Research map / Adversarial robustness and security / Attacks on text and graphs
TVChain: Leveraging Textual-Visual Prompt Chains for Jailbreaking Large Vision-Language Models
Hao Yu, Ke Liang, Junxian Duan, Jun Wang, Siwei Wang, Chuan Ma, Xinwang Liu
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 |
|---|---|---|---|
| use TVChain sub-image chain plus CoT prompt instead of HADES single-image attack | Qwen2.5-VL · HADES | ASR: 30.27 -> 95.87 | Improved |
| use TVChain instead of VisCRA | OpenAI o4-mini · HADES | ASR: 11.73 -> 23.86 | Improved |
| use TVChain instead of VisCRA | GPT-4o · MM-Safety | ASR: 45.88 -> 65.99 | Improved |
| use TVChain instead of QR | Qwen2.5-VL · MM-Safety | ASR: 49.73 -> 87.85 | Improved |
| use TVChain instead of VisCRA | R1-Onevision · HADES (Self Harm) | ASR: 79.33 -> 79.33 | No clear effect |
| add visual prompt chain (full vs w/o visual) (ablation) | TVChain · HADES | ASR: 22.53 -> 82.00 | Improved |
| add textual CoT prompt chain (full vs w/o textual) (ablation) | TVChain · HADES | ASR: 46.27 -> 82.00 | Improved |
| use TVChain prompts instead of HADES prompts | Llama Guard 3 · HADES | detection rate: 82.67 -> 34.93 | Improved |
| use TVChain prompts instead of HADES prompts | JailGuard · HADES | detection rate: 82.13 -> 73.60 | Improved |
| add system prompt hardening or visual noise (ablation) | TVChain · HADES/MM-Safety | ASR | Mixed |
| target reasoning-safe model | TVChain · HADES | ASR: 95.87 (Qwen2.5-VL) vs 23.86 (o4-mini) | Hurt |
Closest papers by meaning
- FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts (AAAI 2025 · similarity 0.92)
- Jailbreaking Vision-Language Models Through the Visual Modality (ICML 2026 · similarity 0.91)
- Safe + Safe = Unsafe? Exploring How Safe Images Can Be Exploited to Jailbreak Large Vision-Language Models (NeurIPS 2025 · similarity 0.91)
- VERA-V: Variational Inference Framework for Jailbreaking Vision-Language Models (ICML 2026 · similarity 0.91)
- MIDAS: Multi-Image Dispersion and Semantic Reconstruction for Jailbreaking MLLMs (ICLR 2026 · similarity 0.91)
- When the Prompt Becomes Visual: Vision-Centric Jailbreak Attacks for Large Image Editing Models (ICML 2026 · similarity 0.91)
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.
