Jailbreak Attack Success Evaluation on RedTeam2K SD+TYPO
51Attack Success Rate (ASR)CMRM
Evaluation Results
| Method | Links | |
|---|---|---|
| CMRMVLM=LLaVA-1.5-7B2026.03 | 51 | |
| LLaVA-1.5-7B (Baseline)VLM=LLaVA-1.5-7B2026.03 | 49.8 | |
| InternVL-Chat-19B (Baseline)VLM=InternVL-Chat-19B2026.03 | 43.6 | |
| ShiftDCVLM=LLaVA-1.5-7B2026.03 | 42.5 | |
| ShareGPT4V-7B (Baseline)VLM=ShareGPT4V-7B2026.03 | 39.2 | |
| ECSOVLM=LLaVA-1.5-7B2026.03 | 28.6 | |
| ShiftDCVLM=InternVL-Chat-19B2026.03 | 27.6 | |
| ECSOVLM=ShareGPT4V-7B2026.03 | 26.1 | |
| AdaShieldVLM=LLaVA-1.5-7B2026.03 | 22.2 | |
| JRS-RemVLM=LLaVA-1.5-7B2026.03 | 21.9 | |
| CMRMVLM=InternVL-Chat-19B2026.03 | 21.4 | |
| ShiftDCVLM=ShareGPT4V-7B2026.03 | 20.1 | |
| ECSOVLM=InternVL-Chat-19B2026.03 | 19.8 | |
| AdaShieldVLM=InternVL-Chat-19B2026.03 | 19.6 | |
| AdaShieldVLM=ShareGPT4V-7B2026.03 | 15 | |
| CMRMVLM=ShareGPT4V-7B2026.03 | 12.1 | |
| JRS-RemVLM=InternVL-Chat-19B2026.03 | 9.6 | |
| JRS-RemVLM=ShareGPT4V-7B2026.03 | 9.1 |