Large Language Models are Advanced Anonymizers

2024-02-21

arXiv.org

AI Summary

Existing anonymization methods fail to withstand strong inference attacks by large language models (LLMs) on online text, leading to severe privacy breaches. To address this, we propose the first LLM-aware anonymization paradigm specifically designed to counteract LLM-driven inference. Our method introduces an LLM-powered adversarial anonymization framework: it explicitly models the attacker’s capabilities using LLMs and jointly optimizes controllable text generation with semantic preservation to achieve co-enhancement of privacy protection and textual utility. Furthermore, we design the first benchmark tailored to evaluate anonymization robustness against LLM-based attackers. Extensive experiments on both real-world and synthetic datasets demonstrate that our approach significantly outperforms state-of-the-art industrial anonymization tools in two critical dimensions—resistance to LLM inference (privacy) and performance on downstream NLP tasks (utility)—thereby establishing the first robust anonymization solution effective against LLM-level adversaries.

Technology Category

Machine Learning: PrivacyNatural Language Processing: Safety and RobustnessComputer Vision: Adversarial Attacks & Robustness

Application Category

User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSearch and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved informationSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recent work in privacy research on large language models has shown that they achieve near human-level performance at inferring personal data from real-world online texts. With consistently increasing model capabilities, existing text anonymization methods are currently lacking behind regulatory requirements and adversarial threats. This raises the question of how individuals can effectively protect their personal data in sharing online texts. In this work, we take two steps to answer this question: We first present a new setting for evaluating anonymizations in the face of adversarial LLMs inferences, allowing for a natural measurement of anonymization performance while remedying some of the shortcomings of previous metrics. We then present our LLM-based adversarial anonymization framework leveraging the strong inferential capabilities of LLMs to inform our anonymization procedure. In our experimental evaluation, we show on real-world and synthetic online texts how adversarial anonymization outperforms current industry-grade anonymizers both in terms of the resulting utility and privacy.

Problem

• Privacy Protection
• Language Models
• Anonymization Methods

Innovation

• Anonymization
• Advanced Language Models
• Privacy Enhancement