Agent Security (Indirect Prompt Injection) on AgentDojo Travel (test)
1.3ASRBenchmark
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BenchmarkModel=GLM-5.12026.05 | 1.3 | 56 | 1.3 | |
| AgentVigilModel=GLM-5.12026.05 | 1.3 | 66.7 | 0 | |
| IterInjectModel=GLM-5.12026.05 | 1.3 | 60 | 1.3 | |
| BenchmarkModel=MiniMax-M2.72026.05 | 9.3 | 26.7 | 1.3 | |
| AgentVigilModel=MiniMax-M2.72026.05 | 9.3 | 78.7 | 1.3 | |
| IterInjectModel=MiniMax-M2.72026.05 | 13.3 | 34.7 | 4 | |
| BenchmarkModel=DS-V4-Flash2026.05 | 14.7 | 56 | 0 | |
| BenchmarkModel=Qwen3.5-27B2026.05 | 20 | 60 | 2.7 | |
| AgentVigilModel=Qwen3.5-27B2026.05 | 20 | 70.7 | 2.7 | |
| AgentVigilModel=DS-V4-Flash2026.05 | 21.3 | 60 | 0 | |
| IterInjectModel=Qwen3.5-27B2026.05 | 21.3 | 68 | 2.7 | |
| IterInjectModel=DS-V4-Flash2026.05 | 33.3 | 60 | 1.3 |