Agent attack defense on ClawTrojan (clean negative/borderline)
0FPRGLM-5.1
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GLM-5.1defense=none2026.05 | 0 | 0 | 100 | |
| DS-V4-Flashdefense=none2026.05 | 0 | 0 | 100 | |
| GPT-5.4defense=none2026.05 | 1.1 | 1.1 | 98.9 | |
| StruQ2026.05 | 1.1 | 1.1 | 98.9 | |
| ClawKeeper2026.05 | 2.2 | 2.2 | 97.8 | |
| MELON2026.05 | 2.2 | 1.1 | 97.8 | |
| PromptShieldParameters=8B2026.05 | 3.3 | 0 | 96.7 | |
| CaMeL2026.05 | 5.4 | 1.1 | 94.6 | |
| PromptShieldParameters=1B2026.05 | 8.7 | 1.1 | 91.3 | |
| DASGuard2026.05 | 13 | 6.5 | 87 |