A Survey on Privacy Risks and Protection in Large Language Models

2025-05-04

AI Summary

This paper systematically uncovers privacy leakage mechanisms of large language models (LLMs) in training data extraction, membership inference, and model inversion. It introduces the first holistic analytical framework integrating multi-dimensional attack paradigms with cross-layer defense strategies. Methodologically, it unifies federated learning, confidential computing, backdoor mitigation, and inference-time detection to establish a structured risk taxonomy and an evaluation benchmark for defense efficacy. Key contributions are: (1) the first three-dimensional analytical model spanning risk mechanisms, defense techniques, and governance pathways; (2) identification of three emerging research directions—privacy risk assessment, secure knowledge transfer, and privacy-governance intersection; and (3) theoretical foundations and actionable guidelines for trustworthy deployment and regulatory compliance of LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Although Large Language Models (LLMs) have become increasingly integral to diverse applications, their capabilities raise significant privacy concerns. This survey offers a comprehensive overview of privacy risks associated with LLMs and examines current solutions to mitigate these challenges. First, we analyze privacy leakage and attacks in LLMs, focusing on how these models unintentionally expose sensitive information through techniques such as model inversion, training data extraction, and membership inference. We investigate the mechanisms of privacy leakage, including the unauthorized extraction of training data and the potential exploitation of these vulnerabilities by malicious actors. Next, we review existing privacy protection against such risks, such as inference detection, federated learning, backdoor mitigation, and confidential computing, and assess their effectiveness in preventing privacy leakage. Furthermore, we highlight key practical challenges and propose future research directions to develop secure and privacy-preserving LLMs, emphasizing privacy risk assessment, secure knowledge transfer between models, and interdisciplinary frameworks for privacy governance. Ultimately, this survey aims to establish a roadmap for addressing escalating privacy challenges in the LLMs domain.

Problem

• Analyzing privacy leakage risks in LLMs from attacks like data extraction.
• Reviewing protection methods such as federated learning and confidential computing.
• Proposing future research directions for secure, privacy-preserving LLMs.

Innovation

• Analyzing privacy leakage and attacks in LLMs
• Reviewing privacy protection techniques like federated learning
• Proposing future research for secure privacy-preserving LLMs