List your company

Safety Performance on JBB

53Refusal Score (CR)CAST
-2.1212.1926.540.81Oct 9, 2025

Evaluation results

MethodLinks
CASTModel=QWEN 3 1.7B2025.1053—
AdaSteerModel=QWEN 3 1.7B2025.1053—
AlphaSteerModel=QWEN 3 1.7B2025.1052—
SurgicalModel=QWEN 3 1.7B2025.1051—
QWEN 3 1.7BModel=QWEN 3 1.7B2025.1049—
ELSModel=QWEN 3 1.7B2025.1043—
system promptModel=QWEN 3 1.7B2025.1027—
CASTModel=QWEN 3 14B2025.1017—
SurgicalModel=QWEN 3 14B2025.1016—
AdaSteerModel=QWEN 3 14B2025.1016—
CASTModel=QWEN 3 8B2025.1014—
QWEN 3 14BModel=QWEN 3 14B2025.1014—
AlphaSteerModel=QWEN 3 14B2025.1014—
AdaSteerModel=LLAMA3.1-8B-INST2025.1013—
SurgicalModel=QWEN 3 8B2025.1013—
AdaSteerModel=QWEN 3 8B2025.1013—
CASTModel=LLAMA3.1-8B-INST2025.1012—
QWEN 3 8BModel=QWEN 3 8B2025.1012—
AlphaSteerModel=QWEN 3 8B2025.1012—
SurgicalModel=LLAMA3.1-8B-INST2025.1011—
AlphaSteerModel=LLAMA3.1-8B-INST2025.1011—
ELSModel=QWEN 3 8B2025.1011—
LLAMA3.1-8B-INSTModel=LLAMA3.1-8B-INST2025.1010—
ELSModel=LLAMA3.1-8B-INST2025.1010—
ELSModel=QWEN 3 14B2025.1010—
CASTModel=LLAMA2-7B-CHAT2025.107—
AlphaSteerModel=LLAMA2-7B-CHAT2025.106—
system promptModel=QWEN 3 8B2025.106—
SurgicalModel=LLAMA2-7B-CHAT2025.105—
AdaSteerModel=LLAMA2-7B-CHAT2025.105—
system promptModel=LLAMA3.1-8B-INST2025.103—
LLAMA2-7B-CHATModel=LLAMA2-7B-CHAT2025.103—
ELSModel=LLAMA2-7B-CHAT2025.103—
system promptModel=QWEN 3 14B2025.103—
system promptModel=LLAMA2-7B-CHAT2025.100—
CAABase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.12—30.1
Gradient CuffBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.12—68
GSAEBase Model=Llama 3 8B, Method Category=Main Method, Random Seeds=52025.12—76.2
GSAE-1DBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.12—40.1
Input Gate OnlyBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.12—70.3
No gatingBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.12—64.1
No SteeringBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.12—-9.1
Prompt guardrailsBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.12—10.2
Random graphsBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.12—44.2
SAE steeringBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.12—36
SafeSwitchBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.12—51.4