Agent Security (Indirect Prompt Injection) on AgentDojo Overall (test)
11.6ASRBenchmark
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BenchmarkModel=GLM-5.12026.05 | 11.6 | 64.7 | 7.3 | |
| BenchmarkModel=MiniMax-M2.72026.05 | 16.1 | 61 | 8.2 | |
| IterInjectModel=GLM-5.12026.05 | 17.5 | 70.4 | 9.8 | |
| AgentVigilModel=GLM-5.12026.05 | 18.2 | 77.5 | 9.6 | |
| AgentVigilModel=MiniMax-M2.72026.05 | 23.1 | 70.8 | 8.8 | |
| BenchmarkModel=Qwen3.5-27B2026.05 | 26.3 | 70.6 | 10.2 | |
| IterInjectModel=MiniMax-M2.72026.05 | 26.5 | 65.1 | 12.4 | |
| AgentVigilModel=Qwen3.5-27B2026.05 | 29 | 76.3 | 11 | |
| IterInjectModel=Qwen3.5-27B2026.05 | 32.4 | 74.1 | 12.2 | |
| BenchmarkModel=DS-V4-Flash2026.05 | 32.9 | 71.4 | 21 | |
| AgentVigilModel=DS-V4-Flash2026.05 | 39.2 | 62.4 | 16.5 | |
| IterInjectModel=DS-V4-Flash2026.05 | 47.8 | 75.3 | 28.8 |