Red-teaming Sample Quality Evaluation on Red-teaming Attack Samples
88Textual DiversityTrust-VLM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Trust-VLM2026.03 | 88 | 50 | 11.67 | 51 | — | |
| TreeTeaming2026.03 | 83 | 64.1 | 6.63 | 9.86 | 2.718 | |
| Arondight2026.03 | 58 | 56 | — | — | — | |
| Hades2026.03 | 47 | 38.6 | 91.46 | 97.63 | — | |
| MM-safety2026.03 | 34 | 46.4 | 36.25 | 78.81 | — | |
| SI-Attack2026.03 | 17 | 44.7 | 28.69 | 11.44 | — | |
| Fig-stepFixed textual prompts=true2026.03 | 0 | 7.5 | 0 | 60.52 | — | |
| MMLFixed textual prompts=true2026.03 | 0 | 7.8 | 0 | 15.81 | — | |
| CS-DJFixed textual prompts=true2026.03 | 0 | 29.8 | 0 | 22.71 | — | |
| JOODFixed textual prompts=true2026.03 | 0 | 40.5 | 0 | 32.15 | — |