Research Automation on DRB
38.5RACEw/o Attack
Evaluation Results
| Method | Links | |
|---|---|---|
| w/o AttackArchitecture=Hier.2026.05 | 38.5 | |
| Forced OutputArchitecture=Hier.2026.05 | 36.9 | |
| Prompt InfectionArchitecture=Hier.2026.05 | 36.2 | |
| w/o AttackArchitecture=Flat2026.05 | 34.8 | |
| Web FraudArchitecture=Hier.2026.05 | 34.1 | |
| InjecAgentArchitecture=Hier.2026.05 | 32.4 | |
| w/o AttackArchitecture=Chain2026.05 | 32.2 | |
| Forced OutputArchitecture=Flat2026.05 | 31.3 | |
| Forced OutputArchitecture=Chain2026.05 | 30.7 | |
| Prompt InfectionArchitecture=Chain2026.05 | 30.1 | |
| Prompt InfectionArchitecture=Flat2026.05 | 29.3 | |
| Web FraudArchitecture=Flat2026.05 | 27.9 | |
| Web FraudArchitecture=Chain2026.05 | 27.1 | |
| InjecAgentArchitecture=Flat2026.05 | 26.5 | |
| Evo-AttackerArchitecture=Hier.2026.05 | 25.9 | |
| InjecAgentArchitecture=Chain2026.05 | 25.4 | |
| Evo-AttackerArchitecture=Flat2026.05 | 22.1 | |
| Evo-AttackerArchitecture=Chain2026.05 | 21.1 |