List your company

Indirect Prompt Injection Defense on Wildchat

0.33Attack Success Rate (ASR)CachePrune
-0.29683.93418.16512.3959Apr 29, 2025

Evaluation results

MethodLinks
CachePruneModel=Mistral-7B2025.040.333.9
DelimitingModel=Mistral-7B2025.040.53.93
Encode_Base64Model=Mistral-7B2025.040.61.24
DatamarkingModel=Mistral-7B2025.041.53.91
CachePruneModel=Phi-3.5-mini-instruct (3.8B)2025.041.893.6
CachePruneModel=LLama3-8B2025.0423.32
VanillaModel=Mistral-7B2025.0423.88
SandwichModel=Mistral-7B2025.042.53.85
DatamarkingModel=Phi-3.5-mini-instruct (3.8B)2025.042.533.71
DelimitingModel=Phi-3.5-mini-instruct (3.8B)2025.043.23.84
VanillaModel=Phi-3.5-mini-instruct (3.8B)2025.043.323.78
Encode_Base64Model=Phi-3.5-mini-instruct (3.8B)2025.043.561.09
SandwichModel=Phi-3.5-mini-instruct (3.8B)2025.044.373.56
Encode_Base64Model=LLama3-8B2025.045.51.52
DatamarkingModel=LLama3-8B2025.047.52.98
SandwichModel=LLama3-8B2025.0413.013.22
VanillaModel=LLama3-8B2025.0414.53.32
DelimitingModel=LLama3-8B2025.04163.12