Enhancing Data Privacy in Large Language Models through Private Association Editing

TL;DR

This paper introduces Private Association Editing (PAE), a novel method to prevent private data leakage in large language models by removing PII without retraining, enhancing data privacy in text-generation applications.

Contribution

The paper presents PAE, a new approach that effectively removes PII from LLM outputs without requiring model retraining, addressing privacy concerns in large language models.

Findings

PAE effectively removes PII from LLM outputs.

PAE outperforms baseline methods in privacy preservation.

PAE maintains model utility while enhancing privacy.

Abstract

Large language models (LLMs) require a significant redesign in solutions to preserve privacy in data-intensive applications due to their text-generation capabilities. Indeed, LLMs tend to memorize and emit private information when maliciously prompted. In this paper, we introduce Private Association Editing (PAE) as a novel defense approach for private data leakage. PAE is designed to effectively remove Personally Identifiable Information (PII) without retraining the model. Experimental results demonstrate the effectiveness of PAE with respect to alternative baseline methods. We believe PAE will serve as a critical tool in the ongoing effort to protect data privacy in LLMs, encouraging the development of safer models for real-world applications.