List your company
ResearchFormerly WizWand

Jailbreak Defense on Dictionary-based Jailbreak Evaluation

0GCG ASRIntention Analysis
-3.9222.544975.46Aug 9, 2025
Updated 4mo ago

Evaluation results

MethodLinks
Intention AnalysisBase LLM=Vicuna2025.0800102046
Self-ReminderBase LLM=Llama22025.08026142—
Intention AnalysisBase LLM=Llama22025.0800000—
Safe DecodingBase LLM=Llama22025.08001040—
Context FilteringBase LLM=Llama22025.0800020—
Self-ReminderBase LLM=ChatGPT2025.08006247286
ICDBase LLM=ChatGPT2025.080224080
Self-ExaminationBase LLM=ChatGPT2025.0800446028
Intention AnalysisBase LLM=ChatGPT2025.08000200
Context FilteringBase LLM=ChatGPT2025.080002828
ICDBase LLM=Llama22025.0820400—
Safe DecodingBase LLM=Vicuna2025.0840204096
No DefenseBase LLM=ChatGPT2025.084420348294
Context FilteringBase LLM=Vicuna2025.086210181048
Self-ExaminationBase LLM=Vicuna2025.0812424128888
Self-ExaminationBase LLM=Llama22025.08120202—
No DefenseBase LLM=Llama22025.08322218100
Self-ReminderBase LLM=Vicuna2025.084868444610098
ICDBase LLM=Vicuna2025.08728058404096
No DefenseBase LLM=Vicuna2025.0898885688100100