DRIP: Defending Prompt Injection via Token-wise Representation Editing and Residual Fusion
DRIP aims to achieve a new security-utility balance by “de-instructionalization”, i.e., precisely removing the instruction semantics from tokens in the data section while preserving their data semantics and robustly maintain the effectiveness of the intended instruction, even against strong adversarial content.
Zhiyong Huang, Jin Dong
· 0 citations