List your company

Safety and Utility Evaluation on MaliciousGen & WildChat

97.69Rule AdherenceStep-Level
97.294897.397497.597.6026Jan 12, 2026

Evaluation results

MethodLinks
Step-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0197.6995.38-0.85.24
Shadow-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0197.6996.35-0.775.68
Client-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.0197.3195.77-0.765.33