List your company

Safety Evaluation on WildGuard (Unsafe)

54.9Refusal RateTarget feature
-2.19612.62727.4542.273Jun 25, 2026

Evaluation results

MethodLinks
Target featureModel=Qwen3-4B-it2026.0654.9
No SteeringModel=Qwen3-4B-it2026.0640.5
Target featureModel=Qwen3-1.7B-it2026.0640.1
No SteeringModel=Qwen3-1.7B-it2026.0626.3
Random featureModel=Qwen3-1.7B-it2026.060.9
Random featureModel=Qwen3-4B-it2026.060