Agent Safety Evaluation on AgentDojo
0ASRECA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ECATasks=2,9202026.05 | 0 | 100 | 0 | — | |
| prompt–response verification framework2026.06 | 5.1 | — | — | 93.6 | |
| Prompt-only safetyTasks=2,9202026.05 | 52.06 | 96.18 | 52.06 | — | |
| PromptGuard-2 / InjecGuardType=Strongest Applicable Baseline2026.06 | — | — | — | 85.9 |