Research Automation on MAB.research
85.4TSw/o Attack
Evaluation Results
| Method | Links | |
|---|---|---|
| w/o AttackArchitecture=Hier.2026.05 | 85.4 | |
| w/o AttackArchitecture=Chain2026.05 | 81.2 | |
| w/o AttackArchitecture=Flat2026.05 | 80 | |
| Prompt InfectionArchitecture=Hier.2026.05 | 79.6 | |
| Web FraudArchitecture=Hier.2026.05 | 78.2 | |
| InjecAgentArchitecture=Hier.2026.05 | 77.5 | |
| Forced OutputArchitecture=Hier.2026.05 | 76.8 | |
| InjecAgentArchitecture=Flat2026.05 | 69.8 | |
| Web FraudArchitecture=Chain2026.05 | 67.4 | |
| InjecAgentArchitecture=Chain2026.05 | 66.9 | |
| Web FraudArchitecture=Flat2026.05 | 66.2 | |
| Forced OutputArchitecture=Chain2026.05 | 65.9 | |
| Forced OutputArchitecture=Flat2026.05 | 64.8 | |
| Evo-AttackerArchitecture=Hier.2026.05 | 63.8 | |
| Prompt InfectionArchitecture=Chain2026.05 | 63.1 | |
| Prompt InfectionArchitecture=Flat2026.05 | 61.4 | |
| Evo-AttackerArchitecture=Flat2026.05 | 54.6 | |
| Evo-AttackerArchitecture=Chain2026.05 | 53.1 |