SynBench: A Benchmark for Differentially Private Text Generation

2025-09-17

AI Summary

In high-stakes domains such as healthcare and finance, sharing sensitive textual data is hindered by privacy leakage risks and the absence of a trusted, standardized benchmark for evaluating differentially private (DP) text generation. Method: We propose the first comprehensive evaluation framework for DP text synthesis, encompassing nine domain-specific datasets; introduce a novel membership inference attack tailored to synthetic text to quantify privacy leakage; and design a dual-dimensional metric jointly assessing utility (e.g., domain fidelity, semantic coherence) and privacy (ε-DP guarantees). Results: Experiments reveal that existing DP text generation methods suffer substantial utility degradation in complex domains and exhibit severe privacy–utility trade-off imbalances, underscoring the necessity of domain-aware evaluation. This work establishes a standardized benchmark, introduces a new attack paradigm, and provides foundational empirical insights for advancing DP text generation.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Privacy
• Application Domains: Humanities & Computational Social Science

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems

Abstract

Data-driven decision support in high-stakes domains like healthcare and finance faces significant barriers to data sharing due to regulatory, institutional, and privacy concerns. While recent generative AI models, such as large language models, have shown impressive performance in open-domain tasks, their adoption in sensitive environments remains limited by unpredictable behaviors and insufficient privacy-preserving datasets for benchmarking. Existing anonymization methods are often inadequate, especially for unstructured text, as redaction and masking can still allow re-identification. Differential Privacy (DP) offers a principled alternative, enabling the generation of synthetic data with formal privacy assurances. In this work, we address these challenges through three key contributions. First, we introduce a comprehensive evaluation framework with standardized utility and fidelity metrics, encompassing nine curated datasets that capture domain-specific complexities such as technical jargon, long-context dependencies, and specialized document structures. Second, we conduct a large-scale empirical study benchmarking state-of-the-art DP text generation methods and LLMs of varying sizes and different fine-tuning strategies, revealing that high-quality domain-specific synthetic data generation under DP constraints remains an unsolved challenge, with performance degrading as domain complexity increases. Third, we develop a membership inference attack (MIA) methodology tailored for synthetic text, providing first empirical evidence that the use of public datasets - potentially present in pre-training corpora - can invalidate claimed privacy guarantees. Our findings underscore the urgent need for rigorous privacy auditing and highlight persistent gaps between open-domain and specialist evaluations, informing responsible deployment of generative AI in privacy-sensitive, high-stakes settings.

Problem

• Evaluating differential privacy methods for synthetic text generation
• Benchmarking privacy-utility tradeoffs in domain-specific datasets
• Assessing privacy risks from pretraining data contamination

Innovation

• Differential Privacy for synthetic text generation
• Comprehensive evaluation framework with standardized metrics
• Membership inference attack methodology for synthetic text