Injection Task Success on AgentDojo
63.7ITSR (Workspace)VeriGrey
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| VeriGreyLLM=Qwen-3 235B2026.03 | 63.7 | 100 | 79.9 | 100 | 93.6 | 41.2 | 54.6 | 81.7 | |
| VeriGreyLLM=GPT-4.12026.03 | 48.2 | 87.1 | 68.1 | 100 | 88.1 | 30.1 | 43.3 | 70.7 | |
| BaselineLLM=Qwen-3 235B2026.03 | 40.2 | 90.7 | 64.6 | 100 | 88.7 | 28.1 | 32 | 67.6 | |
| BaselineLLM=GPT-4.12026.03 | 16.1 | 40.7 | 34.7 | 78.1 | 55.2 | 7.6 | 11.3 | 37.7 | |
| VeriGreyLLM=Gemini-2.5 Flash2026.03 | 15.9 | 77.1 | 32.6 | 90.5 | 63.2 | 16.2 | 17.5 | 47.4 | |
| BaselineLLM=Gemini-2.5 Flash2026.03 | 9.8 | 65 | 14.6 | 78.1 | 49.5 | 11.7 | 11.9 | 36.8 |