hku-tasr / imperio Goto Github PK
View Code? Open in Web Editor NEW[IJCAI 2024] Imperio is an LLM-powered backdoor attack. It allows the adversary to issue language-guided instructions to control the victim model's prediction for arbitrary targets.
Home Page: https://khchow.com/Imperio/
License: MIT License