List your company

Safety Refusal Evaluation on PolyRefuse am 1.0

93.2Harmful Refusal RateAdaSteer (HRL + 32 LRL)
0.84824.82448.872.776May 31, 2026

Evaluation results

MethodLinks
AdaSteer (HRL + 32 LRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=AdaSteer (HRL + 32 LRL)2026.0593.284.29
Ours (HRL + 32 LRL)Backbone=Gemma-2-9B, Steering Method=HRL + 32 LRL2026.0591.616.475.2
Ours (HRL)Backbone=Gemma-2-9B, Steering Method=HRL2026.0584.312.671.7
CAST (Original)Backbone=Gemma-2-9B, Steering Method=CAST (Original)2026.0577.61859.6
CAST (HRL + 32 LRL)Backbone=Gemma-2-9B, Steering Method=CAST (HRL + 32 LRL)2026.0574.315.858.5
Gemma-2-9BBackbone=Gemma-2-9B, Steering Method=None2026.0572.714.458.3
AdaSteer (Original)Backbone=Qwen2.5-7B-Instruct, Steering Method=AdaSteer (Original)2026.0571.981.6-9.7
AdaSteer (Original)Backbone=Gemma-2-9B, Steering Method=AdaSteer (Original)2026.0570.529.640.9
CAST (Original)Backbone=Llama-3.1-8B-Instruct, Steering Method=CAST (Original)2026.0558.9535.9
AdaSteer (HRL + 32 LRL)Backbone=Gemma-2-9B, Steering Method=AdaSteer (HRL + 32 LRL)2026.054072.4-32.4
Ours (HRL + 32 LRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=HRL + 32 LRL2026.0532.38.423.9
Ours (HRL + 32 LRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=HRL + 32 LRL2026.0532.210.421.8
CAST (HRL + 32 LRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=CAST (HRL + 32 LRL)2026.0526.418.87.6
Ours (HRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=HRL2026.0523.36.217.1
AdaSteer (Original)Backbone=Llama-3.1-8B-Instruct, Steering Method=AdaSteer (Original)2026.0516.8151.8
CAST (Original)Backbone=Qwen2.5-7B-Instruct, Steering Method=CAST (Original)2026.0510.310.20.1
Llama-3.1-8B-InstructBackbone=Llama-3.1-8B-Instruct, Steering Method=None2026.059.872.8
AdaSteer (HRL + 32 LRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=AdaSteer (HRL + 32 LRL)2026.059.320.6-11.3
Ours (HRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=HRL2026.058.962.9
CAST (HRL + 32 LRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=CAST (HRL + 32 LRL)2026.056.160.1
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct, Steering Method=None2026.054.45.4-1