List your company

Safety and Utility Evaluation on BeaverTails & LMSYS-Chat

97.88Rule ScoreShadow-Level
67.886475.673283.4691.2468Jan 12, 2026

Evaluation results

MethodLinks
Shadow-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0197.8896.92-0.875.17
Step-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0170.3858.46-2.425.04
Client-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0169.0456.73-2.525.14