Safety Evaluation on Agent-SafetyBench
48Availability RateSafeMCP
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| SafeMCPmode=Zero-shot2026.06 | 48 | 90 | 77 | 67 | 75 | 100 | 83 | 80 | 77.6 | |
| AgentMonitormode=Zero-shot2026.06 | 10 | 79 | 22 | 26 | 40 | 100 | 26 | 32 | 41.9 | |
| LlamaGuard-3-8Bmode=Zero-shot2026.06 | 4 | 80 | 21 | 31 | 42 | 100 | 27 | 37 | 42.8 | |
| w/o Defensemode=Zero-shot2026.06 | 2 | 78 | 8 | 14 | 17 | 100 | 19 | 11 | 31.2 |