Research map / Adversarial robustness and security / Attacks on text and graphs
IDEATOR: Jailbreaking and Benchmarking Large Vision-Language Models Using Themselves
Ruofan Wang, Juncheng Li, Yixu Wang, Bo Wang, Xiaosen Wang, Yan Teng, Yingchun Wang, Xingjun Ma, Yu-Gang Jiang
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 IDEATOR (black-box) instead of no attack | MiniGPT-4 · AdvBench | ASR (%): 35.0 -> 94.0 | Improved |
| use IDEATOR instead of MM-SafetyBench | MiniGPT-4 · AdvBench | ASR (%): 66.0 -> 94.0 | Improved |
| use IDEATOR instead of MM-SafetyBench for transfer to LLaVA | IDEATOR · AdvBench transfer | ASR (%): 46.0 -> 82.0 | Improved |
| use IDEATOR instead of MM-SafetyBench for transfer to Chameleon | IDEATOR · AdvBench transfer | ASR (%): 22.0 -> 75.0 | Improved |
| increase exploration breadth and depth from Nb=1,Nd=1 to Nb=7,Nd=3 (ablation) | IDEATOR · AdvBench | ASR (%): 45.0 -> 94.0 | Improved |
| increase breadth from 1 to 7 at depth 1 (ablation) | IDEATOR · AdvBench | ASR (%): 45.0 -> 85.0 | Improved |
| increase depth from 1 to 3 at breadth 1 (ablation) | IDEATOR · AdvBench | ASR (%): 45.0 -> 68.0 | Improved |
| add text component to image-only attack (Adv I+T vs Adv Img) (ablation) | IDEATOR · AdvBench | ASR (%): 85.0 -> 94.0 | Improved |
| add image component to text-only attack (Adv I+T vs Adv Text) (ablation) | IDEATOR · AdvBench | ASR (%): 86.0 -> 94.0 | Improved |
| apply AdaShield-S defense against IDEATOR on MiniGPT-4 | IDEATOR · AdvBench (defense) | ASR (%): 94.0 -> 84.0 | Hurt |
| apply AdaShield-S defense against MM-SafetyBench on LLaVA | MM-SafetyBench · AdvBench (defense) | ASR (%): 46.0 -> 5.0 | Hurt |
| use Gemini-based IDEATOR challenge set vs base set on Claude-3.5-Sonnet | VLJailbreakBench · VLJailbreakBench | ASR (%): 1.09 -> 19.65 | Improved |
Later papers that compared with this method
1 later accepted paper reports IDEATOR as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Adversarial Style Optimization: Enhancing VLM Jailbreaks by GRPO-based Stylistic Triggers Optimization (CVPR 2026 · 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
- MMJ-Bench: A Comprehensive Study on Jailbreak Attacks and Defenses for Vision Language Models (AAAI 2025 · similarity 0.91)
- BlueSuffix: Reinforced Blue Teaming for Vision-Language Models Against Jailbreak Attacks (ICLR 2025 · similarity 0.90)
- JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (NeurIPS 2024 · similarity 0.90)
- Failures to Find Transferable Image Jailbreaks Between Vision-Language Models (ICLR 2025 · similarity 0.89)
- SafePTR: Token-Level Jailbreak Defense in Multimodal LLMs via Prune-then-Restore Mechanism (NeurIPS 2025 · similarity 0.89)
- JailBound: Jailbreaking Internal Safety Boundaries of Vision-Language Models (NeurIPS 2025 · similarity 0.89)
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.
