List your company

Jailbreak Attack on AdvBench 14 (test)

97.02ASRShallowJail
-3.776822.391648.5674.7284Feb 6, 2026

Evaluation results

MethodLinks
ShallowJailVictim Model=Llama-3.1-8B-Instruct, tau=150, alpha=0.8, beta=0.52026.0297.0215.654
ShallowJailVictim Model=Qwen3-4B-Instruct-2507, tau=150, alpha=5, beta=0.52026.0290.194.3697
ShallowJailVictim Model=Qwen2.5-7B-Instruct, tau=150, alpha=6.5, beta=0.52026.0286.154.3208
DirectVictim Model=Llama-3.1-8B-Instruct, tau=150, alpha=0.8, beta=0.52026.026.351.5404
DirectVictim Model=Qwen2.5-7B-Instruct, tau=150, alpha=6.5, beta=0.52026.020.481.8692
DirectVictim Model=Qwen3-4B-Instruct-2507, tau=150, alpha=5, beta=0.52026.020.11.3062