Indirect Prompt Injection on Agent Action Subset 2
0IRUser Review
Evaluation Results
| Method | Links | |
|---|---|---|
| User ReviewMemory regime=Tool, Model=Sonnet-4.62026.05 | 0 | |
| User ReviewMemory regime=External, Model=Sonnet-4.62026.05 | 0 | |
| User ReviewMemory regime=External, Model=GPT-5.42026.05 | 1 | |
| User ReviewMemory regime=External, Model=GPT-5.52026.05 | 1.5 | |
| User ReviewMemory regime=External, Model=Gemini-3.12026.05 | 1.5 | |
| User ReviewMemory regime=External, Model=Kimi-K2.62026.05 | 1.5 | |
| User ReviewMemory regime=Tool, Model=GPT-5.52026.05 | 2.5 | |
| User ReviewMemory regime=Tool, Model=GPT-5.42026.05 | 3 | |
| User ReviewMemory regime=External, Model=DeepSeek-v42026.05 | 5 | |
| Actor-CriticMemory regime=External, Model=Sonnet-4.62026.05 | 5.5 | |
| Actor-CriticMemory regime=Tool, Model=Sonnet-4.62026.05 | 6.5 | |
| User ReviewMemory regime=Tool, Model=Gemini-3.12026.05 | 37 | |
| User ReviewMemory regime=Tool, Model=Kimi-K2.62026.05 | 42.5 | |
| Actor-CriticMemory regime=External, Model=GPT-5.52026.05 | 46.5 | |
| Actor-CriticMemory regime=External, Model=Gemini-3.12026.05 | 59 | |
| User ReviewMemory regime=Tool, Model=DeepSeek-v42026.05 | 59.5 | |
| Actor-CriticMemory regime=External, Model=GPT-5.42026.05 | 61.5 | |
| Actor-CriticMemory regime=External, Model=Kimi-K2.62026.05 | 65 | |
| Actor-CriticMemory regime=Tool, Model=Gemini-3.12026.05 | 67 | |
| Actor-CriticMemory regime=Tool, Model=Kimi-K2.62026.05 | 81 | |
| Actor-CriticMemory regime=External, Model=DeepSeek-v42026.05 | 82.5 | |
| Actor-CriticMemory regime=Tool, Model=DeepSeek-v42026.05 | 88.5 | |
| Actor-CriticMemory regime=Tool, Model=GPT-5.52026.05 | 91.5 | |
| Actor-CriticMemory regime=Tool, Model=GPT-5.42026.05 | 97 |