Research map / Adversarial robustness and security / Attacks on text and graphs
Omni-Attack: Adversarial Attacks on Open-Ended VQA in Black-Box Multimodal LLMs
Kai Hu, Weichen Yu, Li Zhang, Alexander Robey, Andy Zou, Haoqi Hu, Chengming Xu, Matt Fredrikson
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add cycle verification to five textual targets (ablation) | Omni-Attack · MMBench, GPT-4.1 | ASR: 67.6 -> 69.8 | Improved |
| add cycle verification to single visual target (ablation) | Omni-Attack · MMBench, GPT-4.1 | ASR: 65.3 -> 66.8 | Improved |
| increase number of textual targets from 1 to 5 (cycle) (ablation) | Omni-Attack · MMBench, GPT-4.1 | ASR: 67.1 -> 69.8 | Improved |
| fuse textual and visual targets (ablation) | Omni-Attack · MMBench, GPT-4.1 | ASR: 69.8 -> 71.8 | Improved |
| replace one CLIP surrogate with DINOv2 (ablation) | Omni-Attack surrogates · MMBench, GPT-4.1 | ASR: 71.8 -> 65.6 | Hurt |
| replace one CLIP surrogate with adversarially trained AdvXL (ablation) | Omni-Attack surrogates · MMBench, GPT-4.1 | ASR: 71.8 -> 60.0 | Hurt |
| use smaller 224-resolution CLIP surrogates (ablation) | Omni-Attack surrogates · MMBench, GPT-4.1 | ASR: 71.8 -> 56.8 | Hurt |
| use 6 CLIP surrogates instead of 3 (ablation) | Omni-Attack surrogates · MMBench, GPT-4.1 | ASR: 71.8 -> 71.9 | No clear effect |
| add DropPath (ablation) | Omni-Attack · MMBench, Claude 3.7 | ASR: 11.1 -> 15.5 | Improved |
| add PatchDrop (ablation) | Omni-Attack · MMBench, GPT-4.1 | ASR: 69.3 -> 71.8 | Improved |
| add perturbation moving average (ablation) | Omni-Attack · MMBench, Claude 3.7 | ASR: 11.1 -> 15.5 | Improved |
| add random JPEGify (ablation) | Omni-Attack · MMBench, GPT-4.1 | ASR: 71.6 -> 71.8 | No clear effect |
Closest papers by meaning
- VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models (AAAI 2024 · similarity 0.90)
- On the Robustness of Large Multimodal Models Against Image Adversarial Attacks (CVPR 2024 · similarity 0.89)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.88)
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Backdooring Vision-Language Models with Out-Of-Distribution Data (ICLR 2025 · similarity 0.86)
- A Frustratingly Simple Yet Highly Effective Attack Baseline: Over 90% Success Rate Against the Strong Black-box Models of GPT-4.5/4o/o1 (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
