Mixture-of-Experts (MoE) architectures are rapidly becoming the standard for building scalable and efficient large language models (LLMs), thanks to their routing mechanism that activates only a subset of experts for each input. However, the modular structure and sparse activation of experts also introduce new security risks, as manipulating a small portion of experts may influence the behavior. In this work, we study a white-box, task-specific model-supply-chain setting and propose BadMoE, a routing-aware backdoor attack against MoE LLMs. We motivate BadMoE through a routing-aware decomposition of the MoE-layer output, which separates the contributions of the selected target experts from those of the remaining experts. This analysis highlights two measurable factors underlying the attack: the routing mass assigned to the target experts and their relative contribution to the layer output. Guided by this insight, BadMoE selects task-underutilized experts and utilizes routing-aware triggers to activate them, enabling reliable activation of task-specific, predefined backdoor behaviors. Specifically, BadMoE involves three steps: 1) identifying experts with low usage on a target-task calibration set, 2) optimizing a routing-aware trigger toward these experts, and 3) associating the selected-expert activations with attacker-specified behavior. We evaluate BadMoE on Mixtral, OLMoE, and Deepseek across sentiment, topic, and emotion classification, sentiment steering, and targeted refusal.
Graphical User Interface (GUI) agents powered by Large Vision-Language Models (LVLMs) have emerged as a revolutionary approach to automating human-machine interactions, capable of autonomously operating personal devices (e.g., mobile phones) or applications within the device to perform complex real-world tasks in a human-like manner. However, their close integration with personal devices raises significant security concerns, with many threats, including backdoor attacks, remaining largely unexplored. This work reveals that the visual grounding of GUI agent-mapping textual plans to GUI elements-can introduce vulnerabilities, enabling new types of backdoor attacks. With backdoor attack targeting visual grounding, the agent's behavior can be compromised even when given correct task-solving plans. To validate this vulnerability, we propose VisualTrap, a method that can hijack the grounding by misleading the agent to locate textual plans to trigger locations instead of the intended targets. VisualTrap uses the common method of injecting poisoned data for attacks, and does so during the pre-training of visual grounding to ensure practical feasibility of attacking. Empirical results show that VisualTrap can effectively hijack visual grounding with as little as 5% poisoned data and highly stealthy visual triggers (invisible to the human eye); and the attack can be generalized to downstream tasks, even after clean fine-tuning. Moreover, the injected trigger can remain effective across different GUI environments, e.g., being trained on mobile/web and generalizing to desktop environments. These findings underscore the urgent need for further research on backdoor attack risks in GUI agents.
Deep Reinforcement Learning (DRL) systems are increasingly used in safety-critical applications, yet their security remains severely underexplored. This work investigates backdoor attacks, which implant hidden triggers that cause malicious actions only when specific inputs appear in the observation space. Existing DRL backdoor research focuses solely on training-time attacks requiring unrealistic access to the training pipeline. In contrast, we reveal critical vulnerabilities across the DRL supply chain where backdoors can be embedded with significantly reduced adversarial privileges. We introduce two novel attacks: (1) TrojanentRL, which exploits component-level flaws to implant a persistent backdoor that survives full model retraining; and (2) InfrectroRL, a post-training backdoor attack which requires no access to training, validation, nor test data. Empirical and analytical evaluations across six Atari environments show our attacks rival state-of-the-art training-time backdoor attacks while operating under much stricter adversarial constraints. We also demonstrate that InfrectroRL further evades two leading DRL backdoor defenses. These findings challenge the current research focus and highlight the urgent need for robust defenses.
Backdoor attacks have emerged as a critical security threat against deep neural networks in recent years. The majority of existing backdoor attacks focus on targeted backdoor attacks, where trigger is strongly associated to specific malicious behavior. Various backdoor detection methods depend on this inherent property and shows effective results in identifying and mitigating such targeted attacks. However, a purely untargeted attack in backdoor scenarios is, in some sense, self-weakening, since the target nature is what makes backdoor attacks so powerful. In light of this, we introduce a novel Constrained Untargeted Backdoor Attack (CUBA), which combines the flexibility of untargeted attacks with the intentionality of targeted attacks. The compromised model, when presented with backdoor images, will classify them into random classes within a constrained range of target classes selected by the attacker. This combination of randomness and determinedness enables the proposed untargeted backdoor attack to natively circumvent existing backdoor defense methods. To implement the untargeted backdoor attack under controlled flexibility, we propose to apply logit normalization on cross-entropy loss with flipped one-hot labels. By constraining the logit during training, the compromised model will show a uniform distribution across selected target classes, resulting in controlled untargeted attack. Extensive experiments demonstrate the effectiveness of the proposed CUBA on different datasets.
With the growing integration of vision-language models (VLMs), mobile agents are now widely used for tasks like UI automation and camera-based user assistance. These agents are often fine-tuned on limited user-generated datasets, leaving them vulnerable to covert threats during the training process. In this work we present GHOST, the first clean-label backdoor attack specifically designed for mobile agents built upon VLMs. Our method manipulates only the visual inputs of a portion of the training samples - without altering their corresponding labels or instructions - thereby injecting malicious behaviors into the model. Once fine-tuned with this tampered data, the agent will exhibit attacker-controlled responses when a specific visual trigger is introduced at inference time. The core of our approach lies in aligning the gradients of poisoned samples with those of a chosen target instance, embedding backdoor-relevant features into the poisoned training data. To maintain stealth and enhance robustness, we develop three realistic visual triggers: static visual patches, dynamic motion cues, and subtle low-opacity overlays. We evaluate our method across six real-world Android apps and three VLM architectures adapted for mobile use. Results show that our attack achieves high attack success rates (up to 94.67 percent) while maintaining high clean-task performance (FSR up to 95.85 percent). Additionally, ablation studies shed light on how various design choices affect the efficacy and concealment of the attack. Overall, this work is the first to expose critical security flaws in VLM-based mobile agents, highlighting their susceptibility to clean-label backdoor attacks and the urgent need for effective defense mechanisms in their training pipelines.
The capability of generative diffusion models (DMs) like Stable Diffusion (SD) in replicating training data could be taken advantage of by attackers to launch the Copyright Infringement Attack, with duplicated poisoned image-text pairs. SilentBadDiffusion (SBD) is a method proposed recently, which shew outstanding performance in attacking SD in text-to-image tasks. However, the feasible data resources in this area are still limited, some of them are even constrained or prohibited due to the issues like copyright ownership or inappropriate contents; And not all of the images in current datasets are suitable for the proposed attacking methods; Besides, the state-of-the-art (SoTA) performance of SBD is far from ideal when few generated poisoning samples could be adopted for attacks. In this paper, we raised new datasets accessible for researching in attacks like SBD, and proposed Multi-Element (ME) attack method based on SBD by increasing the number of poisonous visual-text elements per poisoned sample to enhance the ability of attacking, while importing Discrete Cosine Transform (DCT) for the poisoned samples to maintain the stealthiness. The Copyright Infringement Rate (CIR) / First Attack Epoch (FAE) we got on the two new datasets were 16.78% / 39.50 and 51.20% / 23.60, respectively close to or even outperformed benchmark Pokemon and Mijourney datasets. In condition of low subsampling ratio (5%, 6 poisoned samples), MESI and DCT earned CIR / FAE of 0.23% / 84.00 and 12.73% / 65.50, both better than original SBD, which failed to attack at all.
Vision Language Models (VLMs) have shown remarkable performance, but are also vulnerable to backdoor attacks whereby the adversary can manipulate the model's outputs through hidden triggers. Prior attacks primarily rely on single-modality triggers, leaving the crucial cross-modal fusion nature of VLMs largely unexplored. Unlike prior work, we identify a novel attack surface that leverages cross-modal semantic mismatches as implicit triggers. Based on this insight, we propose BadSem (Backdoor Attack with Semantic Manipulation), a data poisoning attack that injects stealthy backdoors by deliberately misaligning image-text pairs during training. To perform the attack, we construct SIMBad, a dataset tailored for semantic manipulation involving color and object attributes. Extensive experiments across four widely used VLMs show that BadSem achieves over 98% average ASR, generalizes well to out-of-distribution datasets, and can transfer across poisoning modalities. Our detailed analysis using attention visualization shows that backdoored models focus on semantically sensitive regions under mismatched conditions while maintaining normal behavior on clean inputs. To mitigate the attack, we try two defense strategies based on system prompt and supervised fine-tuning but find that both of them fail to mitigate the semantic backdoor. Our findings highlight the urgent need to address semantic vulnerabilities in VLMs for their safer deployment.
Deep speech classification tasks, including keyword spotting and speaker verification, are vital in speech-based human-computer interaction. Recently, the security of these technologies has been revealed to be susceptible to backdoor attacks. Specifically, attackers use noisy disruption triggers and speech element triggers to produce poisoned speech samples that train models to become vulnerable. However, these methods typically create only a limited number of backdoors due to the inherent constraints of the trigger function. In this paper, we propose that speech backdoor attacks can strategically focus on speech elements such as timbre and emotion, leveraging the Speech Large Language Model (SLLM) to generate diverse triggers. Increasing the number of triggers may disproportionately elevate the poisoning rate, resulting in higher attack costs and a lower success rate per trigger. We introduce the Multiple Gradient Descent Algorithm (MGDA) as a mitigation strategy to address this challenge. The proposed attack is called the Speech Prompt Backdoor Attack (SPBA). Building on this foundation, we conducted attack experiments on two speech classification tasks, demonstrating that SPBA shows significant trigger effectiveness and achieves exceptional performance in attack metrics.
Graph recommendation systems have been widely studied due to their ability to effectively capture the complex interactions between users and items. However, these systems also exhibit certain vulnerabilities when faced with attacks. The prevailing shilling attack methods typically manipulate recommendation results by injecting a large number of fake nodes and edges. However, such attack strategies face two primary challenges: low stealth and high destructiveness. To address these challenges, this paper proposes a novel graph backdoor attack method that aims to enhance the exposure of target items to the target user in a covert manner, without affecting other unrelated nodes. Specifically, we design a single-node trigger generator, which can effectively expose multiple target items to the target user by inserting only one fake user node. Additionally, we introduce constraint conditions between the target nodes and irrelevant nodes to mitigate the impact of fake nodes on the recommendation system's performance. Experimental results show that the exposure of the target items reaches no less than 50% in 99% of the target users, while the impact on the recommendation system's performance is controlled within approximately 5%.
Diffusion models have achieved remarkable progress in both image generation and editing. However, recent studies have revealed their vulnerability to backdoor attacks, in which specific patterns embedded in the input can manipulate the model's behavior. Most existing research in this area has proposed attack frameworks focused on the image generation pipeline, leaving backdoor attacks in image editing relatively unexplored. Among the few studies targeting image editing, most utilize visible triggers, which are impractical because they introduce noticeable alterations to the input image before editing. In this paper, we propose a novel attack framework that embeds invisible triggers into the image editing process via poisoned training data. We leverage off-the-shelf deep watermarking models to encode imperceptible watermarks as backdoor triggers. Our goal is to make the model produce the predefined backdoor target when it receives watermarked inputs, while editing clean images normally according to the given prompt. With extensive experiments across different watermarking models, the proposed method achieves promising attack success rates. In addition, the analysis results of the watermark characteristics in term of backdoor attack further support the effectiveness of our approach. The code is available at:https://github.com/aiiu-lab/BackdoorImageEditing
Heterogeneous Graph Neural Networks (HGNNs) excel in modeling complex, multi-typed relationships across diverse domains, yet their vulnerability to backdoor attacks remains unexplored. To address this gap, we conduct the first investigation into the susceptibility of HGNNs to existing graph backdoor attacks, revealing three critical issues: (1) high attack budget required for effective backdoor injection, (2) inefficient and unreliable backdoor activation, and (3) inaccurate attack effectiveness evaluation. To tackle these issues, we propose the Heterogeneous Graph Backdoor Attack (HGBA), the first backdoor attack specifically designed for HGNNs, introducing a novel relation-based trigger mechanism that establishes specific connections between a strategically selected trigger node and poisoned nodes via the backdoor metapath. HGBA achieves efficient and stealthy backdoor injection with minimal structural modifications and supports easy backdoor activation through two flexible strategies: Self-Node Attack and Indiscriminate Attack. Additionally, we improve the ASR measurement protocol, enabling a more accurate assessment of attack effectiveness. Extensive experiments demonstrate that HGBA far surpasses multiple state-of-the-art graph backdoor attacks in black-box settings, efficiently attacking HGNNs with low attack budgets. Ablation studies show that the strength of HBGA benefits from our trigger node selection method and backdoor metapath selection strategy. In addition, HGBA shows superior robustness against node feature perturbations and multiple types of existing graph backdoor defense mechanisms. Finally, extension experiments demonstrate that the relation-based trigger mechanism can effectively extend to tasks in homogeneous graph scenarios, thereby posing severe threats to broader security-critical domains.
Robots need task planning methods to achieve goals that require more than individual actions. Recently, large language models (LLMs) have demonstrated impressive performance in task planning. LLMs can generate a step-by-step solution using a description of actions and the goal. Despite the successes in LLM-based task planning, there is limited research studying the security aspects of those systems. In this paper, we develop Robo-Troj, the first multi-trigger backdoor attack for LLM-based task planners, which is the main contribution of this work. As a multi-trigger attack, Robo-Troj is trained to accommodate the diversity of robot application domains. For instance, one can use unique trigger words, e.g., "herical", to activate a specific malicious behavior, e.g., cutting hand on a kitchen robot. In addition, we develop an optimization method for selecting the trigger words that are most effective. Through demonstrating the vulnerability of LLM-based planners, we aim to promote the development of secured robot systems.
Backdoor attacks embed malicious triggers into training data, enabling attackers to manipulate neural network behavior during inference while maintaining high accuracy on benign inputs. However, existing backdoor attacks face limitations manifesting in excessive reliance on training data, poor stealth, and instability, which hinder their effectiveness in real-world applications. Therefore, this paper introduces ShadowPrint, a versatile backdoor attack that targets feature embeddings within neural networks to achieve high ASRs and stealthiness. Unlike traditional approaches, ShadowPrint reduces reliance on training data access and operates effectively with exceedingly low poison rates (as low as 0.01%). It leverages a clustering-based optimization strategy to align feature embeddings, ensuring robust performance across diverse scenarios while maintaining stability and stealth. Extensive evaluations demonstrate that ShadowPrint achieves superior ASR (up to 100%), steady CA (with decay no more than 1% in most cases), and low DDR (averaging below 5%) across both clean-label and dirty-label settings, and with poison rates ranging from as low as 0.01% to 0.05%, setting a new standard for backdoor attack capabilities and emphasizing the need for advanced defense strategies focused on feature space manipulations.
Backdoor attacks, or trojans, pose a security risk by concealing undesirable behavior in deep neural network models. Open-source neural networks are downloaded from the internet daily, possibly containing backdoors, and third-party model developers are common. To advance research on backdoor attack mitigation, we develop several trojans for deep reinforcement learning (DRL) agents. We focus on in-distribution triggers, which occur within the agent's natural data distribution, since they pose a more significant security threat than out-of-distribution triggers due to their ease of activation by the attacker during model deployment. We implement backdoor attacks in four reinforcement learning (RL) environments: LavaWorld, Randomized LavaWorld, Colorful Memory, and Modified Safety Gymnasium. We train various models, both clean and backdoored, to characterize these attacks. We find that in-distribution triggers can require additional effort to implement and be more challenging for models to learn, but are nevertheless viable threats in DRL even using basic data poisoning attacks.
In recent years, deep learning-based Monocular Depth Estimation (MDE) models have been widely applied in fields such as autonomous driving and robotics. However, their vulnerability to backdoor attacks remains unexplored. To fill the gap in this area, we conduct a comprehensive investigation of backdoor attacks against MDE models. Typically, existing backdoor attack methods can not be applied to MDE models. This is because the label used in MDE is in the form of a depth map. To address this, we propose BadDepth, the first backdoor attack targeting MDE models. BadDepth overcomes this limitation by selectively manipulating the target object's depth using an image segmentation model and restoring the surrounding areas via depth completion, thereby generating poisoned datasets for object-level backdoor attacks. To improve robustness in physical world scenarios, we further introduce digital-to-physical augmentation to adapt to the domain gap between the physical world and the digital domain. Extensive experiments on multiple models validate the effectiveness of BadDepth in both the digital domain and the physical world, without being affected by environmental factors.
Vision-Language-Action (VLA) models have advanced robotic control by enabling end-to-end decision-making directly from multimodal inputs. However, their tightly coupled architectures expose novel security vulnerabilities. Unlike traditional adversarial perturbations, backdoor attacks represent a stealthier, persistent, and practically significant threat-particularly under the emerging Training-as-a-Service paradigm-but remain largely unexplored in the context of VLA models. To address this gap, we propose BadVLA, a backdoor attack method based on Objective-Decoupled Optimization, which for the first time exposes the backdoor vulnerabilities of VLA models. Specifically, it consists of a two-stage process: (1) explicit feature-space separation to isolate trigger representations from benign inputs, and (2) conditional control deviations that activate only in the presence of the trigger, while preserving clean-task performance. Empirical results on multiple VLA benchmarks demonstrate that BadVLA consistently achieves near-100% attack success rates with minimal impact on clean task accuracy. Further analyses confirm its robustness against common input perturbations, task transfers, and model fine-tuning, underscoring critical security vulnerabilities in current VLA deployments. Our work offers the first systematic investigation of backdoor vulnerabilities in VLA models, highlighting an urgent need for secure and trustworthy embodied model design practices. We have released the project page at https://badvla-project.github.io/.
Traffic sign recognition (TSR) systems are crucial for autonomous driving but are vulnerable to backdoor attacks. Existing physical backdoor attacks either lack stealth, provide inflexible attack control, or ignore emerging Vision-Large-Language-Models (VLMs). In this paper, we introduce FIGhost, the first physical-world backdoor attack leveraging fluorescent ink as triggers. Fluorescent triggers are invisible under normal conditions and activated stealthily by ultraviolet light, providing superior stealthiness, flexibility, and untraceability. Inspired by real-world graffiti, we derive realistic trigger shapes and enhance their robustness via an interpolation-based fluorescence simulation algorithm. Furthermore, we develop an automated backdoor sample generation method to support three attack objectives. Extensive evaluations in the physical world demonstrate FIGhost's effectiveness against state-of-the-art detectors and VLMs, maintaining robustness under environmental variations and effectively evading existing defenses.
Recent studies have revealed that text-to-image diffusion models are vulnerable to backdoor attacks, where attackers implant stealthy textual triggers to manipulate model outputs. Previous backdoor detection methods primarily focus on the static features of backdoor samples. However, a vital property of diffusion models is their inherent dynamism. This study introduces a novel backdoor detection perspective named Dynamic Attention Analysis (DAA), showing that these dynamic characteristics serve as better indicators for backdoor detection. Specifically, by examining the dynamic evolution of cross-attention maps, we observe that backdoor samples exhibit distinct feature evolution patterns at the $<$EOS$>$ token compared to benign samples. To quantify these dynamic anomalies, we first introduce DAA-I, which treats the tokens' attention maps as spatially independent and measures dynamic feature using the Frobenius norm. Furthermore, to better capture the interactions between attention maps and refine the feature, we propose a dynamical system-based approach, referred to as DAA-S. This model formulates the spatial correlations among attention maps using a graph-based state equation and we theoretically analyze the global asymptotic stability of this method. Extensive experiments across five representative backdoor attack scenarios demonstrate that our approach significantly surpasses existing detection methods, achieving an average F1 Score of 79.49% and an AUC of 87.67%. The code is available at https://github.com/Robin-WZQ/DAA.
Federated learning (FL) enhances privacy and reduces communication cost for resource-constrained edge clients by supporting distributed model training at the edge. However, the heterogeneous nature of such devices produces diverse, non-independent, and identically distributed (non-IID) data, making the detection of backdoor attacks more challenging. In this paper, we propose a novel federated representative-attention-based defense mechanism, named FeRA, that leverages cross-client attention over internal feature representations to distinguish benign from malicious clients. FeRA computes an anomaly score based on representation reconstruction errors, effectively identifying clients whose internal activations significantly deviate from the group consensus. Our evaluation demonstrates FeRA's robustness across various FL scenarios, including challenging non-IID data distributions typical of edge devices. Experimental results show that it effectively reduces backdoor attack success rates while maintaining high accuracy on the main task. The method is model-agnostic, attack-agnostic, and does not require labeled reference data, making it well suited to heterogeneous and resource-limited edge deployments.
This paper focuses on implanting multiple heterogeneous backdoor triggers in bridge-based diffusion models designed for complex and arbitrary input distributions. Existing backdoor formulations mainly address single-attack scenarios and are limited to Gaussian noise input models. To fill this gap, we propose MixBridge, a novel diffusion Schr\"odinger bridge (DSB) framework to cater to arbitrary input distributions (taking I2I tasks as special cases). Beyond this trait, we demonstrate that backdoor triggers can be injected into MixBridge by directly training with poisoned image pairs. This eliminates the need for the cumbersome modifications to stochastic differential equations required in previous studies, providing a flexible tool to study backdoor behavior for bridge models. However, a key question arises: can a single DSB model train multiple backdoor triggers? Unfortunately, our theory shows that when attempting this, the model ends up following the geometric mean of benign and backdoored distributions, leading to performance conflict across backdoor tasks. To overcome this, we propose a Divide-and-Merge strategy to mix different bridges, where models are independently pre-trained for each specific objective (Divide) and then integrated into a unified model (Merge). In addition, a Weight Reallocation Scheme (WRS) is also designed to enhance the stealthiness of MixBridge. Empirical studies across diverse generation tasks speak to the efficacy of MixBridge.