List your company

Safety and Utility Evaluation on MaliciousGen & LMSYS-Chat

97.31Rule ScoreShadow-Level
82.708486.499290.2994.0808Jan 12, 2026

Evaluation results

MethodLinks
Shadow-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0197.3196.35-0.925.06
Step-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0193.0892.12-1.055.16
Client-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0183.2773.27-1.715.01