TokenSwap: Backdoor Attack on the Compositional Understanding of Large Vision-Language Models

• URL: http://arxiv.org/abs/2509.24566v1
• Date: Mon, 29 Sep 2025 10:19:22 GMT
• Title: TokenSwap: Backdoor Attack on the Compositional Understanding of Large Vision-Language Models
• Authors: Zhifang Zhang, Qiqi Tao, Jiaqi Lv, Na Zhao, Lei Feng, Joey Tianyi Zhou,
• Abstract summary: We introduce TokenSwap, a more evasive and stealthy backdoor attack on large vision-language models (LVLMs)<n>Instead of enforcing a fixed targeted content, TokenSwap subtly disrupts the understanding of object relationships in text.<n> TokenSwap achieves high attack success rates while maintaining superior evasiveness and stealthiness.
• Score: 57.32952956674526
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large vision-language models (LVLMs) have achieved impressive performance across a wide range of vision-language tasks, while they remain vulnerable to backdoor attacks. Existing backdoor attacks on LVLMs aim to force the victim model to generate a predefined target pattern, which is either inserted into or replaces the original content. We find that these fixed-pattern attacks are relatively easy to detect, because the attacked LVLM tends to memorize such frequent patterns in the training dataset, thereby exhibiting overconfidence on these targets given poisoned inputs. To address these limitations, we introduce TokenSwap, a more evasive and stealthy backdoor attack that focuses on the compositional understanding capabilities of LVLMs. Instead of enforcing a fixed targeted content, TokenSwap subtly disrupts the understanding of object relationships in text. Specifically, it causes the backdoored model to generate outputs that mention the correct objects in the image but misrepresent their relationships (i.e., bags-of-words behavior). During training, TokenSwap injects a visual trigger into selected samples and simultaneously swaps the grammatical roles of key tokens in the corresponding textual answers. However, the poisoned samples exhibit only subtle differences from the original ones, making it challenging for the model to learn the backdoor behavior. To address this, TokenSwap employs an adaptive token-weighted loss that explicitly emphasizes the learning of swapped tokens, such that the visual triggers and bags-of-words behavior are associated. Extensive experiments demonstrate that TokenSwap achieves high attack success rates while maintaining superior evasiveness and stealthiness across multiple benchmarks and various LVLM architectures.

Related papers

• MTAttack: Multi-Target Backdoor Attacks against Large Vision-Language Models [52.37749859972453]
  We propose MTAttack, the first multi-target backdoor attack framework for enforcing accurate multiple trigger-target mappings in LVLMs.<n> Experiments on popular benchmarks demonstrate a high success rate of MTAttack for multi-target attacks.<n>Our attack exhibits strong generalizability across datasets and robustness against backdoor defense strategies.
  arXiv  Detail & Related papers  (2025-11-13T09:00:21Z)
• Backdoor Attack on Vision Language Models with Stealthy Semantic Manipulation [32.24294112337828]
  BadSem is a data poisoning attack that injects backdoors by deliberately misaligning image-text pairs during training.<n>Our experiments show that BadSem achieves over 98% average ASR, generalizes well to out-of-distribution datasets, and can transfer across poisoning modalities.<n>Our findings highlight the urgent need to address semantic vulnerabilities in Vision Language Models for their safer deployment.
  arXiv  Detail & Related papers  (2025-06-08T16:40:40Z)
• SRD: Reinforcement-Learned Semantic Perturbation for Backdoor Defense in VLMs [57.880467106470775]
  Attackers can inject imperceptible perturbations into the training data, causing the model to generate malicious, attacker-controlled captions.<n>We propose Semantic Reward Defense (SRD), a reinforcement learning framework that mitigates backdoor behavior without prior knowledge of triggers.<n>SRD uses a Deep Q-Network to learn policies for applying discrete perturbations to sensitive image regions, aiming to disrupt the activation of malicious pathways.
  arXiv  Detail & Related papers  (2025-06-05T08:22:24Z)
• Robust Anti-Backdoor Instruction Tuning in LVLMs [53.766434746801366]
  We introduce a lightweight, certified-agnostic defense framework for large visual language models (LVLMs)<n>Our framework finetunes only adapter modules and text embedding layers under instruction tuning.<n>Experiments against seven attacks on Flickr30k and MSCOCO demonstrate that ours reduces their attack success rate to nearly zero.
  arXiv  Detail & Related papers  (2025-06-04T01:23:35Z)
• VEAttack: Downstream-agnostic Vision Encoder Attack against Large Vision Language Models [33.120141513366136]
  Large Vision-Language Models (LVLMs) have demonstrated remarkable capabilities in multimodal understanding and generation.<n>Existing effective attacks always focus on task-specific white-box settings.<n>We propose a simple yet effective Vision Attack (VEAttack) which targets the vision encoder of LVLMs only.
  arXiv  Detail & Related papers  (2025-05-23T03:46:04Z)
• Break the Visual Perception: Adversarial Attacks Targeting Encoded Visual Tokens of Large Vision-Language Models [15.029014337718849]
  Large vision-language models (LVLMs) integrate visual information into large language models, showcasing remarkable multi-modal conversational capabilities. In general, LVLMs rely on vision encoders to transform images into visual tokens, which are crucial for the language models to perceive image contents effectively. We propose a non-targeted attack method referred to as VT-Attack, which constructs adversarial examples from multiple perspectives.
  arXiv  Detail & Related papers  (2024-10-09T09:06:56Z)
• Revisiting Backdoor Attacks against Large Vision-Language Models from Domain Shift [104.76588209308666]
  This paper explores backdoor attacks in LVLM instruction tuning across mismatched training and testing domains.<n>We introduce a new evaluation dimension, backdoor domain generalization, to assess attack robustness.<n>We propose a multimodal attribution backdoor attack (MABA) that injects domain-agnostic triggers into critical areas.
  arXiv  Detail & Related papers  (2024-06-27T02:31:03Z)
• VL-Trojan: Multimodal Instruction Backdoor Attacks against Autoregressive Visual Language Models [65.23688155159398]
  Autoregressive Visual Language Models (VLMs) showcase impressive few-shot learning capabilities in a multimodal context. Recently, multimodal instruction tuning has been proposed to further enhance instruction-following abilities. Adversaries can implant a backdoor by injecting poisoned samples with triggers embedded in instructions or images. We propose a multimodal instruction backdoor attack, namely VL-Trojan.
  arXiv  Detail & Related papers  (2024-02-21T14:54:30Z)
• InstructTA: Instruction-Tuned Targeted Attack for Large Vision-Language Models [13.21813503235793]
  Large vision-language models (LVLMs) have demonstrated their incredible capability in image understanding and response generation. In this paper, we formulate a novel and practical targeted attack scenario that the adversary can only know the vision encoder of the victim LVLM. We propose an instruction-tuned targeted attack (dubbed textscInstructTA) to deliver the targeted adversarial attack on LVLMs with high transferability.
  arXiv  Detail & Related papers  (2023-12-04T13:40:05Z)
• CleanCLIP: Mitigating Data Poisoning Attacks in Multimodal Contrastive Learning [63.72975421109622]
  CleanCLIP is a finetuning framework that weakens the learned spurious associations introduced by backdoor attacks. CleanCLIP maintains model performance on benign examples while erasing a range of backdoor attacks on multimodal contrastive learning.
  arXiv  Detail & Related papers  (2023-03-06T17:48:32Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.