Commonsense Reasoning on CSQA (TSR and ASR)
82.5Task Success Rate (TSR)STAR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| STARBackbone=Qwen-Plus, Attack Setting=Cooperative2026.05 | 82.5 | 6 | |
| STARBackbone=Qwen-Plus, Attack Setting=Independent2026.05 | 81 | 8.5 | |
| ARGUSBackbone=Qwen-Plus, Attack Setting=Cooperative2026.05 | 74.75 | 17.5 | |
| STARBackbone=GPT-3.5-Turbo, Attack Setting=Cooperative2026.05 | 74.25 | 12.25 | |
| STARBackbone=GPT-3.5-Turbo, Attack Setting=Independent2026.05 | 73 | 13.25 | |
| ARGUSBackbone=Qwen-Plus, Attack Setting=Independent2026.05 | 71.25 | 21.5 | |
| Blind-GuardBackbone=Qwen-Plus, Attack Setting=Independent2026.05 | 69 | 24.75 | |
| G-SafeguardBackbone=Qwen-Plus, Attack Setting=Cooperative2026.05 | 68.5 | 23.5 | |
| G-SafeguardBackbone=Qwen-Plus, Attack Setting=Independent2026.05 | 67.75 | 25.25 | |
| Blind-GuardBackbone=Qwen-Plus, Attack Setting=Cooperative2026.05 | 67.75 | 25.5 | |
| C&IBackbone=Qwen-Plus, Attack Setting=Independent2026.05 | 64.75 | 28.25 | |
| STARBackbone=GPT-5-Nano, Attack Setting=Independent2026.05 | 64 | 23.5 | |
| Blind-GuardBackbone=GPT-3.5-Turbo, Attack Setting=Cooperative2026.05 | 62.75 | 31.25 | |
| C&IBackbone=Qwen-Plus, Attack Setting=Cooperative2026.05 | 62.5 | 31.25 | |
| Blind-GuardBackbone=GPT-3.5-Turbo, Attack Setting=Independent2026.05 | 61 | 34 | |
| STARBackbone=GPT-5-Nano, Attack Setting=Cooperative2026.05 | 59.5 | 28 | |
| C&IBackbone=GPT-3.5-Turbo, Attack Setting=Cooperative2026.05 | 58.25 | 34.5 | |
| C&IBackbone=GPT-3.5-Turbo, Attack Setting=Independent2026.05 | 54 | 40.25 | |
| Blind-GuardBackbone=GPT-5-Nano, Attack Setting=Independent2026.05 | 46.75 | 46.25 | |
| G-SafeguardBackbone=GPT-5-Nano, Attack Setting=Cooperative2026.05 | 45.25 | 48 | |
| ARGUSBackbone=GPT-3.5-Turbo, Attack Setting=Cooperative2026.05 | 42 | 52.25 | |
| ARGUSBackbone=GPT-3.5-Turbo, Attack Setting=Independent2026.05 | 40.75 | 54.25 | |
| G-SafeguardBackbone=GPT-5-Nano, Attack Setting=Independent2026.05 | 34.25 | 60 | |
| G-SafeguardBackbone=GPT-3.5-Turbo, Attack Setting=Cooperative2026.05 | 33.75 | 62 | |
| ARGUSBackbone=GPT-5-Nano, Attack Setting=Independent2026.05 | 29.75 | 66.5 | |
| Blind-GuardBackbone=GPT-5-Nano, Attack Setting=Cooperative2026.05 | 28.75 | 67 | |
| G-SafeguardBackbone=GPT-3.5-Turbo, Attack Setting=Independent2026.05 | 26 | 69.5 | |
| ARGUSBackbone=GPT-5-Nano, Attack Setting=Cooperative2026.05 | 24.5 | 71.5 | |
| C&IBackbone=GPT-5-Nano, Attack Setting=Independent2026.05 | 23.25 | 74 | |
| C&IBackbone=GPT-5-Nano, Attack Setting=Cooperative2026.05 | 11 | 88.5 |