Learning to Refuse: Towards Mitigating Privacy Risks in LLMs

TL;DR

This paper introduces a new dataset and framework for enabling large language models to unlearn specific private data, effectively balancing privacy protection with model utility.

Contribution

It presents the Name-Aware Unlearning Framework (NAUF) and a real-world dataset for evaluating machine unlearning in privacy-sensitive scenarios.

Findings

NAUF achieves a 5.65 point improvement in unlearning score over baselines.

The framework effectively protects individual privacy without degrading unrelated question answering.

The dataset enables realistic evaluation of unlearning methods in practical settings.

Abstract

Large language models (LLMs) exhibit remarkable capabilities in understanding and generating natural language. However, these models can inadvertently memorize private information, posing significant privacy risks. This study addresses the challenge of enabling LLMs to protect specific individuals' private data without the need for complete retraining. We propose \return, a Real-world pErsonal daTa UnleaRNing dataset, comprising 2,492 individuals from Wikipedia with associated QA pairs, to evaluate machine unlearning (MU) methods for protecting personal data in a realistic scenario. Additionally, we introduce the Name-Aware Unlearning Framework (NAUF) for Privacy Protection, which enables the model to learn which individuals' information should be protected without affecting its ability to answer questions related to other unrelated individuals. Our extensive experiments demonstrate that NAUF achieves a state-of-the-art average unlearning score, surpassing the best baseline method by 5.65 points, effectively protecting target individuals' personal data while maintaining the model's general capabilities.