Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Style Optimization: Enhancing VLM Jailbreaks by GRPO-based Stylistic Triggers Optimization
Bingjun Luo, Jialin Guo, Yue Yao, Xinpeng Ding
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 GRPO-based Enhance stage after Probing (ablation) | SI Attack on LLaVA-OV-1.5 · MM-SafetyBench | ASR (%): 39.96% -> 44.25% | Improved |
| add Probing (naive best style) to base attack (ablation) | SI Attack on Qwen3-VL · MM-SafetyBench | ASR (%): 39.31% -> 40.62% | Improved |
| add Probing then Enhance (full ASO) to base attack | QR Attack on Qwen3-VL · MM-SafetyBench | ASR (%): 38.99% -> 42.98% | Improved |
| add full ASO | QR Attack on Gemini-2.5-Flash · MM-SafetyBench | ASR (%): 55.04% -> 62.79% | Improved |
| add full ASO | HIMRD on GPT-4.1-mini · MM-SafetyBench | ASR (%): 72.80% -> 71.77% | Hurt |
| add full ASO | HIMRD on Qwen3-VL · MM-SafetyBench | ASR (%): 87.38% -> 89.52% | Improved |
| add full ASO | IDEATOR on GPT-4.1-mini · VLBreakBench | ASR (%): 74.82% -> 75.54% | Improved |
| add full ASO | SI Attack on Gemini-2.5-Flash · VLBreakBench | ASR (%): 58.99% -> 65.47% | Improved |
| add full ASO | QR Attack on Fraud category · MM-SafetyBench categories | ASR (%): 5.2% -> 9.7% | Improved |
| add full ASO | SI Attack on Gov. decision category (GD) · MM-SafetyBench categories | ASR (%): 55.0% -> 56.4% | Improved |
| add full ASO | HIMRD on Economic Harm · MM-SafetyBench categories | ASR (%): 95.1% -> 97.5% | Improved |
| add full ASO | HIMRD on Economic Harm · MM-SafetyBench categories | Harmfulness Score: 9.6 -> 10.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- IDEATOR — IDEATOR: Jailbreaking and Benchmarking Large Vision-Language Models Using Themselves (ICCV 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Jailbreaking Vision-Language Models via Dissonance-Guided Suffix Optimization and Image-Phrase Injection (CVPR 2026 · similarity 0.89)
- Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models (ICLR 2024 · similarity 0.89)
- Failures to Find Transferable Image Jailbreaks Between Vision-Language Models (ICLR 2025 · similarity 0.89)
- Distraction is All You Need for Multimodal Large Language Model Jailbreaking (CVPR 2025 · similarity 0.89)
- JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (NeurIPS 2024 · similarity 0.88)
- Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks (CVPR 2025 · similarity 0.88)
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.
