List your company

Safety Evaluation on JailbreakBench (test)

1.12ASR (Llama-Guard-3-8B)MLP
-2.170420.039842.2564.4602May 6, 2026

Evaluation results

MethodLinks
MLPBackbone=Qwen2.5-72B2026.051.124.51
MLPBackbone=Qwen2.5-7B2026.051.623.621.25
MLPBackbone=Llama-3.1-70B2026.054.122.752.62
MoEBackbone=Gemma-2-9B2026.0554.386.25
MoEBackbone=Gemma-2-27B2026.055.014.625.75
MoEBackbone=Llama-3.1-8B2026.055.384.386
MoEBackbone=Mistral-7B-v0.32026.0510.387.3810.12
FMBackbone=Gemma-2-27B2026.0511.19.6214.88
MLPBackbone=Gemma-2-27B2026.0511.129.8813.25
MoEBackbone=Qwen2.5-72B2026.0513.121314
FMBackbone=Gemma-2-9B2026.05149.2515.75
CVAEBackbone=Qwen2.5-72B2026.05151611.38
MoEBackbone=Llama-3.1-70B2026.051616.3814.75
FMBackbone=Llama-3.1-8B2026.0518.1215.7525.25
MLPBackbone=Gemma-2-9B2026.0518.3817.3816.38
MoEBackbone=Qwen2.5-7B2026.0519.525.124.88
MLPBackbone=Mistral-7B-v0.32026.0521.75108.5
MLPBackbone=Llama-3.1-8B2026.0527.3815.7517.38
FMBackbone=Qwen2.5-7B2026.0530.1620.2526.12
Safe LoRABackbone=Qwen2.5-72B2026.0530.2939.2528.25
CVAEBackbone=Gemma-2-9B2026.0535.2534.3831.87
FMBackbone=Qwen2.5-72B2026.0535.882948
FMBackbone=Mistral-7B-v0.32026.0537.525.1240.12
AEBackbone=Gemma-2-27B2026.0539.3631.2542.75
CVAEBackbone=Qwen2.5-7B2026.0542.6254.8748.38
FMBackbone=Llama-3.1-70B2026.0544.538.6249.62
Safe LoRABackbone=Gemma-2-9B2026.0545.1245.2548.38
AEBackbone=Gemma-2-9B2026.0546.493955
AEBackbone=Qwen2.5-72B2026.0546.7540.2557.25
CVAEBackbone=Llama-3.1-8B2026.0549.555.8855.75
CVAEBackbone=Mistral-7B-v0.32026.0549.6245.2553.12
Safe LoRABackbone=Llama-3.1-8B2026.0550.2549.3864
Safe LoRABackbone=Mistral-7B-v0.32026.0552.3841.8857
AEBackbone=Qwen2.5-7B2026.0553.1439.7552.75
CVAEBackbone=Gemma-2-27B2026.0555.2542.3852.25
AEBackbone=Llama-3.1-8B2026.0556.8254.3767.88
Safe LoRABackbone=Qwen2.5-7B2026.0559.5765.8858.5
Unsafe (Finetuned)Backbone=Qwen2.5-72B2026.0560.255080.5
SaLoRABackbone=Qwen2.5-72B2026.0562.2553.6273.38
Unsafe (Finetuned)Backbone=Gemma-2-9B2026.0563.0451.8889.88
CVAEBackbone=Llama-3.1-70B2026.0563.6252.548.25
AEBackbone=Mistral-7B-v0.32026.0565.1247.7569
Unsafe (Finetuned)Backbone=Llama-3.1-70B2026.0565.8158.8186.38
Unsafe (Finetuned)Backbone=Gemma-2-27B2026.0566.3753.4486.94
Unsafe (Finetuned)Backbone=Llama-3.1-8B2026.0566.8560.5690.44
AEBackbone=Llama-3.1-70B2026.0567.3857.3870.75
Safe LoRABackbone=Llama-3.1-70B2026.0568.1263.6261.12
Safe LoRABackbone=Gemma-2-27B2026.057267.3863.12
SaLoRABackbone=Gemma-2-9B2026.0573.095790.62
Unsafe (Finetuned)Backbone=Qwen2.5-7B2026.0573.3658.8181.69
Unsafe (Finetuned)Backbone=Mistral-7B-v0.32026.0574.2553.8792.06
SaLoRABackbone=Llama-3.1-8B2026.0575.8470.2590.12
SaLoRABackbone=Gemma-2-27B2026.0576.2561.2584
SaLoRABackbone=Llama-3.1-70B2026.0578.3867.6285.5
SaLoRABackbone=Qwen2.5-7B2026.0579.9763.1278.12
SaLoRABackbone=Mistral-7B-v0.32026.0583.386289.88