List your company

Prompt Injection Attack on SQuAD

10.32Attack Success Rate (ASR)Vanilla
1.19923.56715.9358.3029Apr 29, 2025

Evaluation results

MethodLinks
VanillaModel=Phi3.5-mini-instruct, Attack Type=Adaptive, Attack Tokens (K)=10, Epochs (E)=202025.0410.3226.13
DelimitingModel=Phi3.5-mini-instruct, Attack Type=Adaptive, Attack Tokens (K)=10, Epochs (E)=202025.049.0925.01
DaramarkingModel=Phi3.5-mini-instruct, Attack Type=Adaptive, Attack Tokens (K)=10, Epochs (E)=202025.045.5326.67
CachePruneModel=Phi3.5-mini-instruct, Attack Type=Adaptive, Attack Tokens (K)=10, Epochs (E)=20, N=82025.041.5525.98