AI Paper Atlas

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

AAAI 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
use TVChain sub-image chain plus CoT prompt instead of HADES single-image attackQwen2.5-VL · HADESASR: 30.27 -> 95.87Improved
use TVChain instead of VisCRAOpenAI o4-mini · HADESASR: 11.73 -> 23.86Improved
use TVChain instead of VisCRAGPT-4o · MM-SafetyASR: 45.88 -> 65.99Improved
use TVChain instead of QRQwen2.5-VL · MM-SafetyASR: 49.73 -> 87.85Improved
use TVChain instead of VisCRAR1-Onevision · HADES (Self Harm)ASR: 79.33 -> 79.33No clear effect
add visual prompt chain (full vs w/o visual) (ablation)TVChain · HADESASR: 22.53 -> 82.00Improved
add textual CoT prompt chain (full vs w/o textual) (ablation)TVChain · HADESASR: 46.27 -> 82.00Improved
use TVChain prompts instead of HADES promptsLlama Guard 3 · HADESdetection rate: 82.67 -> 34.93Improved
use TVChain prompts instead of HADES promptsJailGuard · HADESdetection rate: 82.13 -> 73.60Improved
add system prompt hardening or visual noise (ablation)TVChain · HADES/MM-SafetyASRMixed
target reasoning-safe modelTVChain · HADESASR: 95.87 (Qwen2.5-VL) vs 23.86 (o4-mini)Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy