List your company

Jailbreak Attack Evaluation on JBB sampled harmful behaviors

100PAIR Success RateM
-4235077Nov 11, 2025

Evaluation results

MethodLinks
MModel Backbone=Gemma-9B, Variant=Unaligned baseline instruction tuned model2025.1110010010010
MsafepromptModel Backbone=Gemma-9B, Variant=Safe prompt baseline2025.1110010010010
MModel Backbone=Mistral-7B, Variant=Unaligned baseline instruction tuned model2025.1110010010010
MsafepromptModel Backbone=Mistral-7B, Variant=Safe prompt baseline2025.1110010010010
MModel Backbone=Llama-3-8B, Variant=Unaligned baseline instruction tuned model2025.1110010010010
MsafepromptModel Backbone=Llama-3-8B, Variant=Safe prompt baseline2025.1110010010010
M+Model Backbone=Gemma-9B, Variant=Policy patch / PEFT safety adapter2025.110001
M'Model Backbone=Gemma-9B, Variant=Improved model variant2025.110001
M+Model Backbone=Mistral-7B, Variant=Policy patch / PEFT safety adapter2025.110001
M'Model Backbone=Mistral-7B, Variant=Improved model variant2025.110001
M+Model Backbone=Llama-3-8B, Variant=Policy patch / PEFT safety adapter2025.110001
M'Model Backbone=Llama-3-8B, Variant=Improved model variant2025.110001