Agent Safety Evaluation on ToolEmu
99SafetySafeMCP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SafeMCPAgent=GPT-4o2026.06 | 99 | 22 | 60 | 44 | |
| SafeMCPAgent=GPT-4o-mini2026.06 | 98 | 15 | 56 | 40 | |
| RL-GuardAgent=Claude-3.5-Sonnet2026.06 | 94 | 39 | 67 | 57 | |
| SafeMCPAgent=Claude-3.5-Sonnet2026.06 | 94 | 42 | 68 | 59 | |
| RL-GuardAgent=GPT-4o2026.06 | 89 | 9 | 49 | 35 | |
| AgentMonitorAgent=Claude-3.5-Sonnet2026.06 | 89 | 28 | 59 | 48 | |
| RL-GuardAgent=GPT-4o-mini2026.06 | 88 | 9 | 49 | 35 | |
| AgentMonitorAgent=Llama-3.1-8B2026.06 | 88 | 3 | 45 | 31 | |
| SafeMCPAgent=Llama-3.1-8B2026.06 | 85 | 7 | 46 | 33 | |
| AgentMonitorAgent=GPT-4o-mini2026.06 | 83 | 10 | 47 | 35 | |
| AgentMonitorAgent=GPT-4o2026.06 | 81 | 22 | 52 | 43 | |
| ChainGuardAgent=Claude-3.5-Sonnet2026.06 | 77 | 40 | 59 | 55 | |
| RL-GuardAgent=Llama-3.1-8B2026.06 | 77 | 3 | 40 | 30 | |
| NeMoGuard-8BAgent=Claude-3.5-Sonnet2026.06 | 76 | 43 | 60 | 56 | |
| Qwen3Guard-8BAgent=Claude-3.5-Sonnet2026.06 | 72 | 43 | 57 | 55 | |
| ChainGuardAgent=GPT-4o2026.06 | 71 | 15 | 43 | 36 | |
| Llama Guard 3Agent=Claude-3.5-Sonnet2026.06 | 69 | 47 | 58 | 57 | |
| ChainGuardAgent=GPT-4o-mini2026.06 | 65 | 11 | 38 | 32 | |
| SafironAgent=Claude-3.5-Sonnet2026.06 | 65 | 53 | 59 | 59 | |
| Qwen3Guard-8BAgent=GPT-4o2026.06 | 63 | 21 | 42 | 38 | |
| w/o defenseAgent=Claude-3.5-Sonnet2026.06 | 63 | 54 | 58 | 58 | |
| ChainGuardAgent=Llama-3.1-8B2026.06 | 60 | 5 | 33 | 27 | |
| NeMoGuard-8BAgent=GPT-4o2026.06 | 58 | 22 | 40 | 37 | |
| Qwen3Guard-8BAgent=GPT-4o-mini2026.06 | 58 | 15 | 36 | 33 | |
| NeMoGuard-8BAgent=Llama-3.1-8B2026.06 | 52 | 7 | 30 | 26 | |
| Llama Guard 3Agent=Llama-3.1-8B2026.06 | 51 | 7 | 29 | 26 | |
| Qwen3Guard-8BAgent=Llama-3.1-8B2026.06 | 49 | 6 | 28 | 25 | |
| SafironAgent=GPT-4o2026.06 | 45 | 24 | 34 | 33 | |
| SafironAgent=GPT-4o-mini2026.06 | 45 | 17 | 31 | 29 | |
| Llama Guard 3Agent=GPT-4o-mini2026.06 | 44 | 17 | 31 | 29 | |
| NeMoGuard-8BAgent=GPT-4o-mini2026.06 | 43 | 14 | 29 | 27 | |
| w/o defenseAgent=GPT-4o2026.06 | 42 | 25 | 34 | 33 | |
| w/o defenseAgent=GPT-4o-mini2026.06 | 42 | 17 | 30 | 28 | |
| Llama Guard 3Agent=GPT-4o2026.06 | 41 | 20 | 31 | 30 | |
| SafironAgent=Llama-3.1-8B2026.06 | 36 | 7 | 22 | 20 | |
| w/o defenseAgent=Llama-3.1-8B2026.06 | 33 | 7 | 20 | 19 |