from being triggered. However, existing detection methods often require the defenders to have high accessibility to victim models, extra clean data, or knowledge about the appearance of backdoor triggers, limiting their practicality. In this paper, we propose the test-time corruption robustness consistency evaluation (TeCo)11https://github.com/CGCL-codes/TeCo, a novel test-time trigger sample detection method that only needs the hard-label outputs of the victim models without any extra information. Our journey begins with the intriguing observation that the backdoor-infected models have similar performance across different image corruptions for the clean images, but perform discrepantly for the trigger samples. Based on this phenomenon, we design TeCo to evaluate test-time robustness consistency by calculating the deviation of severity that leads to predictions' transition across different corruptions. Extensive experiments demonstrate that compared with state-of-the-art defenses, which even require either certain information about the trigger types or accessibility of clean data, TeCo outperforms them on different backdoor attacks, datasets, and model architectures, enjoying a higher AUROC by 10% and 5 times of stability.

Backdoor attacks for point clouds have elicited mounting interest with the proliferation of deep learning. The point cloud classifiers can be vulnerable to malicious actors who seek to manipulate or fool the model with specific backdoor triggers. Detecting and rejecting backdoor samples during the inference stage can effectively alleviate backdoor attacks. Recently, some black-box test-time backdoor sample detection methods have been proposed in the 2D image domain, without any underlying assumptions about the backdoor triggers. However, upon examination, we have found that these detection techniques are not effective for 3D point clouds. As a result, there is a pressing need to bridge the gap for the development of a universal approach that is specifically designed for 3D point clouds.

While deep face recognition (FR) systems have shown amazing performance in identification and verification, they also arouse privacy concerns for their excessive surveillance on users, especially for public face images widely spread on social networks. Recently, some studies adopt adversarial examples to protect photos from being identified by unauthorized face recognition systems. However, existing methods of generating adversarial face images suffer from many limitations, such as awkward visual, white-box setting, weak transferability, making them difficult to be applied to protect face privacy in reality. In this paper, we propose adversarial makeup transfer GAN (AMT-GAN)11https://github.com/CGCL-codes/AMT-GAN, a novel face protection method aiming at constructing adversarial face images that preserve stronger black-box transferability and better visual quality simultaneously. AMT-GAN leverages generative adversarial networks (GAN) to synthesize adversarial face images with makeup transferred from reference images. In particular, we introduce a new regularization module along with a joint training strategy to reconcile the conflicts between the adversarial noises and the cycle consistence loss in makeup transfer, achieving a desirable balance between the attack strength and visual changes. Extensive experiments verify that compared with state of the arts, AMT-GAN can not only preserve a comfortable visual quality, but also achieve a higher attack success rate over commercial FR APIs, including Face++, Aliyun, and Microsoft.

In this paper, we propose AdvHash, the first targeted mismatch attack on deep hashing through adversarial patch. After superimposed with the same adversarial patch, any query image with a chosen label will retrieve a set of irrelevant images with the target label. Concretely, we first formulate a set-to-set problem, where a set of samples are pushed into a predefined clustered area in the Hamming space. Then we obtain a target anchor hash code and transform the attack to a set-to-point optimization. In order to generate a image-agnostic stable adversarial patch for a chosen label more efficiently, we propose a product-based weighted gradient aggregation strategy to dynamically adjust the gradient directions of the patch, by exploiting the Hamming distances between training samples and the target anchor hash code and assigning different weights to discriminatively aggregate gradients. Extensive experiments on benchmark datasets verify that AdvHash is highly effective at attacking two state-of-the-art deep hashing schemes. Our codes are available at: https://github.com/CGCL-codes/AdvHash.