Jailbreak Attack Success Evaluation on HADES SD+TYPO
2.1Attack Success Rate (ASR)JRS-Rem
Evaluation Results
| Method | Links | |
|---|---|---|
| JRS-RemVLM=ShareGPT4V-7B2026.03 | 2.1 | |
| CMRMVLM=ShareGPT4V-7B2026.03 | 5.4 | |
| JRS-RemVLM=InternVL-Chat-19B2026.03 | 6.1 | |
| JRS-RemVLM=LLaVA-1.5-7B2026.03 | 12.4 | |
| AdaShieldVLM=InternVL-Chat-19B2026.03 | 14.5 | |
| AdaShieldVLM=ShareGPT4V-7B2026.03 | 14.6 | |
| ECSOVLM=InternVL-Chat-19B2026.03 | 18.4 | |
| ShiftDCVLM=InternVL-Chat-19B2026.03 | 21.4 | |
| ShiftDCVLM=ShareGPT4V-7B2026.03 | 26.5 | |
| AdaShieldVLM=LLaVA-1.5-7B2026.03 | 29 | |
| ECSOVLM=LLaVA-1.5-7B2026.03 | 31.6 | |
| InternVL-Chat-19B (Baseline)VLM=InternVL-Chat-19B2026.03 | 31.7 | |
| CMRMVLM=InternVL-Chat-19B2026.03 | 36.8 | |
| ECSOVLM=ShareGPT4V-7B2026.03 | 42.2 | |
| ShareGPT4V-7B (Baseline)VLM=ShareGPT4V-7B2026.03 | 71.7 | |
| ShiftDCVLM=LLaVA-1.5-7B2026.03 | 72.9 | |
| CMRMVLM=LLaVA-1.5-7B2026.03 | 74.5 | |
| LLaVA-1.5-7B (Baseline)VLM=LLaVA-1.5-7B2026.03 | 84.9 |