Group Collusive Attack Detection on SVAMP
92Detection AccuracyGroupGuard
Evaluation Results
| Method | Links | |
|---|---|---|
| GroupGuardAttack Type=Targeted Suppression, Model=Qwen3-32B2026.03 | 92 | |
| GroupGuardAttack Type=Group Exclusion, Model=Qwen3-235B-A22B2026.03 | 88 | |
| G-safeguardAttack Type=Group Exclusion, Model=Qwen3-235B-A22B2026.03 | 87 | |
| GroupGuardAttack Type=Targeted Suppression, Model=Deepseek-V3.12026.03 | 85 | |
| G-safeguardAttack Type=Targeted Suppression, Model=Deepseek-V3.12026.03 | 83.7 | |
| G-safeguardAttack Type=Targeted Suppression, Model=Qwen3-32B2026.03 | 83.4 | |
| AutoDefenseAttack Type=Group Exclusion, Model=Deepseek-V3.12026.03 | 81.2 | |
| G-safeguardAttack Type=Group Exclusion, Model=Deepseek-V3.12026.03 | 81.2 | |
| GroupGuardAttack Type=False Consensus, Model=Qwen3-32B2026.03 | 81 | |
| GroupGuardAttack Type=False Consensus, Model=Deepseek-V3.12026.03 | 80 | |
| G-safeguardAttack Type=False Consensus, Model=Qwen3-32B2026.03 | 80 | |
| AutoDefenseAttack Type=Targeted Suppression, Model=Deepseek-V3.12026.03 | 79.6 | |
| GroupGuardAttack Type=Group Exclusion, Model=Deepseek-V3.12026.03 | 79 | |
| G-safeguardAttack Type=False Consensus, Model=Deepseek-V3.12026.03 | 78.6 | |
| AutoDefenseAttack Type=Group Exclusion, Model=Qwen3-235B-A22B2026.03 | 77.2 | |
| AutoDefenseAttack Type=False Consensus, Model=Qwen3-235B-A22B2026.03 | 72.2 | |
| G-safeguardAttack Type=False Consensus, Model=Qwen3-235B-A22B2026.03 | 71.8 | |
| GroupGuardAttack Type=False Consensus, Model=Qwen3-235B-A22B2026.03 | 71 | |
| GroupGuardAttack Type=Targeted Suppression, Model=Qwen3-235B-A22B2026.03 | 71 | |
| GroupGuardAttack Type=Group Exclusion, Model=Qwen3-32B2026.03 | 70 | |
| G-safeguardAttack Type=Targeted Suppression, Model=Qwen3-235B-A22B2026.03 | 68.8 | |
| G-safeguardAttack Type=Group Exclusion, Model=Qwen3-32B2026.03 | 68.4 | |
| AutoDefenseAttack Type=False Consensus, Model=Qwen3-32B2026.03 | 64.6 | |
| AutoDefenseAttack Type=Group Exclusion, Model=Qwen3-32B2026.03 | 64.6 | |
| AutoDefenseAttack Type=False Consensus, Model=Deepseek-V3.12026.03 | 61.8 | |
| AutoDefenseAttack Type=Targeted Suppression, Model=Qwen3-32B2026.03 | 60.4 | |
| AutoDefenseAttack Type=Targeted Suppression, Model=Qwen3-235B-A22B2026.03 | 60 |