Multi-agent System Safety and Welfare Evaluation on Governance Simulation Environment
30ToxicityBaseline
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaselineScenario=Baseline2026.03 | 30 | 181.38 | 172.6 | 100 | |
| Strict GovernanceScenario=Strict Governance2026.03 | 30 | 108.5 | 147.6 | 60 | |
| Adversarial Red TeamScenario=Adversarial Red Team2026.03 | 30.8 | 110.12 | 154.4 | 100 | |
| Misalignment SweepScenario=Misalignment Sweep2026.03 | 31.5 | 163.24 | 419.4 | 100 | |
| Adaptive GovernanceScenario=Adaptive Governance2026.03 | 34.1 | 184.14 | 355 | 0 | |
| Threshold DancerScenario=Threshold Dancer2026.03 | 35.3 | 354.8 | 1,009 | 0 | |
| Collusion DetectionScenario=Collusion Detection2026.03 | 35.7 | 157.9 | 270.6 | 100 |