List your company

Safety Evaluation on DAN

91Safety Score (DAN)VCL
-3.0482421.3681345.784570.20087Mar 4, 2026Mar 14, 2026Mar 25, 2026Apr 5, 2026Apr 15, 2026Apr 26, 2026May 7, 2026

Evaluation results

MethodLinks
VCLBase Model=Llama-3.1-8B2026.0391———
Vector AblationBase Model=Llama-3.1-8B2026.0386———
DROBase Model=Llama-3.1-8B2026.0383———
Llama-3.1-8B-SFTBase Model=Llama-3.1-8B2026.0382———
System PromptBase Model=Llama-3.1-8B2026.0382———
Self-CDBase Model=Llama-3.1-8B2026.0380———
Causal-DRPolicy Model=LLaMA2-7B2026.030.80712.2——
Causal-IPSPolicy Model=LLaMA2-7B2026.030.79510.5——
SelectiveRMBackbone=LLaMA2-7B2026.050.7951.66——
ILDEBackbone=LLaMA2-7B2026.050.7860.51——
CSGNBackbone=LLaMA2-7B2026.050.7840.18——
SDRPolicy Model=LLaMA2-7B2026.030.7828.7——
NaiveBackbone=LLaMA2-7B2026.050.782———
ILDEPolicy Model=LLaMA2-7B2026.030.7767.8——
IPSPolicy Model=LLaMA2-7B2026.030.7514.4——
NaivePolicy Model=LLaMA2-7B2026.030.72———
SelectiveRMBackbone=Qwen2.5-7B2026.050.6843.89——
Causal-DRPolicy Model=Qwen2.5-7B2026.030.6819.6——
ILDEBackbone=Qwen2.5-7B2026.050.6711.94——
CSGNBackbone=Qwen2.5-7B2026.050.6681.5——
Causal-IPSPolicy Model=Qwen2.5-7B2026.030.66316.7——
NaiveBackbone=Qwen2.5-7B2026.050.658———
ILDEPolicy Model=Qwen2.5-7B2026.030.65615.4——
SDRPolicy Model=Qwen2.5-7B2026.030.65314.9——
IPSPolicy Model=Qwen2.5-7B2026.030.6025.9——
NaivePolicy Model=Qwen2.5-7B2026.030.569———
AdvGameAttacker=Qwen2.5-7B2026.06———0.103
AdvGRPO (ST)Attacker=Qwen2.5-7B2026.06———0.073
AdvGRPO (ST)Attacker=Qwen2.5-14B2026.06———0.04
BaseAttacker=None2026.06———0.393
KLSelection Criterion=KL2026.04——19.1—
Latent SimilaritySelection Criterion=Latent Similarity, Layer=122026.04——28.2—
Latent SimilaritySelection Criterion=Latent Similarity, Layer=Best Layer2026.04——42.2—
PerplexitySelection Criterion=Perplexity2026.04——-26.2—
RandomSelection Criterion=Random2026.04——-1.5—
Self CertaintySelection Criterion=Self Certainty2026.04——-46.2—
Self-RedTeamAttacker=Qwen2.5-7B2026.06———0.366