Attack Success Rate on InjecAgent
88.3Attack Success Rate (ASR)Llama-4
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-4Prompt=InjecPrompt, Technique=ChatInject + tool2025.09 | 88.3 | |
| Llama-4Prompt=Multi-turn, Technique=ChatInject2025.09 | 88.3 | |
| Llama-4Prompt=InjecPrompt, Technique=ChatInject2025.09 | 79.4 | |
| GLM-4.5Prompt=InjecPrompt, Technique=ChatInject + tool2025.09 | 72.2 | |
| Kimi-K2Prompt=InjecPrompt, Technique=ChatInject + tool2025.09 | 72.2 | |
| GLM-4.5Prompt=Multi-turn, Technique=ChatInject2025.09 | 71.5 | |
| GLM-4.5Prompt=InjecPrompt, Technique=ChatInject + think2025.09 | 69.3 | |
| Kimi-K2Prompt=InjecPrompt, Technique=ChatInject2025.09 | 67.4 | |
| Qwen-3Prompt=Multi-turn, Technique=ChatInject2025.09 | 65.9 | |
| Kimi-K2Prompt=Multi-turn, Technique=ChatInject2025.09 | 61 | |
| GLM-4.5Prompt=InjecPrompt, Technique=ChatInject2025.09 | 57.3 | |
| Llama-4Prompt=InjecPrompt, Technique=default2025.09 | 50.1 | |
| Qwen-3Prompt=InjecPrompt, Technique=ChatInject + tool2025.09 | 42.1 | |
| Qwen-3Prompt=InjecPrompt, Technique=ChatInject + think2025.09 | 40.1 | |
| Qwen-3Prompt=InjecPrompt, Technique=ChatInject2025.09 | 39.4 | |
| GPT-ossPrompt=InjecPrompt, Technique=ChatInject + tool2025.09 | 19.1 | |
| Grok-2Prompt=InjecPrompt, Technique=ChatInject2025.09 | 17.7 | |
| Kimi-K2Prompt=Multi-turn, Technique=default2025.09 | 17.2 | |
| GPT-ossPrompt=Multi-turn, Technique=ChatInject2025.09 | 16.9 | |
| GPT-ossPrompt=InjecPrompt, Technique=ChatInject + think2025.09 | 16.7 | |
| Llama-4Prompt=Multi-turn, Technique=default2025.09 | 16.6 | |
| Grok-2Prompt=InjecPrompt, Technique=default2025.09 | 16.5 | |
| Kimi-K2Prompt=InjecPrompt, Technique=default2025.09 | 15.7 | |
| GPT-ossPrompt=InjecPrompt, Technique=ChatInject2025.09 | 14.2 | |
| Qwen-3Prompt=Multi-turn, Technique=default2025.09 | 10.7 | |
| Grok-2Prompt=Multi-turn, Technique=ChatInject2025.09 | 10.4 | |
| Qwen-3Prompt=InjecPrompt, Technique=default2025.09 | 8.5 | |
| Grok-2Prompt=Multi-turn, Technique=default2025.09 | 1.6 | |
| GPT-ossPrompt=Multi-turn, Technique=default2025.09 | 0.1 | |
| GLM-4.5Prompt=Multi-turn, Technique=default2025.09 | 0.1 | |
| GPT-ossPrompt=InjecPrompt, Technique=default2025.09 | 0 | |
| GLM-4.5Prompt=InjecPrompt, Technique=default2025.09 | 0 |