List your company

Jailbreak Attack on Advbench Harmful Behaviors (subset 100 samples)

100Attack Success Rate AKSCAV
80.39685.485590.57595.6645Jul 8, 2025

Evaluation results

MethodLinks
SCAVVictim LLM=Llama3.1-8B2025.0710090.6587.1195.1999.23
SCAVVictim LLM=Qwen2.5-7B2025.0799.8587.5478.6598.07100
SCAVVictim LLM=Mistral-8B2025.0799.2478.2682.380.7685.19
CAVGANVictim LLM=Qwen2.5-7B2025.0798.9883.8870.4186.7399.8
CAVGANVictim LLM=Llama3.1-8B2025.0798.7888.3878.1688.3899.38
CAVGANVictim LLM=Mistral-8B2025.0795.5194.2988.7895.199.18
JREVictim LLM=Qwen2.5-7B2025.0783.547060.9661.1555
JREVictim LLM=Mistral-8B2025.0782.886561.9255.7656.73
JREVictim LLM=Llama3.1-8B2025.0781.156762.8856.3456.92