Adversarial Attack on 16 malicious prompts
0ASRBaseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BaselineTarget Model=Qwen3-30B-A3B-Instruct-25072026.05 | 0 | |
| GCGTarget Model=Qwen3-30B-A3B-Instruct-25072026.05 | 0 | |
| SteerMoETarget Model=Qwen3-30B-A3B-Instruct-25072026.05 | 0 | |
| BaselineTarget Model=Hunyuan-A13B-Instruct2026.05 | 2.6 | |
| BaselineTarget Model=Phi-3.5-MoE-Instruct2026.05 | 3.2 | |
| BaselineTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.05 | 4.2 | |
| SteerMoETarget Model=Qwen1.5-MoE-A2.7B-Chat2026.05 | 4.2 | |
| BaselineTarget Model=Pangu-Pro-MoE-72B2026.05 | 5.2 | |
| GCGTarget Model=Phi-3.5-MoE-Instruct2026.05 | 6.2 | |
| BaselineTarget Model=Average2026.05 | 7.1 | |
| GCGTarget Model=Pangu-Pro-MoE-72B2026.05 | 7.4 | |
| GCGTarget Model=Hunyuan-A13B-Instruct2026.05 | 8 | |
| SteerMoETarget Model=Phi-3.5-MoE-Instruct2026.05 | 11.2 | |
| BaselineTarget Model=Mixtral-8x7B-Instruct-v0.12026.05 | 14.6 | |
| SteerMoETarget Model=Hunyuan-A13B-Instruct2026.05 | 14.6 | |
| SteerMoETarget Model=Average2026.05 | 17 | |
| SteerMoETarget Model=Mixtral-8x7B-Instruct-v0.12026.05 | 19.2 | |
| BaselineTarget Model=DeepSeek-MoE-16B-Chat2026.05 | 19.8 | |
| GCGTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.05 | 21.1 | |
| GCGTarget Model=Average2026.05 | 21.9 | |
| SAFExTarget Model=Phi-3.5-MoE-Instruct2026.05 | 26.8 | |
| SAFExTarget Model=Qwen3-30B-A3B-Instruct-25072026.05 | 28.4 | |
| SAFExTarget Model=Hunyuan-A13B-Instruct2026.05 | 28.4 | |
| SAFExTarget Model=Pangu-Pro-MoE-72B2026.05 | 30 | |
| SteerMoETarget Model=Pangu-Pro-MoE-72B2026.05 | 32.2 | |
| GCGTarget Model=Mixtral-8x7B-Instruct-v0.12026.05 | 32.3 | |
| SAFExTarget Model=Average2026.05 | 33.3 | |
| RouteHijackTarget Model=Phi-3.5-MoE-Instruct2026.05 | 34.7 | |
| SAFExTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.05 | 35 | |
| SAFExTarget Model=DeepSeek-MoE-16B-Chat2026.05 | 35.6 | |
| SteerMoETarget Model=DeepSeek-MoE-16B-Chat2026.05 | 37.5 | |
| SAFExTarget Model=Mixtral-8x7B-Instruct-v0.12026.05 | 48.8 | |
| RouteHijackTarget Model=Pangu-Pro-MoE-72B2026.05 | 55.9 | |
| RouteHijackTarget Model=Average2026.05 | 69.3 | |
| RouteHijackTarget Model=Qwen3-30B-A3B-Instruct-25072026.05 | 70.6 | |
| RouteHijackTarget Model=Hunyuan-A13B-Instruct2026.05 | 72.2 | |
| RouteHijackTarget Model=Mixtral-8x7B-Instruct-v0.12026.05 | 75.1 | |
| GCGTarget Model=DeepSeek-MoE-16B-Chat2026.05 | 78.6 | |
| RouteHijackTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.05 | 87.2 | |
| RouteHijackTarget Model=DeepSeek-MoE-16B-Chat2026.05 | 89.1 |