List your company

Jailbreak Detection on Base64

100AccuracyIMAG
-4235077Dec 3, 2025

Evaluation results

MethodLinks
IMAGTarget LLM=Llama3-8B2025.12100100
PPLTarget LLM=Mistral-7B2025.129595
IMAGTarget LLM=Mistral-7B2025.129597
PPLTarget LLM=Vicuna-7B2025.129595
PPLTarget LLM=Vicuna-13B2025.129595
PPLTarget LLM=Llama2-7B2025.129595
PPLTarget LLM=Llama3-8B2025.129595
IMAGTarget LLM=Vicuna-7B2025.129195
IMAGTarget LLM=Vicuna-13B2025.129195
IMAGTarget LLM=Llama2-7B2025.126075
LlamaGTarget LLM=Mistral-7B2025.125067
LlamaGTarget LLM=Vicuna-7B2025.124965
LlamaGTarget LLM=Llama2-7B2025.124965
LlamaGTarget LLM=Vicuna-13B2025.124864
LlamaGTarget LLM=Llama3-8B2025.123451
Self-ExTarget LLM=Mistral-7B2025.123249
Self-ExTarget LLM=Llama2-7B2025.122032
Self-ExTarget LLM=Llama3-8B2025.121221
Self-ExTarget LLM=Vicuna-7B2025.1212
OAPITarget LLM=Mistral-7B2025.1200
GradSafeTarget LLM=Mistral-7B2025.1200
OAPITarget LLM=Vicuna-7B2025.1200
GradSafeTarget LLM=Vicuna-7B2025.1200
OAPITarget LLM=Vicuna-13B2025.1200
Self-ExTarget LLM=Vicuna-13B2025.1200
GradSafeTarget LLM=Vicuna-13B2025.1200
OAPITarget LLM=Llama2-7B2025.1200
GradSafeTarget LLM=Llama2-7B2025.1200
OAPITarget LLM=Llama3-8B2025.1200
GradSafeTarget LLM=Llama3-8B2025.1200