VLSBench: Unveiling Visual Leakage in Multimodal Safety

• URL: http://arxiv.org/abs/2411.19939v3
• Date: Sat, 17 May 2025 15:14:14 GMT
• Title: VLSBench: Unveiling Visual Leakage in Multimodal Safety
• Authors: Xuhao Hu, Dongrui Liu, Hao Li, Xuanjing Huang, Jing Shao,
• Abstract summary: Safety concerns of Multimodal large language models (MLLMs) have gradually become an important problem in various applications.<n>Previous works indicate a counterintuitive phenomenon that using textual unlearning to align MLLMs achieves comparable safety performances with MLLMs aligned with image text pairs.
• Score: 39.344623032631475
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Safety concerns of Multimodal large language models (MLLMs) have gradually become an important problem in various applications. Surprisingly, previous works indicate a counterintuitive phenomenon that using textual unlearning to align MLLMs achieves comparable safety performances with MLLMs aligned with image text pairs. To explain such a phenomenon, we discover a Visual Safety Information Leakage (VSIL) problem in existing multimodal safety benchmarks, i.e., the potentially risky content in the image has been revealed in the textual query. Thus, MLLMs can easily refuse these sensitive image-text pairs according to textual queries only, leading to unreliable cross-modality safety evaluation of MLLMs. We also conduct a further comparison experiment between textual alignment and multimodal alignment to highlight this drawback. To this end, we construct multimodal Visual Leakless Safety Bench (VLSBench) with 2.2k image-text pairs through an automated data pipeline. Experimental results indicate that VLSBench poses a significant challenge to both open-source and close-source MLLMs, e.g., LLaVA, Qwen2-VL and GPT-4o. Besides, we empirically compare textual and multimodal alignment methods on VLSBench and find that textual alignment is effective enough for multimodal safety scenarios with VSIL, while multimodal alignment is preferable for safety scenarios without VSIL. Code and data are released under https://github.com/AI45Lab/VLSBench

Related papers

• Text2VLM: Adapting Text-Only Datasets to Evaluate Alignment Training in Visual Language Models [0.0]
  Existing evaluation datasets lean towards text-only prompts, leaving visual vulnerabilities under evaluated.<n>We propose Text2VLM, a novel multi-stage pipeline that adapts text-only datasets into multimodal formats.<n>Text2VLM provides a scalable tool for comprehensive safety assessment, contributing to the development of more robust safety mechanisms for Visual Language Models.
  arXiv  Detail & Related papers  (2025-07-28T10:57:44Z)
• Con Instruction: Universal Jailbreaking of Multimodal Large Language Models via Non-Textual Modalities [76.9327488986162]
  Existing attacks against multimodal language models (MLLMs) primarily communicate instructions through text accompanied by adversarial images.<n>We exploit the capabilities of MLLMs to interpret non-textual instructions, specifically, adversarial images or audio generated by our novel method, Con Instruction.<n>Our method achieves the highest attack success rates, reaching 81.3% and 86.6% on LLaVA-v1.5 (13B)
  arXiv  Detail & Related papers  (2025-05-31T13:11:14Z)
• Unlearning Sensitive Information in Multimodal LLMs: Benchmark and Attack-Defense Evaluation [88.78166077081912]
  We introduce a multimodal unlearning benchmark, UnLOK-VQA, and an attack-and-defense framework to evaluate methods for deleting specific multimodal knowledge from MLLMs.<n>Our results show multimodal attacks outperform text- or image-only ones, and that the most effective defense removes answer information from internal model states.
  arXiv  Detail & Related papers  (2025-05-01T01:54:00Z)
• SEA: Low-Resource Safety Alignment for Multimodal Large Language Models via Synthetic Embeddings [32.661752596399204]
  Multimodal Large Language Models (MLLMs) have serious security vulnerabilities. Existing low-resource security alignment methods, including textual alignment, have been found to struggle with the security risks posed by additional modalities. We propose Synthetic Embedding augmented safety alignment (SEA), which optimize embeddings of additional modality through gradient updates.
  arXiv  Detail & Related papers  (2025-02-18T05:57:35Z)
• Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs [56.440345471966666]
  Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images. This paper introduces MMSafeAware, the first comprehensive multimodal safety awareness benchmark designed to evaluate MLLMs across 29 safety scenarios. MMSafeAware includes both unsafe and over-safety subsets to assess models abilities to correctly identify unsafe content and avoid over-sensitivity that can hinder helpfulness.
  arXiv  Detail & Related papers  (2025-02-16T16:12:40Z)
• Universal Adversarial Attack on Aligned Multimodal LLMs [1.429392440481971]
  We propose a universal adversarial attack on multimodal Large Language Models (LLMs)<n>We craft a synthetic image that forces the model to respond with a targeted phrase or otherwise unsafe content.<n>We will release code and datasets under the Apache-2.0 license.
  arXiv  Detail & Related papers  (2025-02-11T22:07:47Z)
• MSTS: A Multimodal Safety Test Suite for Vision-Language Models [35.86658426268927]
  Vision-language models (VLMs) are increasingly integrated into chat assistants and other consumer AI applications. Despite clear hazards, little work has evaluated VLM safety and the novel risks created by multimodal inputs. M MSTS comprises 400 test prompts across 40 fine-grained hazard categories. We translate MSTS into ten languages, showing non-English prompts to increase the rate of unsafe model responses.
  arXiv  Detail & Related papers  (2025-01-17T09:22:35Z)
• Multimodal Situational Safety [73.63981779844916]
  We present the first evaluation and analysis of a novel safety challenge termed Multimodal Situational Safety. For an MLLM to respond safely, whether through language or action, it often needs to assess the safety implications of a language query within its corresponding visual context. We develop the Multimodal Situational Safety benchmark (MSSBench) to assess the situational safety performance of current MLLMs.
  arXiv  Detail & Related papers  (2024-10-08T16:16:07Z)
• TrojVLM: Backdoor Attack Against Vision Language Models [50.87239635292717]
  This study introduces TrojVLM, the first exploration of backdoor attacks aimed at Vision Language Models (VLMs) TrojVLM inserts predetermined target text into output text when encountering poisoned images. A novel semantic preserving loss is proposed to ensure the semantic integrity of the original image content.
  arXiv  Detail & Related papers  (2024-09-28T04:37:09Z)
• CoCA: Regaining Safety-awareness of Multimodal Large Language Models with Constitutional Calibration [90.36429361299807]
  multimodal large language models (MLLMs) have demonstrated remarkable success in engaging in conversations involving visual inputs. The integration of visual modality has introduced a unique vulnerability: the MLLM becomes susceptible to malicious visual inputs. We introduce a technique termed CoCA, which amplifies the safety-awareness of the MLLM by calibrating its output distribution.
  arXiv  Detail & Related papers  (2024-09-17T17:14:41Z)
• AIM: Let Any Multi-modal Large Language Models Embrace Efficient In-Context Learning [15.770849688170477]
  In-context learning (ICL) facilitates Large Language Models exhibiting emergent ability on downstream tasks without updating billions of parameters. Most primary MLLMs are only trained on single-image datasets, making them unable to read multi-modal demonstrations. We propose a general and light-weighted framework textbfAIM to tackle the mentioned problems through textbfAggregating textbfImage information of textbfMultimodal demonstrations to the dense latent space of the corresponding linguistic part.
  arXiv  Detail & Related papers  (2024-06-11T08:12:43Z)
• Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models [107.88745040504887]
  We study the harmlessness alignment problem of multimodal large language models (MLLMs) Inspired by this, we propose a novel jailbreak method named HADES, which hides and amplifies the harmfulness of the malicious intent within the text input. Experimental results show that HADES can effectively jailbreak existing MLLMs, which achieves an average Attack Success Rate (ASR) of 90.26% for LLaVA-1.5 and 71.60% for Gemini Pro Vision.
  arXiv  Detail & Related papers  (2024-03-14T18:24:55Z)
• Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation [98.02846901473697]
  We propose ECSO (Eyes Closed, Safety On), a training-free protecting approach that exploits the inherent safety awareness of MLLMs. ECSO generates safer responses via adaptively transforming unsafe images into texts to activate the intrinsic safety mechanism of pre-aligned LLMs.
  arXiv  Detail & Related papers  (2024-03-14T17:03:04Z)
• MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models [41.708401515627784]
  We observe that Multimodal Large Language Models (MLLMs) can be easily compromised by query-relevant images. We introduce MM-SafetyBench, a framework designed for conducting safety-critical evaluations of MLLMs against such image-based manipulations. Our work underscores the need for a concerted effort to strengthen and enhance the safety measures of open-source MLLMs against potential malicious exploits.
  arXiv  Detail & Related papers  (2023-11-29T12:49:45Z)
• LION : Empowering Multimodal Large Language Model with Dual-Level Visual Knowledge [58.82222646803248]
  Multimodal Large Language Models (MLLMs) have endowed LLMs with the ability to perceive and understand multi-modal signals. Most of the existing MLLMs mainly adopt vision encoders pretrained on coarsely aligned image-text pairs, leading to insufficient extraction and reasoning of visual knowledge. We propose a dual-Level vIsual knedgeOwl eNhanced Multimodal Large Language Model (LION), which empowers the MLLM by injecting visual knowledge in two levels.
  arXiv  Detail & Related papers  (2023-11-20T15:56:44Z)

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.