List your company

Safety Evaluation on XSTest Safe (Refusal Rate)

51.6Refusal RateTarget feature
-2.06411.86825.839.732Jun 25, 2026

Evaluation results

MethodLinks
Target featureModel=Qwen3-4B-it2026.0651.6
Target featureModel=Qwen3-1.7B-it2026.069.6
No SteeringModel=Qwen3-1.7B-it2026.064.4
No SteeringModel=Qwen3-4B-it2026.063.2
Random featureModel=Qwen3-4B-it2026.063.1
Random featureModel=Qwen3-1.7B-it2026.060