Attack Success Rate Reduction on AgentRedBench closed subset 215-scenario panel
67.5ASR ReductionAGENTREDGUARD-DeBERTa
Evaluation Results
| Method | Links | |
|---|---|---|
| AGENTREDGUARD-DeBERTaParams=142M2026.06 | 67.5 | |
| AGENTREDGUARD-MiniLMParams=23M2026.06 | 67.5 | |
| ProtectAI PIParams=184M2026.06 | 20 | |
| Llama Guard 3Params=8B2026.06 | 0.1 | |
| WildGuardParams=7B2026.06 | 0.1 | |
| PromptGuard 2Params=22M2026.06 | 0 |