List your company

Jailbreak Detection on PAIR

98AccuracyIMAG
-3.9222.544975.46Dec 3, 2025

Evaluation results

MethodLinks
IMAGTarget LLM=Vicuna-13B2025.129899
IMAGTarget LLM=Mistral-7B2025.129396
IMAGTarget LLM=Vicuna-7B2025.128893
IMAGTarget LLM=Llama3-8B2025.127887
IMAGTarget LLM=Llama2-7B2025.127687
LlamaGTarget LLM=Vicuna-7B2025.127585
LlamaGTarget LLM=Vicuna-13B2025.127585
LlamaGTarget LLM=Mistral-7B2025.127485
LlamaGTarget LLM=Llama3-8B2025.127082
GradSafeTarget LLM=Llama2-7B2025.126277
LlamaGTarget LLM=Llama2-7B2025.125369
Self-ExTarget LLM=Mistral-7B2025.124663
Self-ExTarget LLM=Llama2-7B2025.123235
Self-ExTarget LLM=Llama3-8B2025.121627
PPLTarget LLM=Llama2-7B2025.121018
OAPITarget LLM=Mistral-7B2025.12713
OAPITarget LLM=Vicuna-13B2025.12510
OAPITarget LLM=Vicuna-7B2025.1249
Self-ExTarget LLM=Vicuna-7B2025.1236
GradSafeTarget LLM=Vicuna-7B2025.1236
OAPITarget LLM=Llama2-7B2025.1235
OAPITarget LLM=Llama3-8B2025.1236
PPLTarget LLM=Vicuna-13B2025.1212
PPLTarget LLM=Mistral-7B2025.1200
GradSafeTarget LLM=Mistral-7B2025.1200
PPLTarget LLM=Vicuna-7B2025.1200
Self-ExTarget LLM=Vicuna-13B2025.1200
GradSafeTarget LLM=Vicuna-13B2025.1200
PPLTarget LLM=Llama3-8B2025.1200
GradSafeTarget LLM=Llama3-8B2025.1200