Yige Li

Yunhao Feng, Yifan Ding, Xiaohu Du et al.

Computer-use agents extend language models from text generation to sustained interaction with files, terminals, browsers, and external tools. This shift creates safety risks that are difficult to detect from isolated prompts or final responses, because harm often emerges only through multi-step execution traces whose individual actions appear locally benign. We introduce BraveGuard, a self-evolving defense framework for training guard models from open-world threat signals and realistic agent trajectories. BraveGuard mines recent research sources to identify emerging risks and attack patterns, instantiates them as executable computer-use tasks, collects agent rollouts, and derives trajectory-level supervision for guard model training. As new threats and validation failures appear, the pipeline can be repeated, yielding an adaptive defense loop rather than a static, benchmark-driven training process. We instantiate BraveGuard by training multiple guard backbones, including Qwen3-Guard and Llama-Guard variants, and evaluate the resulting guards on trajectory-level agent-safety benchmarks. BraveGuard consistently improves safety detection across computer-use trajectories. On AgentHazard, it substantially improves detection accuracy over off-the-shelf guard models, with accuracy increasing from 38.79% to 82.38% under the averaged guard-model setting. These results show that guard supervision grounded in open-world threat discovery and realistic agent execution can improve safety monitoring beyond fixed taxonomies and synthetic prompt-level data. BraveGuard offers a scalable path toward adaptive defenses for computer-use agents facing evolving real-world risks.

Yige Li, Hanxun Huang, Yunhan Zhao et al.

Generative large language models (LLMs) have achieved state-of-the-art results on a wide range of tasks, yet they remain susceptible to backdoor attacks: carefully crafted triggers in the input can manipulate the model to produce adversary-specified outputs. While prior research has predominantly focused on backdoor risks in vision and classification settings, the vulnerability of LLMs in open-ended text generation remains underexplored. To fill this gap, we introduce BackdoorLLM (Our BackdoorLLM benchmark was awarded First Prize in the SafetyBench competition, https://www.mlsafety.org/safebench/winners, organized by the Center for AI Safety, https://safe.ai/.), the first comprehensive benchmark for systematically evaluating backdoor threats in text-generation LLMs. BackdoorLLM provides: (i) a unified repository of benchmarks with a standardized training and evaluation pipeline; (ii) a diverse suite of attack modalities, including data poisoning, weight poisoning, hidden-state manipulation, and chain-of-thought hijacking; (iii) over 200 experiments spanning 8 distinct attack strategies, 7 real-world scenarios, and 6 model architectures; (iv) key insights into the factors that govern backdoor effectiveness and failure modes in LLMs; and (v) a defense toolkit encompassing 7 representative mitigation techniques. Our code and datasets are available at https://github.com/bboylyg/BackdoorLLM. We will continuously incorporate emerging attack and defense methodologies to support the research in advancing the safety and reliability of LLMs.

Yutao Wu, Xiao Liu, Yifeng Gao et al.

This work identifies a critical failure mode in frontier large language models (LLMs), which we term Internal Safety Collapse (ISC): under certain task conditions, models enter a state in which they continuously generate harmful content while executing otherwise benign tasks. We introduce TVD (Task, Validator, Data), a framework that triggers ISC through domain tasks where generating harmful content is the only valid completion, and construct ISC-Bench containing 53 scenarios across 8 professional disciplines. Evaluated on JailbreakBench, three representative scenarios yield worst-case safety failure rates averaging 95.3% across four frontier LLMs (including GPT-5.2 and Claude Sonnet 4.5), substantially exceeding standard jailbreak attacks. Frontier models are more vulnerable than earlier LLMs: the very capabilities that enable complex task execution become liabilities when tasks intrinsically involve harmful content. This reveals a growing attack surface: almost every professional domain uses tools that process sensitive data, and each new dual-use tool automatically expands this vulnerability--even without any deliberate attack. Despite substantial alignment efforts, frontier LLMs retain inherently unsafe internal capabilities: alignment reshapes observable outputs but does not eliminate the underlying risk profile. These findings underscore the need for caution when deploying LLMs in high-stakes settings. Source code: https://github.com/wuyoscar/ISC-Bench

Yige Li, Yunhao Feng, Jun Sun

AI safety is still largely framed as alignment: training models to follow human preferences, safety policies, and normative constraints. That framing has improved the behavior of modern language models, but aligned behavior does not by itself guarantee that a deployed agent can be stopped, overridden, or constrained once it operates in open-ended, interactive, and tool-using environments. A system may be safe in expectation and still fail to yield to explicit runtime authority under conflicting instructions, long-horizon execution, adversarial inputs, or risky tool use. This position paper argues that AI safety therefore requires controllability as a first-class objective. We define \emph{controllability} as the ability of an AI system to remain reliably interruptible, overridable, redirectable, and constrainable by explicit control signals at runtime while preserving ordinary utility when such signals are absent. To study this gap, we introduce \controlbench{}, a benchmark for evaluating controllability failures in high-risk agentic scenarios. Experiments with OpenClaw-based agents show that current alignment and guardrail mechanisms reduce risk, but often fail to provide persistent, authoritative, and enforceable runtime control. We therefore propose a control-centric architectural framework that highlights explicit control planes, runtime intervention pathways, persistent control states, and auditable decision interfaces as key design principles for future controllable AI systems.

Zhe Li, Wei Zhao, Yige Li et al.

Influence functions are important for quantifying the impact of individual training data points on a model's predictions. Although extensive research has been conducted on influence functions in traditional machine learning models, their application to large language models (LLMs) has been limited. In this work, we conduct a systematic study to address a key question: do influence functions work on LLMs? Specifically, we evaluate influence functions across multiple tasks and find that they consistently perform poorly in most settings. Our further investigation reveals that their poor performance can be attributed to: (1) inevitable approximation errors when estimating the iHVP component due to the scale of LLMs, (2) uncertain convergence during fine-tuning, and, more fundamentally, (3) the definition itself, as changes in model parameters do not necessarily correlate with changes in LLM behavior. Thus, our study suggests the need for alternative approaches for identifying influential samples.

Jiaming Zhang, Lingyu Qiu, Qi Yi et al.

The vulnerability of Deep Neural Networks (DNNs) to adversarial attacks poses a significant challenge to their deployment in safety-critical applications. While extensive research has addressed various attack scenarios, the no-box attack setting where adversaries have no prior knowledge, including access to training data of the target model, remains relatively underexplored despite its practical relevance. This work presents a systematic investigation into leveraging large-scale Vision-Language Models (VLMs), particularly CLIP, as surrogate models for executing no-box attacks. Our theoretical and empirical analyses reveal a key limitation in the execution of no-box attacks stemming from insufficient discriminative capabilities for direct application of vanilla CLIP as a surrogate model. To address this limitation, we propose MF-CLIP: a novel framework that enhances CLIP's effectiveness as a surrogate model through margin-aware feature space optimization. Comprehensive evaluations across diverse architectures and datasets demonstrate that MF-CLIP substantially advances the state-of-the-art in no-box attacks, surpassing existing baselines by 15.23% on standard models and achieving a 9.52% improvement on adversarially trained models. Our code will be made publicly available to facilitate reproducibility and future research in this direction.

Yunhan Zhao, Xiang Zheng, Lin Luo et al.

In this paper, we focus on black-box defense for VLMs against jailbreak attacks. Existing black-box defense methods are either unimodal or bimodal. Unimodal methods enhance either the vision or language module of the VLM, while bimodal methods robustify the model through text-image representation realignment. However, these methods suffer from two limitations: 1) they fail to fully exploit the cross-modal information, or 2) they degrade the model performance on benign inputs. To address these limitations, we propose a novel blue-team method BlueSuffix that defends target VLMs against jailbreak attacks without compromising its performance under black-box setting. BlueSuffix includes three key components: 1) a visual purifier against jailbreak images, 2) a textual purifier against jailbreak texts, and 3) a blue-team suffix generator using reinforcement fine-tuning for enhancing cross-modal robustness. We empirically show on four VLMs (LLaVA, MiniGPT-4, InstructionBLIP, and Gemini) and four safety benchmarks (Harmful Instruction, AdvBench, MM-SafetyBench, and RedTeam-2K) that BlueSuffix outperforms the baseline defenses by a significant margin. Our BlueSuffix opens up a promising direction for defending VLMs against jailbreak attacks. Code is available at https://github.com/Vinsonzyh/BlueSuffix.

Yige Li, Jiabo He, Hanxun Huang et al.

Backdoor attacks have become a significant threat to the pre-training and deployment of deep neural networks (DNNs). Although numerous methods for detecting and mitigating backdoor attacks have been proposed, most rely on identifying and eliminating the ``shortcut" created by the backdoor, which links a specific source class to a target class. However, these approaches can be easily circumvented by designing multiple backdoor triggers that create shortcuts everywhere and therefore nowhere specific. In this study, we explore the concept of Multi-Trigger Backdoor Attacks (MTBAs), where multiple adversaries leverage different types of triggers to poison the same dataset. By proposing and investigating three types of multi-trigger attacks including \textit{parallel}, \textit{sequential}, and \textit{hybrid} attacks, we demonstrate that 1) multiple triggers can coexist, overwrite, or cross-activate one another, and 2) MTBAs easily break the prevalent shortcut assumption underlying most existing backdoor detection/removal methods, rendering them ineffective. Given the security risk posed by MTBAs, we have created a multi-trigger backdoor poisoning dataset to facilitate future research on detecting and mitigating these attacks, and we also discuss potential defense strategies against MTBAs. Our code is available at https://github.com/bboylyg/Multi-Trigger-Backdoor-Attacks.

Xiang Zheng, Yutao Wu, Hanxun Huang et al.

Autonomous code agents built on large language models are reshaping software and AI development through tool use, long-horizon reasoning, and self-directed interaction. However, this autonomy introduces a previously unrecognized security risk: agentic interaction fundamentally expands the LLM attack surface, enabling systematic probing and recovery of hidden system prompts that guide model behavior. We identify system prompt extraction as an emergent vulnerability intrinsic to code agents and present \textbf{\textsc{JustAsk}}, a self-evolving framework that autonomously discovers effective extraction strategies through interaction alone. Unlike prior prompt-engineering or dataset-based attacks, \textsc{JustAsk} requires no handcrafted prompts, labeled supervision, or privileged access beyond standard user interaction. It formulates extraction as an online exploration problem, using Upper Confidence Bound-based strategy selection and a hierarchical skill space spanning atomic probes and high-level orchestration. These skills exploit imperfect system-instruction generalization and inherent tensions between helpfulness and safety. Evaluated on \textbf{41} black-box commercial models across multiple providers, \textsc{JustAsk} consistently achieves full or near-complete system prompt recovery, revealing recurring design- and architecture-level vulnerabilities. Our results expose system prompts as a critical yet largely unprotected attack surface in modern agent systems.

Jiayu Li, Yunhan Zhao, Xiang Zheng et al.

Vision-Language-Action (VLA) models enable robots to interpret natural-language instructions and perform diverse tasks, yet their integration of perception, language, and control introduces new safety vulnerabilities. Despite growing interest in attacking such models, the effectiveness of existing techniques remains unclear due to the absence of a unified evaluation framework. One major issue is that differences in action tokenizers across VLA architectures hinder reproducibility and fair comparison. More importantly, most existing attacks have not been validated in real-world scenarios. To address these challenges, we propose AttackVLA, a unified framework that aligns with the VLA development lifecycle, covering data construction, model training, and inference. Within this framework, we implement a broad suite of attacks, including all existing attacks targeting VLAs and multiple adapted attacks originally developed for vision-language models, and evaluate them in both simulation and real-world settings. Our analysis of existing attacks reveals a critical gap: current methods tend to induce untargeted failures or static action states, leaving targeted attacks that drive VLAs to perform precise long-horizon action sequences largely unexplored. To fill this gap, we introduce BackdoorVLA, a targeted backdoor attack that compels a VLA to execute an attacker-specified long-horizon action sequence whenever a trigger is present. We evaluate BackdoorVLA in both simulated benchmarks and real-world robotic settings, achieving an average targeted success rate of 58.4% and reaching 100% on selected tasks. Our work provides a standardized framework for evaluating VLA vulnerabilities and demonstrates the potential for precise adversarial manipulation, motivating further research on securing VLA-based embodied systems.

Hanxun Huang, Sarah Erfani, Yige Li et al.

Contrastive language-image pretraining (CLIP) has been found to be vulnerable to poisoning backdoor attacks where the adversary can achieve an almost perfect attack success rate on CLIP models by poisoning only 0.01\% of the training dataset. This raises security concerns on the current practice of pretraining large-scale models on unscrutinized web data using CLIP. In this work, we analyze the representations of backdoor-poisoned samples learned by CLIP models and find that they exhibit unique characteristics in their local subspace, i.e., their local neighborhoods are far more sparse than that of clean samples. Based on this finding, we conduct a systematic study on detecting CLIP backdoor attacks and show that these attacks can be easily and efficiently detected by traditional density ratio-based local outlier detectors, whereas existing backdoor sample detection methods fail. Our experiments also reveal that an unintentional backdoor already exists in the original CC3M dataset and has been trained into a popular open-source model released by OpenCLIP. Based on our detector, one can clean up a million-scale web dataset (e.g., CC3M) efficiently within 15 minutes using 4 Nvidia A100 GPUs. The code is publicly available in our \href{https://github.com/HanxunH/Detect-CLIP-Backdoor-Samples}{GitHub repository}.

Yunhao Feng, Yige Li, Yutao Wu et al.

Large language model (LLM) agents execute tasks through multi-step workflows that combine planning, memory, and tool use. While this design enables autonomy, it also expands the attack surface for backdoor threats. Backdoor triggers injected into specific stages of an agent workflow can persist through multiple intermediate states and adversely influence downstream outputs. However, existing studies remain fragmented and typically analyze individual attack vectors in isolation, leaving the cross-stage interaction and propagation of backdoor triggers poorly understood from an agent-centric perspective. To fill this gap, we propose \textbf{BackdoorAgent}, a modular and stage-aware framework that provides a unified, agent-centric view of backdoor threats in LLM agents. BackdoorAgent structures the attack surface into three functional stages of agentic workflows, including \textbf{planning attacks}, \textbf{memory attacks}, and \textbf{tool-use attacks}, and instruments agent execution to enable systematic analysis of trigger activation and propagation across different stages. Building on this framework, we construct a standardized benchmark spanning four representative agent applications: \textbf{Agent QA}, \textbf{Agent Code}, \textbf{Agent Web}, and \textbf{Agent Drive}, covering both language-only and multimodal settings. Our empirical analysis shows that \textit{triggers implanted at a single stage can persist across multiple steps and propagate through intermediate states.} For instance, when using a GPT-based backbone, we observe trigger persistence in 43.58\% of planning attacks, 77.97\% of memory attacks, and 60.28\% of tool-stage attacks, highlighting the vulnerabilities of the agentic workflow itself to backdoor threats. To facilitate reproducibility and future research, our code and benchmark are publicly available at GitHub.

Hanxun Huang, Sarah Erfani, Yige Li et al.

As Contrastive Language-Image Pre-training (CLIP) models are increasingly adopted for diverse downstream tasks and integrated into large vision-language models (VLMs), their susceptibility to adversarial perturbations has emerged as a critical concern. In this work, we introduce \textbf{X-Transfer}, a novel attack method that exposes a universal adversarial vulnerability in CLIP. X-Transfer generates a Universal Adversarial Perturbation (UAP) capable of deceiving various CLIP encoders and downstream VLMs across different samples, tasks, and domains. We refer to this property as \textbf{super transferability}--a single perturbation achieving cross-data, cross-domain, cross-model, and cross-task adversarial transferability simultaneously. This is achieved through \textbf{surrogate scaling}, a key innovation of our approach. Unlike existing methods that rely on fixed surrogate models, which are computationally intensive to scale, X-Transfer employs an efficient surrogate scaling strategy that dynamically selects a small subset of suitable surrogates from a large search space. Extensive evaluations demonstrate that X-Transfer significantly outperforms previous state-of-the-art UAP methods, establishing a new benchmark for adversarial transferability across CLIP models. The code is publicly available in our \href{https://github.com/HanxunH/XTransferBench}{GitHub repository}.

Peihai Jiang, Xixiang Lyu, Yige Li et al.

Supervised fine-tuning has become the predominant method for adapting large pretrained models to downstream tasks. However, recent studies have revealed that these models are vulnerable to backdoor attacks, where even a small number of malicious samples can successfully embed backdoor triggers into the model. While most existing defense methods focus on post-training backdoor defense, efficiently defending against backdoor attacks during training phase remains largely unexplored. To address this gap, we propose a novel defense method called Backdoor Token Unlearning (BTU), which proactively detects and neutralizes trigger tokens during the training stage. Our work is based on two key findings: 1) backdoor learning causes distinctive differences between backdoor token parameters and clean token parameters in word embedding layers, and 2) the success of backdoor attacks heavily depends on backdoor token parameters. The BTU defense leverages these properties to identify aberrant embedding parameters and subsequently removes backdoor behaviors using a fine-grained unlearning technique. Extensive evaluations across three datasets and four types of backdoor attacks demonstrate that BTU effectively defends against these threats while preserving the model's performance on primary tasks. Our code is available at https://github.com/XDJPH/BTU.

Wei Zhao, Zhe Li, Yige Li et al.

Large Vision-Language Models (LVLMs) have made significant strides in multimodal comprehension, thanks to extensive pre-training and fine-tuning on large-scale visual datasets. However, despite their robust textual safety mechanisms, they remain vulnerable to harmful visual inputs. Existing safeguards-typically relying on pre-filtering or fine-tuning-incur high costs and diminish overall utility. To address this critical vulnerability, we introduce SafeCLIP, a lightweight method that leverages LVLMs inherent multimodal alignment for zero-shot toxic image detection. By projecting CLIPs discarded CLS token into its text space and matching it with toxic descriptors, SafeCLIP detects harmful content without any architectural changes-adding minimal latency and enabling dynamic safety corrections during inference and fine-tuning.Experiments show that SafeCLIP achieves a 66.9% defense success rate with only 3.2% false positive rate and 7.2% overhead. In contrast, state-of-the-art methods achieve 52.9% success but have a 10.7% false positive rate and 210% overhead. Our work demonstrates that leveraging inherent multimodal alignment can yield efficient, low-cost LVLM safety. Code is available at anonymous.4open.science/r/safeclip-2C01.

Yige Li, Hanxun Huang, Jiaming Zhang et al.

Backdoor attacks covertly implant triggers into deep neural networks (DNNs) by poisoning a small portion of the training data with pre-designed backdoor triggers. This vulnerability is exacerbated in the era of large models, where extensive (pre-)training on web-crawled datasets is susceptible to compromise. In this paper, we introduce a novel two-step defense framework named Expose Before You Defend (EBYD). EBYD unifies existing backdoor defense methods into a comprehensive defense system with enhanced performance. Specifically, EBYD first exposes the backdoor functionality in the backdoored model through a model preprocessing step called backdoor exposure, and then applies detection and removal methods to the exposed model to identify and eliminate the backdoor features. In the first step of backdoor exposure, we propose a novel technique called Clean Unlearning (CUL), which proactively unlearns clean features from the backdoored model to reveal the hidden backdoor features. We also explore various model editing/modification techniques for backdoor exposure, including fine-tuning, model sparsification, and weight perturbation. Using EBYD, we conduct extensive experiments on 10 image attacks and 6 text attacks across 2 vision datasets (CIFAR-10 and an ImageNet subset) and 4 language datasets (SST-2, IMDB, Twitter, and AG's News). The results demonstrate the importance of backdoor exposure for backdoor defense, showing that the exposed models can significantly benefit a range of downstream defense tasks, including backdoor label detection, backdoor trigger recovery, backdoor model detection, and backdoor removal. We hope our work could inspire more research in developing advanced defense frameworks with exposed models. Our code is available at: https://github.com/bboylyg/Expose-Before-You-Defend.

Nay Myat Min, Long H. Pham, Yige Li et al.

Large language models (LLMs) demonstrate remarkable performance across myriad language tasks, yet they remain vulnerable to backdoor attacks, where adversaries implant hidden triggers that systematically manipulate model outputs. Traditional defenses focus on explicit token-level anomalies and therefore overlook semantic backdoors-covert triggers embedded at the conceptual level (e.g., ideological stances or cultural references) that rely on meaning-based cues rather than lexical oddities. We first show, in a controlled finetuning setting, that such semantic backdoors can be implanted with only a small poisoned corpus, establishing their practical feasibility. We then formalize the notion of semantic backdoors in LLMs and introduce a black-box detection framework, RAVEN (short for "Response Anomaly Vigilance for uncovering semantic backdoors"), which combines semantic entropy with cross-model consistency analysis. The framework probes multiple models with structured topic-perspective prompts, clusters the sampled responses via bidirectional entailment, and flags anomalously uniform outputs; cross-model comparison isolates model-specific anomalies from corpus-wide biases. Empirical evaluations across diverse LLM families (GPT-4o, Llama, DeepSeek, Mistral) uncover previously undetected semantic backdoors, providing the first proof-of-concept evidence of these hidden vulnerabilities and underscoring the urgent need for concept-level auditing of deployed language models. We open-source our code and data at https://github.com/NayMyatMin/RAVEN.

Yige Li, Wei Zhao, Zhe Li et al.

Backdoor mechanisms have traditionally been studied as security threats that compromise the integrity of machine learning models. However, the same mechanism -- the conditional activation of specific behaviors through input triggers -- can also serve as a controllable and auditable interface for trustworthy model behavior. In this work, we present \textbf{Backdoor4Good (B4G)}, a unified benchmark and framework for \textit{beneficial backdoor} applications in large language models (LLMs). Unlike conventional backdoor studies focused on attacks and defenses, B4G repurposes backdoor conditioning for Beneficial Tasks that enhance safety, controllability, and accountability. It formalizes beneficial backdoor learning under a triplet formulation $(T, A, U)$, representing the \emph{Trigger}, \emph{Activation mechanism}, and \emph{Utility function}, and implements a benchmark covering four trust-centric applications. Through extensive experiments across Llama3.1-8B, Gemma-2-9B, Qwen2.5-7B, and Llama2-13B, we show that beneficial backdoors can achieve high controllability, tamper-resistance, and stealthiness while preserving clean-task performance. Our findings demonstrate new insights that backdoors need not be inherently malicious; when properly designed, they can serve as modular, interpretable, and beneficial building blocks for trustworthy AI systems. Our code and datasets are available at https://github.com/bboylyg/BackdoorLLM/B4G.

Kaiyuan Cui, Yige Li, Yutao Wu et al.

Vision-language models (VLMs) extend large language models (LLMs) with vision encoders, enabling text generation conditioned on both images and text. However, this multimodal integration expands the attack surface by exposing the model to image-based jailbreaks crafted to induce harmful responses. Existing gradient-based jailbreak methods transfer poorly, as adversarial patterns overfit to a single white-box surrogate and fail to generalise to black-box models. In this work, we propose Universal and transferable jailbreak (UltraBreak), a framework that constrains adversarial patterns through transformations and regularisation in the vision space, while relaxing textual targets through semantic-based objectives. By defining its loss in the textual embedding space of the target LLM, UltraBreak discovers universal adversarial patterns that generalise across diverse jailbreak objectives. This combination of vision-level regularisation and semantically guided textual supervision mitigates surrogate overfitting and enables strong transferability across both models and attack targets. Extensive experiments show that UltraBreak consistently outperforms prior jailbreak methods. Further analysis reveals why earlier approaches fail to transfer, highlighting that smoothing the loss landscape via semantic objectives is crucial for enabling universal and transferable jailbreaks. The code is publicly available in our \href{https://github.com/kaiyuanCui/UltraBreak}{GitHub repository}.

Wei Zhao, Zhe Li, Yige Li et al.

Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in cross-modal understanding, but remain vulnerable to adversarial attacks through visual inputs despite robust textual safety mechanisms. These vulnerabilities arise from two core weaknesses: the continuous nature of visual representations, which allows for gradient-based attacks, and the inadequate transfer of text-based safety mechanisms to visual content. We introduce Q-MLLM, a novel architecture that integrates two-level vector quantization to create a discrete bottleneck against adversarial attacks while preserving multimodal reasoning capabilities. By discretizing visual representations at both pixel-patch and semantic levels, Q-MLLM blocks attack pathways and bridges the cross-modal safety alignment gap. Our two-stage training methodology ensures robust learning while maintaining model utility. Experiments demonstrate that Q-MLLM achieves significantly better defense success rate against both jailbreak attacks and toxic image attacks than existing approaches. Notably, Q-MLLM achieves perfect defense success rate (100\%) against jailbreak attacks except in one arguable case, while maintaining competitive performance on multiple utility benchmarks with minimal inference overhead. This work establishes vector quantization as an effective defense mechanism for secure multimodal AI systems without requiring expensive safety-specific fine-tuning or detection overhead. Code is available at https://github.com/Amadeuszhao/QMLLM.

Juncheng Li, Yige Li, Hanxun Huang et al.

Backdoor attacks undermine the reliability and trustworthiness of machine learning systems by injecting hidden behaviors that can be maliciously activated at inference time. While such threats have been extensively studied in unimodal settings, their impact on multimodal foundation models, particularly vision-language models (VLMs), remains largely underexplored. In this work, we introduce \textbf{BackdoorVLM}, the first comprehensive benchmark for systematically evaluating backdoor attacks on VLMs across a broad range of settings. It adopts a unified perspective that injects and analyzes backdoors across core vision-language tasks, including image captioning and visual question answering. BackdoorVLM organizes multimodal backdoor threats into 5 representative categories: targeted refusal, malicious injection, jailbreak, concept substitution, and perceptual hijack. Each category captures a distinct pathway through which an adversary can manipulate a model's behavior. We evaluate these threats using 12 representative attack methods spanning text, image, and bimodal triggers, tested on 2 open-source VLMs and 3 multimodal datasets. Our analysis reveals that VLMs exhibit strong sensitivity to textual instructions, and in bimodal backdoors the text trigger typically overwhelms the image trigger when forming the backdoor mapping. Notably, backdoors involving the textual modality remain highly potent, with poisoning rates as low as 1\% yielding over 90\% success across most tasks. These findings highlight significant, previously underexplored vulnerabilities in current VLMs. We hope that BackdoorVLM can serve as a useful benchmark for analyzing and mitigating multimodal backdoor threats. Code is available at: https://github.com/bin015/BackdoorVLM .

Yunhao Feng, Yifan Ding, Yingshui Tan et al.

Computer-use agents extend language models from text generation to persistent action over tools, files, and execution environments. Unlike chat systems, they maintain state across interactions and translate intermediate outputs into concrete actions. This creates a distinct safety challenge in that harmful behavior may emerge through sequences of individually plausible steps, including intermediate actions that appear locally acceptable but collectively lead to unauthorized actions. We present \textbf{AgentHazard}, a benchmark for evaluating harmful behavior in computer-use agents. AgentHazard contains \textbf{2,653} instances spanning diverse risk categories and attack strategies. Each instance pairs a harmful objective with a sequence of operational steps that are locally legitimate but jointly induce unsafe behavior. The benchmark evaluates whether agents can recognize and interrupt harm arising from accumulated context, repeated tool use, intermediate actions, and dependencies across steps. We evaluate AgentHazard on Claude Code, OpenClaw, and IFlow using mostly open or openly deployable models from the Qwen3, Kimi, GLM, and DeepSeek families. Our experimental results indicate that current systems remain highly vulnerable. In particular, when powered by Qwen3-Coder, Claude Code exhibits an attack success rate of \textbf{73.63\%}, suggesting that model alignment alone does not reliably guarantee the safety of autonomous agents.

Zhe Li, Wei Zhao, Yige Li et al.

Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their deployment is frequently undermined by undesirable behaviors such as generating harmful content, factual inaccuracies, and societal biases. Diagnosing the root causes of these failures poses a critical challenge for AI safety. Existing attribution methods, particularly those based on parameter gradients, often fall short due to prohibitive noisy signals and computational complexity. In this work, we introduce a novel and efficient framework that diagnoses a range of undesirable LLM behaviors by analyzing representation and its gradients, which operates directly in the model's activation space to provide a semantically meaningful signal linking outputs to their training data. We systematically evaluate our method for tasks that include tracking harmful content, detecting backdoor poisoning, and identifying knowledge contamination. The results demonstrate that our approach not only excels at sample-level attribution but also enables fine-grained token-level analysis, precisely identifying the specific samples and phrases that causally influence model behavior. This work provides a powerful diagnostic tool to understand, audit, and ultimately mitigate the risks associated with LLMs. The code is available at https://github.com/plumprc/RepT.

Yige Li, Xixiang Lyu, Xingjun Ma et al.

Deep neural networks (DNNs) have been found to be vulnerable to backdoor attacks, raising security concerns about their deployment in mission-critical applications. While existing defense methods have demonstrated promising results, it is still not clear how to effectively remove backdoor-associated neurons in backdoored DNNs. In this paper, we propose a novel defense called \emph{Reconstructive Neuron Pruning} (RNP) to expose and prune backdoor neurons via an unlearning and then recovering process. Specifically, RNP first unlearns the neurons by maximizing the model's error on a small subset of clean samples and then recovers the neurons by minimizing the model's error on the same data. In RNP, unlearning is operated at the neuron level while recovering is operated at the filter level, forming an asymmetric reconstructive learning procedure. We show that such an asymmetric process on only a few clean samples can effectively expose and prune the backdoor neurons implanted by a wide range of attacks, achieving a new state-of-the-art defense performance. Moreover, the unlearned model at the intermediate step of our RNP can be directly used to improve other backdoor defense tasks including backdoor removal, trigger recovery, backdoor label detection, and backdoor sample detection. Code is available at \url{https://github.com/bboylyg/RNP}.

Yige Li, Xixiang Lyu, Nodens Koren et al.

Backdoor attack has emerged as a major security threat to deep neural networks (DNNs). While existing defense methods have demonstrated promising results on detecting or erasing backdoors, it is still not clear whether robust training methods can be devised to prevent the backdoor triggers being injected into the trained model in the first place. In this paper, we introduce the concept of \emph{anti-backdoor learning}, aiming to train \emph{clean} models given backdoor-poisoned data. We frame the overall learning process as a dual-task of learning the \emph{clean} and the \emph{backdoor} portions of data. From this view, we identify two inherent characteristics of backdoor attacks as their weaknesses: 1) the models learn backdoored data much faster than learning with clean data, and the stronger the attack the faster the model converges on backdoored data; 2) the backdoor task is tied to a specific class (the backdoor target class). Based on these two weaknesses, we propose a general learning scheme, Anti-Backdoor Learning (ABL), to automatically prevent backdoor attacks during training. ABL introduces a two-stage \emph{gradient ascent} mechanism for standard training to 1) help isolate backdoor examples at an early training stage, and 2) break the correlation between backdoor examples and the target class at a later training stage. Through extensive experiments on multiple benchmark datasets against 10 state-of-the-art attacks, we empirically show that ABL-trained models on backdoor-poisoned data achieve the same performance as they were trained on purely clean data. Code is available at \url{https://github.com/bboylyg/ABL}.

Yige Li, Xixiang Lyu, Nodens Koren et al.

Deep neural networks (DNNs) are known vulnerable to backdoor attacks, a training time attack that injects a trigger pattern into a small proportion of training data so as to control the model's prediction at the test time. Backdoor attacks are notably dangerous since they do not affect the model's performance on clean examples, yet can fool the model to make incorrect prediction whenever the trigger pattern appears during testing. In this paper, we propose a novel defense framework Neural Attention Distillation (NAD) to erase backdoor triggers from backdoored DNNs. NAD utilizes a teacher network to guide the finetuning of the backdoored student network on a small clean subset of data such that the intermediate-layer attention of the student network aligns with that of the teacher network. The teacher network can be obtained by an independent finetuning process on the same clean subset. We empirically show, against 6 state-of-the-art backdoor attacks, NAD can effectively erase the backdoor triggers using only 5\% clean training data without causing obvious performance degradation on clean examples. Code is available in https://github.com/bboylyg/NAD.

Shen Dong, Shaochen Xu, Pengfei He et al.

Agents powered by large language models (LLMs) have demonstrated strong capabilities in a wide range of complex, real-world applications. However, LLM agents with a compromised memory bank may easily produce harmful outputs when the past records retrieved for demonstration are malicious. In this paper, we propose a novel Memory INJection Attack, MINJA, without assuming that the attacker can directly modify the memory bank of the agent. The attacker injects malicious records into the memory bank by only interacting with the agent via queries and output observations. These malicious records are designed to elicit a sequence of malicious reasoning steps corresponding to a different target query during the agent's execution of the victim user's query. Specifically, we introduce a sequence of bridging steps to link victim queries to the malicious reasoning steps. During the memory injection, we propose an indication prompt that guides the agent to autonomously generate similar bridging steps, with a progressive shortening strategy that gradually removes the indication prompt, such that the malicious record will be easily retrieved when processing later victim queries. Our extensive experiments across diverse agents demonstrate the effectiveness of MINJA in compromising agent memory. With minimal requirements for execution, MINJA enables any user to influence agent memory, highlighting the risk.

Nay Myat Min, Long H. Pham, Yige Li et al.

Large Language Models (LLMs) are vulnerable to backdoor attacks that manipulate outputs via hidden triggers. Existing defense methods--designed for vision/text classification tasks--fail for text generation. We propose Internal Consistency Regularization (CROW), a defense leveraging the observation that backdoored models exhibit unstable layer-wise hidden representations when triggered, while clean models show smooth transitions. CROW enforces consistency across layers via adversarial perturbations and regularization during finetuning, neutralizing backdoors without requiring clean reference models or trigger knowledge--only a small clean dataset. Experiments across Llama-2 (7B, 13B), CodeLlama (7B, 13B), and Mistral-7B demonstrate CROW's effectiveness: it achieves significant reductions in attack success rates across diverse backdoor strategies (sentiment steering, targeted refusal, code injection) while preserving generative performance. CROW's architecture-agnostic design enables practical deployment.

Xingjun Ma, Yifeng Gao, Yixu Wang et al.

The rapid advancement of large models, driven by their exceptional abilities in learning and generalization through large-scale pre-training, has reshaped the landscape of Artificial Intelligence (AI). These models are now foundational to a wide range of applications, including conversational AI, recommendation systems, autonomous driving, content generation, medical diagnostics, and scientific discovery. However, their widespread deployment also exposes them to significant safety risks, raising concerns about robustness, reliability, and ethical implications. This survey provides a systematic review of current safety research on large models, covering Vision Foundation Models (VFMs), Large Language Models (LLMs), Vision-Language Pre-training (VLP) models, Vision-Language Models (VLMs), Diffusion Models (DMs), and large-model-powered Agents. Our contributions are summarized as follows: (1) We present a comprehensive taxonomy of safety threats to these models, including adversarial attacks, data poisoning, backdoor attacks, jailbreak and prompt injection attacks, energy-latency attacks, data and model extraction attacks, and emerging agent-specific threats. (2) We review defense strategies proposed for each type of attacks if available and summarize the commonly used datasets and benchmarks for safety research. (3) Building on this, we identify and discuss the open challenges in large model safety, emphasizing the need for comprehensive safety evaluations, scalable and effective defense mechanisms, and sustainable data practices. More importantly, we highlight the necessity of collective efforts from the research community and international collaboration. Our work can serve as a useful reference for researchers and practitioners, fostering the ongoing development of comprehensive defense systems and platforms to safeguard AI models.

Yujing Jiang, Xingjun Ma, Sarah Monazam Erfani et al.

Backdoor attacks present a substantial security concern for deep learning models, especially those utilized in applications critical to safety and security. These attacks manipulate model behavior by embedding a hidden trigger during the training phase, allowing unauthorized control over the model's output during inference time. Although numerous defenses exist for image classification models, there is a conspicuous absence of defenses tailored for time series data, as well as an end-to-end solution capable of training clean models on poisoned data. To address this gap, this paper builds upon Anti-Backdoor Learning (ABL) and introduces an innovative method, End-to-End Anti-Backdoor Learning (E2ABL), for robust training against backdoor attacks. Unlike the original ABL, which employs a two-stage training procedure, E2ABL accomplishes end-to-end training through an additional classification head linked to the shallow layers of a Deep Neural Network (DNN). This secondary head actively identifies potential backdoor triggers, allowing the model to dynamically cleanse these samples and their corresponding labels during training. Our experiments reveal that E2ABL significantly improves on existing defenses and is effective against a broad range of backdoor attacks in both image and time series domains.

Xiao Li, Xiang Zheng, Yifeng Gao et al.

Embodied Artificial Intelligence (Embodied AI) integrates perception, cognition, planning, and interaction into agents that operate in open-world, safety-critical environments. As these systems gain autonomy and enter domains such as transportation, healthcare, and industrial or assistive robotics, ensuring their safety becomes both technically challenging and socially indispensable. Unlike digital AI systems, embodied agents must act under uncertain sensing, incomplete knowledge, and dynamic human-robot interactions, where failures can directly lead to physical harm. This survey provides a comprehensive and structured review of safety research in embodied AI, examining attacks and defenses across the full embodied pipeline, from perception and cognition to planning, action and interaction, and agentic system. We introduce a multi-level taxonomy that unifies fragmented lines of work and connects embodied-specific safety findings with broader advances in vision, language, and multimodal foundation models. Our review synthesizes insights from over 400 papers spanning adversarial, backdoor, jailbreak, and hardware-level attacks; attack detection, safe training and robust inference; and risk-aware human-agent interaction. This analysis reveals several overlooked challenges, including the fragility of multimodal perception fusion, the instability of planning under jailbreak attacks, and the trustworthiness of human-agent interaction in open-ended scenarios. By organizing the field into a coherent framework and identifying critical research gaps, this survey provides a roadmap for building embodied agents that are not only capable and autonomous but also safe, robust, and reliable in real-world deployment.

Yige Li, Peihai Jiang, Jun Sun et al.

Large Language Models (LLMs) have demonstrated significant success across diverse applications. However, enforcing content restrictions remains a significant challenge due to their expansive output space. One aspect of content restriction is preventing LLMs from generating harmful content via model alignment approaches such as supervised fine-tuning (SFT). Yet, the need for content restriction may vary significantly across user groups, change rapidly over time, and not always align with general definitions of harmfulness. Applying SFT to each of these specific use cases is impractical due to the high computational, data, and storage demands. Motivated by this need, we propose a new task called \textit{Adaptive Content Restriction} (AdaCoRe), which focuses on lightweight strategies -- methods without model fine-tuning -- to prevent deployed LLMs from generating restricted terms for specific use cases. We propose the first method for AdaCoRe, named \textit{Suffix Optimization (SOP)}, which appends a short, optimized suffix to any prompt to a) prevent a target LLM from generating a set of restricted terms, while b) preserving the output quality. To evaluate AdaCoRe approaches, including our SOP, we create a new \textit{Content Restriction Benchmark} (CoReBench), which contains 400 prompts for 80 restricted terms across 8 carefully selected categories. We demonstrate the effectiveness of SOP on CoReBench, which outperforms the system-level baselines such as system suffix by 15\%, 17\%, 10\%, 9\%, and 6\% on average restriction rates for Gemma2-2B, Mistral-7B, Vicuna-7B, Llama3-8B, and Llama3.1-8B, respectively. We also demonstrate that SOP is effective on POE, an online platform hosting various commercial LLMs, highlighting its practicality in real-world scenarios.

Yiming Sun, Yige Li, Amy Kuceyeski et al.

Spectral density matrix estimation of multivariate time series is a classical problem in time series and signal processing. In modern neuroscience, spectral density based metrics are commonly used for analyzing functional connectivity among brain regions. In this paper, we develop a non-asymptotic theory for regularized estimation of high-dimensional spectral density matrices of Gaussian and linear processes using thresholded versions of averaged periodograms. Our theoretical analysis ensures that consistent estimation of spectral density matrix of a $p$-dimensional time series using $n$ samples is possible under high-dimensional regime $\log p / n \rightarrow 0$ as long as the true spectral density is approximately sparse. A key technical component of our analysis is a new concentration inequality of average periodogram around its expectation, which is of independent interest. Our estimation consistency results complement existing results for shrinkage based estimators of multivariate spectral density, which require no assumption on sparsity but only ensure consistent estimation in a regime $p^2/n \rightarrow 0$. In addition, our proposed thresholding based estimators perform consistent and automatic edge selection when learning coherence networks among the components of a multivariate time series. We demonstrate the advantage of our estimators using simulation studies and a real data application on functional connectivity analysis with fMRI data.