Jailbreak Attack on AgentHarm
2Attack Success Score (ASS)Base
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BaseTarget Agent=GPT Agent2026.05 | 2 | 4 | |
| ReNeLLMTarget Agent=GPT Agent2026.05 | 2 | 4 | |
| X-TeamingTarget Agent=GPT Agent2026.05 | 5 | 15 | |
| AutoDAN-TurboTarget Agent=GPT Agent2026.05 | 12 | 17 | |
| Red-Agent-ReflectTarget Agent=GPT Agent2026.05 | 15 | 31 | |
| BaseTarget Agent=Gemini Agent2026.05 | 26 | 33 | |
| BaseTarget Agent=DeepSeek Agent2026.05 | 34 | 42 | |
| ReNeLLMTarget Agent=Gemini Agent2026.05 | 38 | 60 | |
| ReNeLLMTarget Agent=DeepSeek Agent2026.05 | 38 | 52 | |
| TRACETarget Agent=GPT Agent2026.05 | 40 | 90 | |
| AutoDAN-TurboTarget Agent=DeepSeek Agent2026.05 | 42 | 73 | |
| X-TeamingTarget Agent=DeepSeek Agent2026.05 | 43 | 52 | |
| X-TeamingTarget Agent=Gemini Agent2026.05 | 46 | 69 | |
| AutoDAN-TurboTarget Agent=Gemini Agent2026.05 | 52 | 77 | |
| Red-Agent-ReflectTarget Agent=Gemini Agent2026.05 | 59 | 87 | |
| Red-Agent-ReflectTarget Agent=DeepSeek Agent2026.05 | 61 | 81 | |
| TRACETarget Agent=DeepSeek Agent2026.05 | 72 | 100 | |
| TRACETarget Agent=Gemini Agent2026.05 | 73 | 98 |