Implicit Tool Poisoning on MCPTox
4.3ASRMCPTox
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MCPToxLLM Agent=Qwen3-8b, Reasoning Mode=false2026.01 | 4.3 | 80.4 | 15.3 | 0 | |
| MCP-ITPLLM Agent=Qwen3-8b, Reasoning Mode=false2026.01 | 7.4 | 66.7 | 24.9 | 0.96 | |
| MCPToxLLM Agent=Qwen3-235b-a22b, Reasoning Mode=false2026.01 | 8 | 47.5 | 43.6 | 0.9 | |
| MCPToxLLM Agent=Qwen3-32b, Reasoning Mode=false2026.01 | 9.5 | 71.5 | 19 | 0 | |
| MCP-ITPLLM Agent=Qwen3-235b-a22b, Reasoning Mode=false2026.01 | 16.5 | 29.8 | 53.7 | 0 | |
| MCPToxLLM Agent=Qwen3-8b, Reasoning Mode=true2026.01 | 20.3 | 70.9 | 8.3 | 0.5 | |
| MCP-ITPLLM Agent=Qwen3-32b, Reasoning Mode=false2026.01 | 28.8 | 48.9 | 21.7 | 0.6 | |
| MCPToxLLM Agent=Qwen3-235b-a22b, Reasoning Mode=true2026.01 | 32.5 | 59.5 | 7.7 | 0.3 | |
| MCP-ITPLLM Agent=Qwen3-8b, Reasoning Mode=true2026.01 | 40.5 | 52.7 | 5.2 | 1.6 | |
| MCPToxLLM Agent=DeepSeek-V3, Reasoning Mode=false2026.01 | 44.8 | 38 | 16.6 | 0.6 | |
| MCPToxLLM Agent=Qwen3-32b, Reasoning Mode=true2026.01 | 45.1 | 46.3 | 7.7 | 0.9 | |
| MCP-ITPLLM Agent=Qwen3-235b-a22b, Reasoning Mode=true2026.01 | 45.3 | 51.5 | 2.3 | 0.9 | |
| MCPToxLLM Agent=Gemini-2.5-flash, Reasoning Mode=false2026.01 | 46 | 40.8 | 11 | 2.2 | |
| MCP-ITPLLM Agent=Gemini-2.5-flash, Reasoning Mode=false2026.01 | 47.6 | 5.5 | 18.8 | 28.1 | |
| MCPToxLLM Agent=GPT-3.5-turbo, Reasoning Mode=false2026.01 | 48.2 | 23.3 | 24.9 | 3.6 | |
| MCPToxLLM Agent=GPT-4o-mini, Reasoning Mode=false2026.01 | 54 | 26.1 | 18.7 | 1.2 | |
| MCPToxLLM Agent=o1-mini, Reasoning Mode=false2026.01 | 59.5 | 19.9 | 15.9 | 4.7 | |
| MCPToxLLM Agent=DeepSeek-R1, Reasoning Mode=false2026.01 | 66 | 12.3 | 12 | 9.7 | |
| MCP-ITPLLM Agent=DeepSeek-R1, Reasoning Mode=false2026.01 | 66.7 | 5.8 | 8.7 | 18.8 | |
| MCP-ITPLLM Agent=Qwen3-32b, Reasoning Mode=true2026.01 | 69.9 | 26.2 | 3.6 | 0.3 | |
| MCP-ITPLLM Agent=GPT-4o-mini, Reasoning Mode=false2026.01 | 72.5 | 7.4 | 16.2 | 3.9 | |
| MCP-ITPLLM Agent=DeepSeek-V3, Reasoning Mode=false2026.01 | 78.4 | 14.8 | 4.8 | 2 | |
| MCP-ITPLLM Agent=o1-mini, Reasoning Mode=false2026.01 | 82.9 | 4.2 | 6.8 | 6.1 | |
| MCP-ITPLLM Agent=GPT-3.5-turbo, Reasoning Mode=false2026.01 | 84.2 | 9.3 | 3.4 | 3.1 |