List your company

Adversarial Attack on 16 malicious prompts

0ASRBaseline
-3.56420.49344.5568.607May 1, 2026

Evaluation results

MethodLinks
BaselineTarget Model=Qwen3-30B-A3B-Instruct-25072026.050
GCGTarget Model=Qwen3-30B-A3B-Instruct-25072026.050
SteerMoETarget Model=Qwen3-30B-A3B-Instruct-25072026.050
BaselineTarget Model=Hunyuan-A13B-Instruct2026.052.6
BaselineTarget Model=Phi-3.5-MoE-Instruct2026.053.2
BaselineTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.054.2
SteerMoETarget Model=Qwen1.5-MoE-A2.7B-Chat2026.054.2
BaselineTarget Model=Pangu-Pro-MoE-72B2026.055.2
GCGTarget Model=Phi-3.5-MoE-Instruct2026.056.2
BaselineTarget Model=Average2026.057.1
GCGTarget Model=Pangu-Pro-MoE-72B2026.057.4
GCGTarget Model=Hunyuan-A13B-Instruct2026.058
SteerMoETarget Model=Phi-3.5-MoE-Instruct2026.0511.2
BaselineTarget Model=Mixtral-8x7B-Instruct-v0.12026.0514.6
SteerMoETarget Model=Hunyuan-A13B-Instruct2026.0514.6
SteerMoETarget Model=Average2026.0517
SteerMoETarget Model=Mixtral-8x7B-Instruct-v0.12026.0519.2
BaselineTarget Model=DeepSeek-MoE-16B-Chat2026.0519.8
GCGTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.0521.1
GCGTarget Model=Average2026.0521.9
SAFExTarget Model=Phi-3.5-MoE-Instruct2026.0526.8
SAFExTarget Model=Qwen3-30B-A3B-Instruct-25072026.0528.4
SAFExTarget Model=Hunyuan-A13B-Instruct2026.0528.4
SAFExTarget Model=Pangu-Pro-MoE-72B2026.0530
SteerMoETarget Model=Pangu-Pro-MoE-72B2026.0532.2
GCGTarget Model=Mixtral-8x7B-Instruct-v0.12026.0532.3
SAFExTarget Model=Average2026.0533.3
RouteHijackTarget Model=Phi-3.5-MoE-Instruct2026.0534.7
SAFExTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.0535
SAFExTarget Model=DeepSeek-MoE-16B-Chat2026.0535.6
SteerMoETarget Model=DeepSeek-MoE-16B-Chat2026.0537.5
SAFExTarget Model=Mixtral-8x7B-Instruct-v0.12026.0548.8
RouteHijackTarget Model=Pangu-Pro-MoE-72B2026.0555.9
RouteHijackTarget Model=Average2026.0569.3
RouteHijackTarget Model=Qwen3-30B-A3B-Instruct-25072026.0570.6
RouteHijackTarget Model=Hunyuan-A13B-Instruct2026.0572.2
RouteHijackTarget Model=Mixtral-8x7B-Instruct-v0.12026.0575.1
GCGTarget Model=DeepSeek-MoE-16B-Chat2026.0578.6
RouteHijackTarget Model=Qwen1.5-MoE-A2.7B-Chat2026.0587.2
RouteHijackTarget Model=DeepSeek-MoE-16B-Chat2026.0589.1