Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DP-MGTD: Privacy-Preserving Machine-Generated Text Detection via Adaptive Differentially Private Entity Sanitization

Lionel Z. Wang, Yusheng Zhao, Jiabin Luo, Xinfeng Li, Lixu Wang, Yinan Peng, Haoyang Li, XiaoFeng Wang, Wei Dong

TL;DR

DP-MGTD introduces an adaptive differential privacy framework that sanitizes sensitive entities in text before machine-generated text detection, balancing privacy with detection utility. The two-stage approach first estimates entity frequencies and then allocates privacy budgets across numerical and textual data using Laplace and Exponential mechanisms, respectively, yielding a DP-compliant sanitized input. Surprisingly, DP-induced perturbations reveal discriminative stability patterns that dramatically improve both metric-based and model-based detectors, achieving near-perfect accuracy on MGTBench-2.0 across domains and architectures. This work demonstrates that privacy mechanisms can transform constraints into advantageous signals for authorship verification, with practical implications for secure, privacy-preserving MGT detection in real-world deployments.

Abstract

The deployment of Machine-Generated Text (MGT) detection systems necessitates processing sensitive user data, creating a fundamental conflict between authorship verification and privacy preservation. Standard anonymization techniques often disrupt linguistic fluency, while rigorous Differential Privacy (DP) mechanisms typically degrade the statistical signals required for accurate detection. To resolve this dilemma, we propose \textbf{DP-MGTD}, a framework incorporating an Adaptive Differentially Private Entity Sanitization algorithm. Our approach utilizes a two-stage mechanism that performs noisy frequency estimation and dynamically calibrates privacy budgets, applying Laplace and Exponential mechanisms to numerical and textual entities respectively. Crucially, we identify a counter-intuitive phenomenon where the application of DP noise amplifies the distinguishability between human and machine text by exposing distinct sensitivity patterns to perturbation. Extensive experiments on the MGTBench-2.0 dataset show that our method achieves near-perfect detection accuracy, significantly outperforming non-private baselines while satisfying strict privacy guarantees.

DP-MGTD: Privacy-Preserving Machine-Generated Text Detection via Adaptive Differentially Private Entity Sanitization

TL;DR

DP-MGTD introduces an adaptive differential privacy framework that sanitizes sensitive entities in text before machine-generated text detection, balancing privacy with detection utility. The two-stage approach first estimates entity frequencies and then allocates privacy budgets across numerical and textual data using Laplace and Exponential mechanisms, respectively, yielding a DP-compliant sanitized input. Surprisingly, DP-induced perturbations reveal discriminative stability patterns that dramatically improve both metric-based and model-based detectors, achieving near-perfect accuracy on MGTBench-2.0 across domains and architectures. This work demonstrates that privacy mechanisms can transform constraints into advantageous signals for authorship verification, with practical implications for secure, privacy-preserving MGT detection in real-world deployments.

Abstract

The deployment of Machine-Generated Text (MGT) detection systems necessitates processing sensitive user data, creating a fundamental conflict between authorship verification and privacy preservation. Standard anonymization techniques often disrupt linguistic fluency, while rigorous Differential Privacy (DP) mechanisms typically degrade the statistical signals required for accurate detection. To resolve this dilemma, we propose \textbf{DP-MGTD}, a framework incorporating an Adaptive Differentially Private Entity Sanitization algorithm. Our approach utilizes a two-stage mechanism that performs noisy frequency estimation and dynamically calibrates privacy budgets, applying Laplace and Exponential mechanisms to numerical and textual entities respectively. Crucially, we identify a counter-intuitive phenomenon where the application of DP noise amplifies the distinguishability between human and machine text by exposing distinct sensitivity patterns to perturbation. Extensive experiments on the MGTBench-2.0 dataset show that our method achieves near-perfect detection accuracy, significantly outperforming non-private baselines while satisfying strict privacy guarantees.
Paper Structure (45 sections, 3 theorems, 17 equations, 1 figure, 2 tables, 1 algorithm)

This paper contains 45 sections, 3 theorems, 17 equations, 1 figure, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Differential Private Text Sanitization
  4. Machine-Generated Text Detection
  5. Preliminaries and Problem Formulation
  6. Task Definition
  7. Threat Model and Privacy Goal
  8. Privacy Goal.
  9. Differential Privacy Basics.
  10. Methodology: DP-MGTD Framework
  11. Overview
  12. Adaptive Differentially Private Entity Sanitization
  13. Stage 1: Noisy Frequency Estimation.
  14. Stage 2: Adaptive Budget Allocation.
  15. Sanitization Execution.
  16. ...and 30 more sections

Key Result

Lemma 3.1

Given a function $f:\mathcal{X}\rightarrow \mathbb{R}^k$, the Laplace Mechanism $\mathcal{M}_L$ is defined as: where $\mathbf{\eta} \sim \text{Lap}(\Delta/\epsilon)^k$. This mechanism satisfies $\epsilon$-DP.

Figures (1)

  • Figure 1: Overview of DP-MGTD. The pipeline transforms unprotected input containing sensitive entities into sanitized representations via Adaptive Differentially Private Entity Sanitization. This core module operates in two stages: (1) Noisy Frequency Estimation to gauge entity density, and (2) Adaptive Budget Allocation to dynamically distribute the privacy budget $\epsilon_{total}$ across text and numerical data using Exponential and Laplace mechanisms, respectively. The sanitized output serves as the input for downstream Metric-based and Model-based detection strategies, enabling robust distinction between machine-generated and human-written text while preserving privacy.

Theorems & Definitions (5)

  • Definition 3.1: Differential Privacy
  • Lemma 3.1: Laplace Mechanism
  • Lemma 3.2: Exponential Mechanism
  • Theorem 3.3: Sequential Composition
  • proof