List your company

Safety Refusal Evaluation on PolyRefuse 1.0 (si)

90.2Harmful Refusal RateOurs (HRL + 32 LRL)
28.94444.84760.7576.653May 31, 2026

Evaluation results

MethodLinks
Ours (HRL + 32 LRL)Backbone=Gemma-2-9B, Steering Method=HRL + 32 LRL2026.0590.2882.2
CAST (Original)Backbone=Llama-3.1-8B-Instruct, Steering Method=CAST (Original)2026.0588.580.67.9
AdaSteer (HRL + 32 LRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=AdaSteer (HRL + 32 LRL)2026.058692.2-6.2
Ours (HRL)Backbone=Gemma-2-9B, Steering Method=HRL2026.0581.85.276.6
CAST (HRL + 32 LRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=CAST (HRL + 32 LRL)2026.0581.157.223.9
CAST (Original)Backbone=Gemma-2-9B, Steering Method=CAST (Original)2026.0575.28.466.8
Ours (HRL + 32 LRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=HRL + 32 LRL2026.0574.725.649.1
AdaSteer (Original)Backbone=Qwen2.5-7B-Instruct, Steering Method=AdaSteer (Original)2026.0573.353.819.5
CAST (HRL + 32 LRL)Backbone=Gemma-2-9B, Steering Method=CAST (HRL + 32 LRL)2026.0572.28.663.6
Gemma-2-9BBackbone=Gemma-2-9B, Steering Method=None2026.05675.861.2
Ours (HRL + 32 LRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=HRL + 32 LRL2026.0564.2856.2
Ours (HRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=HRL2026.0561.216.844.4
AdaSteer (Original)Backbone=Llama-3.1-8B-Instruct, Steering Method=AdaSteer (Original)2026.0558.222.236
AdaSteer (Original)Backbone=Gemma-2-9B, Steering Method=AdaSteer (Original)2026.0557.316.840.5
CAST (Original)Backbone=Qwen2.5-7B-Instruct, Steering Method=CAST (Original)2026.0548.417.431
AdaSteer (HRL + 32 LRL)Backbone=Gemma-2-9B, Steering Method=AdaSteer (HRL + 32 LRL)2026.054745.61.4
AdaSteer (HRL + 32 LRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=AdaSteer (HRL + 32 LRL)2026.0545.620.625
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct, Steering Method=None2026.0541.411.629.8
Llama-3.1-8B-InstructBackbone=Llama-3.1-8B-Instruct, Steering Method=None2026.0540.94.836.1
CAST (HRL + 32 LRL)Backbone=Qwen2.5-7B-Instruct, Steering Method=CAST (HRL + 32 LRL)2026.0536.413.423
Ours (HRL)Backbone=Llama-3.1-8B-Instruct, Steering Method=HRL2026.0531.34.626.7