List your company

Safety Performance on WildJailbreak

90.1Selective Refusal Score (Δs)GSAE
-6.72418.41343.5568.687Dec 7, 2025

Evaluation results

MethodLinks
GSAEBase Model=Llama 3 8B, Method Category=Main Method, Random Seeds=52025.1290.1
Input Gate OnlyBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1282.4
Gradient CuffBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1278.7
No gatingBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1278.1
SafeSwitchBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1265.4
Random graphsBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1260.1
GSAE-1DBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1255.4
SAE steeringBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1248.2
CAABase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1242
Prompt guardrailsBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1218.3
No SteeringBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.12-3