Permit: Permission-Aware Representation Intervention for Controlled Generation in Large Language Models

Pengcheng Sun, Lan Zhang, Zhaopeng Zhang +2

Large language models (LLMs) are increasingly deployed in enterprise settings where they handle sensitive documents and user context, raising acute concerns over security and contr…