Yisong Xiao

Jun Guo, Aishan Liu, Xingyu Zheng et al.

Despite the broad application of Machine Learning models as a Service (MLaaS), they are vulnerable to model stealing attacks. These attacks can replicate the model functionality by using the black-box query process without any prior knowledge of the target victim model. Existing stealing defenses add deceptive perturbations to the victim's posterior probabilities to mislead the attackers. However, these defenses are now suffering problems of high inference computational overheads and unfavorable trade-offs between benign accuracy and stealing robustness, which challenges the feasibility of deployed models in practice. To address the problems, this paper proposes Isolation and Induction (InI), a novel and effective training framework for model stealing defenses. Instead of deploying auxiliary defense modules that introduce redundant inference time, InI directly trains a defensive model by isolating the adversary's training gradient from the expected gradient, which can effectively reduce the inference computational cost. In contrast to adding perturbations over model predictions that harm the benign accuracy, we train models to produce uninformative outputs against stealing queries, which can induce the adversary to extract little useful knowledge from victim models with minimal impact on the benign performance. Extensive experiments on several visual classification datasets (e.g., MNIST and CIFAR10) demonstrate the superior robustness (up to 48% reduction on stealing accuracy) and speed (up to 25.4x faster) of our InI over other state-of-the-art methods. Our codes can be found in https://github.com/DIG-Beihang/InI-Model-Stealing-Defense.

Yisong Xiao, Aishan Liu, Tianyuan Zhang et al.

Quantization has emerged as an essential technique for deploying deep neural networks (DNNs) on devices with limited resources. However, quantized models exhibit vulnerabilities when exposed to various noises in real-world applications. Despite the importance of evaluating the impact of quantization on robustness, existing research on this topic is limited and often disregards established principles of robustness evaluation, resulting in incomplete and inconclusive findings. To address this gap, we thoroughly evaluated the robustness of quantized models against various noises (adversarial attacks, natural corruptions, and systematic noises) on ImageNet. The comprehensive evaluation results empirically provide valuable insights into the robustness of quantized models in various scenarios, for example: (1) quantized models exhibit higher adversarial robustness than their floating-point counterparts, but are more vulnerable to natural corruptions and systematic noises; (2) in general, increasing the quantization bit-width results in a decrease in adversarial robustness, an increase in natural robustness, and an increase in systematic robustness; (3) among corruption methods, \textit{impulse noise} and \textit{glass blur} are the most harmful to quantized models, while \textit{brightness} has the least impact; (4) among systematic noises, the \textit{nearest neighbor interpolation} has the highest impact, while bilinear interpolation, cubic interpolation, and area interpolation are the three least harmful. Our research contributes to advancing the robust quantization of models and their deployment in real-world scenarios.

Tianyuan Zhang, Yisong Xiao, Xiaoya Zhang et al.

Adversarial attacks in the physical world can harm the robustness of detection models. Evaluating the robustness of detection models in the physical world can be challenging due to the time-consuming and labor-intensive nature of many experiments. Thus, virtual simulation experiments can provide a solution to this challenge. However, there is no unified detection benchmark based on virtual simulation environment. To address this challenge, we proposed an instant-level data generation pipeline based on the CARLA simulator. Using this pipeline, we generated the DCI dataset and conducted extensive experiments on three detection models and three physical adversarial attacks. The dataset covers 7 continuous and 1 discrete scenes, with over 40 angles, 20 distances, and 20,000 positions. The results indicate that Yolo v6 had strongest resistance, with only a 6.59% average AP drop, and ASA was the most effective attack algorithm with a 14.51% average AP reduction, twice that of other algorithms. Static scenes had higher recognition AP, and results under different weather conditions were similar. Adversarial attack algorithm improvement may be approaching its 'limitation'.

Quanchen Zou, Moyang Chen, Zonghao Ying et al.

Large Vision-Language Models (LVLMs) undergo safety alignment to suppress harmful content. However, current defenses predominantly target explicit malicious patterns in the input representation, often overlooking the vulnerabilities inherent in compositional reasoning. In this paper, we identify a systemic flaw where LVLMs can be induced to synthesize harmful logic from benign premises. We formalize this attack paradigm as \textit{Reasoning-Oriented Programming}, drawing a structural analogy to Return-Oriented Programming in systems security. Just as ROP circumvents memory protections by chaining benign instruction sequences, our approach exploits the model's instruction-following capability to orchestrate a semantic collision of orthogonal benign inputs. We instantiate this paradigm via \tool{}, an automated framework that optimizes for \textit{semantic orthogonality} and \textit{spatial isolation}. By generating visual gadgets that are semantically decoupled from the harmful intent and arranging them to prevent premature feature fusion, \tool{} forces the malicious logic to emerge only during the late-stage reasoning process. This effectively bypasses perception-level alignment. We evaluate \tool{} on SafeBench and MM-SafetyBench across 7 state-of-the-art 0.LVLMs, including GPT-4o and Claude 3.7 Sonnet. Our results demonstrate that \tool{} consistently circumvents safety alignment, outperforming the strongest existing baseline by an average of 4.67\% on open-source models and 9.50\% on commercial models.

Tony Ma, Songze Li, Yisong Xiao et al.

The transferability of adversarial examples is a crucial aspect of evaluating the robustness of deep learning systems, particularly in black-box scenarios. Although several methods have been proposed to enhance cross-model transferability, little attention has been paid to the transferability of adversarial examples across different tasks. This issue has become increasingly relevant with the emergence of foundational multi-task AI systems such as Visual ChatGPT, rendering the utility of adversarial samples generated by a single task relatively limited. Furthermore, these systems often entail inferential functions beyond mere recognition-like tasks. To address this gap, we propose a novel Visual Relation-based cross-task Adversarial Patch generation method called VRAP, which aims to evaluate the robustness of various visual tasks, especially those involving visual reasoning, such as Visual Question Answering and Image Captioning. VRAP employs scene graphs to combine object recognition-based deception with predicate-based relations elimination, thereby disrupting the visual reasoning information shared among inferential tasks. Our extensive experiments demonstrate that VRAP significantly surpasses previous methods in terms of black-box transferability across diverse visual reasoning tasks.

Yisong Xiao, Tianyuan Zhang, Shunchang Liu et al.

Quantization has emerged as an essential technique for deploying deep neural networks (DNNs) on devices with limited resources. However, quantized models exhibit vulnerabilities when exposed to various noises in real-world applications. Despite the importance of evaluating the impact of quantization on robustness, existing research on this topic is limited and often disregards established principles of robustness evaluation, resulting in incomplete and inconclusive findings. To address this gap, we thoroughly evaluated the robustness of quantized models against various noises (adversarial attacks, natural corruptions, and systematic noises) on ImageNet. Extensive experiments demonstrate that lower-bit quantization is more resilient to adversarial attacks but is more susceptible to natural corruptions and systematic noises. Notably, our investigation reveals that impulse noise (in natural corruptions) and the nearest neighbor interpolation (in systematic noises) have the most significant impact on quantized models. Our research contributes to advancing the robust quantization of models and their deployment in real-world scenarios.

Zonghao Ying, Deyue Zhang, Zonglei Jing et al.

Multi-turn jailbreak attacks simulate real-world human interactions by engaging large language models (LLMs) in iterative dialogues, exposing critical safety vulnerabilities. However, existing methods often struggle to balance semantic coherence with attack effectiveness, resulting in either benign semantic drift or ineffective detection evasion. To address this challenge, we propose Reasoning-Augmented Conversation, a novel multi-turn jailbreak framework that reformulates harmful queries into benign reasoning tasks and leverages LLMs' strong reasoning capabilities to compromise safety alignment. Specifically, we introduce an attack state machine framework to systematically model problem translation and iterative reasoning, ensuring coherent query generation across multiple turns. Building on this framework, we design gain-guided exploration, self-play, and rejection feedback modules to preserve attack semantics, enhance effectiveness, and sustain reasoning-driven attack progression. Extensive experiments on multiple LLMs demonstrate that RACE achieves state-of-the-art attack effectiveness in complex conversational scenarios, with attack success rates (ASRs) increasing by up to 96%. Notably, our approach achieves ASRs of 82% and 92% against leading commercial models, OpenAI o1 and DeepSeek R1, underscoring its potency. We release our code at https://github.com/NY1024/RACE to facilitate further research in this critical domain.

Zonghao Ying, Haowen Dai, Tianyuan Zhang et al.

Self-evolving agents offer a promising path toward scalable autonomy. However, in this work, we show that in competitive environments, self-evolution can instead give rise to a serious and previously underexplored risk: the spontaneous emergence of deception as an evolutionarily stable strategy. We conduct a systematic empirical study on the self-evolution of large language model (LLM) agents in a competitive Bidding Arena, where agents iteratively refine their strategies through interaction-driven reflection. Across different evolutionary paths (\eg, Neutral, Honesty-Guided, and Deception-Guided), we find a consistent pattern: under utility-driven competition, unconstrained self-evolution reliably drifts toward deceptive behaviors, even when honest strategies remain viable. This drift is explained by a fundamental asymmetry in generalization. Deception evolves as a transferable meta-strategy that generalizes robustly across diverse and unseen tasks, whereas honesty-based strategies are fragile and often collapse outside their original contexts. Further analysis of agents internal states reveals the emergence of rationalization mechanisms, through which agents justify or deny deceptive actions to reconcile competitive success with normative instructions. Our paper exposes a fundamental tension between agent self-evolution and alignment, highlighting the risks of deploying self-improving agents in adversarial environments.

Shiji Zhao, Mengyang Wang, Shukun Xiong et al.

With the rapid development and widespread application of Large Language Models (LLMs), their potential safety risks have attracted widespread attention. Reinforcement Learning from Human Feedback (RLHF) has been adopted to enhance the safety performance of LLMs. As a simple and effective alternative to RLHF, Direct Preference Optimization (DPO) is widely used for safety alignment. However, safety alignment still suffers from severe overfitting, which limits its actual performance. This paper revisits the overfitting phenomenon from the perspective of the model's comprehension of the training data. We find that the Imbalanced Preference Comprehension phenomenon exists between responses in preference pairs, which compromises the model's safety performance. To address this, we propose Balanced Direct Preference Optimization (B-DPO), which adaptively modulates optimization strength between preferred and dispreferred responses based on mutual information. A series of experimental results show that B-DPO can enhance the safety capability while maintaining the competitive general capabilities of LLMs on various mainstream benchmarks compared to state-of-the-art methods. \color{red}{Warning: This paper contains examples of harmful texts, and reader discretion is recommended.

Yisong Xiao, Aishan Liu, Siyuan Liang et al.

Large Language Models (LLMs) have demonstrated impressive performance across various tasks, yet they remain vulnerable to generating toxic content, necessitating detoxification strategies to ensure safe and responsible deployment. Test-time detoxification methods, which typically introduce static or dynamic interventions into LLM representations, offer a promising solution due to their flexibility and minimal invasiveness. However, current approaches often suffer from imprecise interventions, primarily due to their insufficient exploration of the transition space between toxic and non-toxic outputs. To address this challenge, we propose \textsc{A}utoregressive \textsc{R}eward \textsc{G}uided \textsc{R}epresentation \textsc{E}diting (ARGRE), a novel test-time detoxification framework that explicitly models toxicity transitions within the latent representation space, enabling stable and precise reward-guided editing. ARGRE identifies non-toxic semantic directions and interpolates between toxic and non-toxic representations to reveal fine-grained transition trajectories. These trajectories transform sparse toxicity annotations into dense training signals, enabling the construction of an autoregressive reward model that delivers stable and precise editing guidance. At inference, the reward model guides an adaptive two-step editing process to obtain detoxified representations: it first performs directional steering based on expected reward gaps to shift representations toward non-toxic regions, followed by lightweight gradient-based refinements. Extensive experiments across 8 widely used LLMs show that ARGRE significantly outperforms leading baselines in effectiveness (-62.21% toxicity) and efficiency (-47.58% inference time), while preserving the core capabilities of the original model with minimal degradation. Our code is available at the website.

Yisong Xiao, Aishan Liu, QianJia Cheng et al.

Large Vision-Language Models (LVLMs) have been widely adopted in various applications; however, they exhibit significant gender biases. Existing benchmarks primarily evaluate gender bias at the demographic group level, neglecting individual fairness, which emphasizes equal treatment of similar individuals. This research gap limits the detection of discriminatory behaviors, as individual fairness offers a more granular examination of biases that group fairness may overlook. For the first time, this paper introduces the GenderBias-\emph{VL} benchmark to evaluate occupation-related gender bias in LVLMs using counterfactual visual questions under individual fairness criteria. To construct this benchmark, we first utilize text-to-image diffusion models to generate occupation images and their gender counterfactuals. Subsequently, we generate corresponding textual occupation options by identifying stereotyped occupation pairs with high semantic similarity but opposite gender proportions in real-world statistics. This method enables the creation of large-scale visual question counterfactuals to expose biases in LVLMs, applicable in both multimodal and unimodal contexts through modifying gender attributes in specific modalities. Overall, our GenderBias-\emph{VL} benchmark comprises 34,581 visual question counterfactual pairs, covering 177 occupations. Using our benchmark, we extensively evaluate 15 commonly used open-source LVLMs (\eg, LLaVA) and state-of-the-art commercial APIs, including GPT-4o and Gemini-Pro. Our findings reveal widespread gender biases in existing LVLMs. Our benchmark offers: (1) a comprehensive dataset for occupation-related gender bias evaluation; (2) an up-to-date leaderboard on LVLM biases; and (3) a nuanced understanding of the biases presented by these models. \footnote{The dataset and code are available at the \href{https://genderbiasvl.github.io/}{website}.}

Aishan Liu, Xinwei Zhang, Yisong Xiao et al.

Pre-trained vision models (PVMs) have become a dominant component due to their exceptional performance when fine-tuned for downstream tasks. However, the presence of backdoors within PVMs poses significant threats. Unfortunately, existing studies primarily focus on backdooring PVMs for the classification task, neglecting potential inherited backdoors in downstream tasks such as detection and segmentation. In this paper, we propose the Pre-trained Trojan attack, which embeds backdoors into a PVM, enabling attacks across various downstream vision tasks. We highlight the challenges posed by cross-task activation and shortcut connections in successful backdoor attacks. To achieve effective trigger activation in diverse tasks, we stylize the backdoor trigger patterns with class-specific textures, enhancing the recognition of task-irrelevant low-level features associated with the target class in the trigger pattern. Moreover, we address the issue of shortcut connections by introducing a context-free learning pipeline for poison training. In this approach, triggers without contextual backgrounds are directly utilized as training data, diverging from the conventional use of clean images. Consequently, we establish a direct shortcut from the trigger to the target class, mitigating the shortcut connection issue. We conducted extensive experiments to thoroughly validate the effectiveness of our attacks on downstream detection and segmentation tasks. Additionally, we showcase the potential of our approach in more practical scenarios, including large vision models and 3D object detection in autonomous driving. This paper aims to raise awareness of the potential threats associated with applying PVMs in practical scenarios. Our codes will be available upon paper publication.

Quanchen Zou, Zonghao Ying, Moyang Chen et al.

The increasing sophistication of large vision-language models (LVLMs) has been accompanied by advances in safety alignment mechanisms designed to prevent harmful content generation. However, these defenses remain vulnerable to sophisticated adversarial attacks. Existing jailbreak methods typically rely on direct and semantically explicit prompts, overlooking subtle vulnerabilities in how LVLMs compose information over multiple reasoning steps. In this paper, we propose a novel and effective jailbreak framework inspired by Return-Oriented Programming (ROP) techniques from software security. Our approach decomposes a harmful instruction into a sequence of individually benign visual gadgets. A carefully engineered textual prompt directs the sequence of inputs, prompting the model to integrate the benign visual gadgets through its reasoning process to produce a coherent and harmful output. This makes the malicious intent emergent and difficult to detect from any single component. We validate our method through extensive experiments on established benchmarks including SafeBench and MM-SafetyBench, targeting popular LVLMs. Results show that our approach consistently and substantially outperforms existing baselines on state-of-the-art models, achieving near-perfect attack success rates (over 0.90 on SafeBench) and improving ASR by up to 0.39. Our findings reveal a critical and underexplored vulnerability that exploits the compositional reasoning abilities of LVLMs, highlighting the urgent need for defenses that secure the entire reasoning process.

Tianyuan Zhang, Lu Wang, Hainan Li et al.

Lane detection (LD) is an essential component of autonomous driving systems, providing fundamental functionalities like adaptive cruise control and automated lane centering. Existing LD benchmarks primarily focus on evaluating common cases, neglecting the robustness of LD models against environmental illusions such as shadows and tire marks on the road. This research gap poses significant safety challenges since these illusions exist naturally in real-world traffic situations. For the first time, this paper studies the potential threats caused by these environmental illusions to LD and establishes the first comprehensive benchmark LanEvil for evaluating the robustness of LD against this natural corruption. We systematically design 14 prevalent yet critical types of environmental illusions (e.g., shadow, reflection) that cover a wide spectrum of real-world influencing factors in LD tasks. Based on real-world environments, we create 94 realistic and customizable 3D cases using the widely used CARLA simulator, resulting in a dataset comprising 90,292 sampled images. Through extensive experiments, we benchmark the robustness of popular LD methods using LanEvil, revealing substantial performance degradation (-5.37% Accuracy and -10.70% F1-Score on average), with shadow effects posing the greatest risk (-7.39% Accuracy). Additionally, we assess the performance of commercial auto-driving systems OpenPilot and Apollo through collaborative simulations, demonstrating that proposed environmental illusions can lead to incorrect decisions and potential traffic accidents. To defend against environmental illusions, we propose the Attention Area Mixing (AAM) approach using hard examples, which witness significant robustness improvement (+3.76%) under illumination effects. We hope our paper can contribute to advancing more robust auto-driving systems in the future. Website: https://lanevil.github.io/.

Yisong Xiao, Aishan Liu, Tianlin Li et al.

Machine learning (ML) systems have achieved remarkable performance across a wide area of applications. However, they frequently exhibit unfair behaviors in sensitive application domains, raising severe fairness concerns. To evaluate and test fairness, engineers often generate individual discriminatory instances to expose unfair behaviors before model deployment. However, existing baselines ignore the naturalness of generation and produce instances that deviate from the real data distribution, which may fail to reveal the actual model fairness since these unnatural discriminatory instances are unlikely to appear in practice. To address the problem, this paper proposes a framework named Latent Imitator (LIMI) to generate more natural individual discriminatory instances with the help of a generative adversarial network (GAN), where we imitate the decision boundary of the target model in the semantic latent space of GAN and further samples latent instances on it. Specifically, we first derive a surrogate linear boundary to coarsely approximate the decision boundary of the target model, which reflects the nature of the original data distribution. Subsequently, to obtain more natural instances, we manipulate random latent vectors to the surrogate boundary with a one-step movement, and further conduct vector calculation to probe two potential discriminatory candidates that may be more closely located in the real decision boundary. Extensive experiments on various datasets demonstrate that our LIMI outperforms other baselines largely in effectiveness ($\times$9.42 instances), efficiency ($\times$8.71 speeds), and naturalness (+19.65%) on average. In addition, we empirically demonstrate that retraining on test samples generated by our approach can lead to improvements in both individual fairness (45.67% on $IF_r$ and 32.81% on $IF_o$) and group fairness (9.86% on $SPD$ and 28.38% on $AOD$}).