Indirect Prompt Injection on InjectAgent
90.83Direct Harm ASRIterInject
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| IterInjectVictim Model=DS-V4-Flash2026.05 | 90.83 | 100 | 89.84 | 90.32 | |
| AgentVigilVictim Model=DS-V4-Flash2026.05 | 89.17 | 98.44 | 93.75 | 91.53 | |
| IterInjectVictim Model=Qwen3.5-27B2026.05 | 69.17 | 62.5 | 60.16 | 64.52 | |
| AgentVigilVictim Model=Qwen3.5-27B2026.05 | 48.33 | 59.38 | 58.59 | 53.63 | |
| IterInjectVictim Model=MiniMax-M2.72026.05 | 45 | 75 | 44.53 | 44.76 | |
| IterInjectVictim Model=GLM-5.12026.05 | 44.17 | 25 | 22.66 | 33.07 | |
| AgentVigilVictim Model=MiniMax-M2.72026.05 | 28.33 | 70.31 | 60.16 | 44.75 | |
| AgentVigilVictim Model=GLM-5.12026.05 | 14.17 | 31.25 | 21.88 | 18.15 | |
| BenchmarkVictim Model=MiniMax-M2.72026.05 | 1.67 | 4.69 | 2.34 | 2.02 | |
| BenchmarkVictim Model=DS-V4-Flash2026.05 | 1.67 | 18.75 | 4.69 | 3.23 | |
| BenchmarkVictim Model=GLM-5.12026.05 | 0 | 0 | 0 | 0 | |
| BenchmarkVictim Model=Qwen3.5-27B2026.05 | 0 | 14.84 | 7.03 | 3.63 |