Prompt Injection Prevention on AgentDojo (test)
36Banking Success RateNone
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| NoneBackbone agent=GPT-4o2026.02 | 36 | 42 | 47 | 45 | |
| PromptArmorBackbone agent=GPT-4o2026.02 | 31 | 37 | 41 | 40 | |
| DataFilterBackbone agent=GPT-4o2026.02 | 5 | 3 | 7 | 9 | |
| RedVisorMistralBackbone agent=GPT-4o, Detection Backbone=Mistral2026.02 | 4 | 6 | 5 | 8 | |
| RedVisorLlamaBackbone agent=GPT-4o, Detection Backbone=Llama2026.02 | 3 | 6 | 4 | 5 | |
| RedVisorQwenBackbone agent=GPT-4o, Detection Backbone=Qwen2026.02 | 3 | 5 | 5 | 7 | |
| PromptLocateBackbone agent=GPT-4o2026.02 | 2 | 5 | 4 | 7 |