Safety Evaluation on CocoNot
0.613Safety ScoreGRAPH ROUTER
Evaluation Results
| Method | Links | |
|---|---|---|
| GRAPH ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Benign)2026.02 | 0.613 | |
| LLM ROUTERMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.575 | |
| LLM ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.569 | |
| LLM ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Benign)2026.02 | 0.568 | |
| LLM ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.549 | |
| LLM ROUTERMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.541 | |
| GRAPH ROUTERMalicious Setting=M4-RL, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.495 | |
| GRAPH ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.492 | |
| GRAPH ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.49 | |
| GRAPH ROUTERMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.485 | |
| GRAPH ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.484 | |
| GRAPH ROUTERMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.46 | |
| LLM ROUTERMalicious Setting=M4-RL, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.429 | |
| LLM ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.424 | |
| LLM ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Malicious)2026.02 | 0.41 | |
| TEXT DEBATEMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.373 | |
| LLM ROUTERMalicious Setting=M4-RL, Mitigation Strategy=None (Malicious)2026.02 | 0.368 | |
| TEXT DEBATEMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.365 | |
| TEXT DEBATEMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.365 | |
| TEXT DEBATEMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.333 | |
| TEXT DEBATEMalicious Setting=M4-RL, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.332 | |
| TEXT FEEDBACKMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.319 | |
| TEXT DEBATEMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.307 | |
| TEXT FEEDBACKMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.304 | |
| TEXT DEBATEMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Benign)2026.02 | 0.298 | |
| TEXT FEEDBACKMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Benign)2026.02 | 0.296 | |
| TEXT FEEDBACKMalicious Setting=M2-Activation Steering, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.291 | |
| TEXT FEEDBACKMalicious Setting=M4-RL, Mitigation Strategy=None (Malicious)2026.02 | 0.254 | |
| TEXT DEBATEMalicious Setting=M4-RL, Mitigation Strategy=None (Malicious)2026.02 | 0.242 | |
| TEXT FEEDBACKMalicious Setting=M4-RL, Mitigation Strategy=supervisor-free mitigation2026.02 | 0.242 | |
| TEXT FEEDBACKMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S1)2026.02 | 0.233 | |
| TEXT FEEDBACKMalicious Setting=M4-RL, Mitigation Strategy=supervisor-based mitigation (S2)2026.02 | 0.232 | |
| TEXT DEBATEMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Malicious)2026.02 | 0.2 | |
| TEXT FEEDBACKMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Malicious)2026.02 | 0.199 | |
| GRAPH ROUTERMalicious Setting=M4-RL, Mitigation Strategy=None (Malicious)2026.02 | 0.184 | |
| GRAPH ROUTERMalicious Setting=M2-Activation Steering, Mitigation Strategy=None (Malicious)2026.02 | 0.168 |