ASR

Creator
Creator
Seonglae ChoSeonglae Cho
Created
Created
2024 Dec 20 23:3
Editor
Edited
Edited
2025 Oct 9 10:3
Refs
Refs

Attack Success Rate

Need for evaluation focused on actual harm Attack Success Rate (ASR) should be measured based on successful information leakage or function call criteria, not just by observing a few prefix words and making quick judgments
 
 
 
 
 
 

Recommendations