Alignment Imprint: Zero-Shot AI-Generated Text Detection via Provable Preference Discrepancy

TL;DR

This paper introduces the Alignment Imprint and LAPD, a new method for detecting AI-generated text by leveraging distributional signatures left by model alignment, achieving significant performance improvements.

Contribution

It provides a theoretical framework for understanding alignment imprint and proposes LAPD, a new detection statistic that outperforms existing methods.

Findings

LAPD achieves 45.82% relative improvement over baselines.

Alignment imprint captures distributional effects of model alignment.

Theoretically, LAPD improves detection when models are close in distribution.

Abstract

Detecting AI-generated text is an important but challenging problem. Existing likelihood-based detection methods are often sensitive to content complexity and may exhibit unstable performance. In this paper, our key insight is that modern Large Language Models (LLMs) undergo alignment (including fine-tuning and preference tuning), leaving a measurable distributional imprint. We theoretically derive this imprint by abstracting the alignment process as a sequence of constrained optimization steps, showing that the log-likelihood ratio can naturally decompose into implicit instructional biases and preference rewards. We refer to this quantity as the Alignment Imprint. Furthermore, to mitigate the instability in high-entropy regions, we introduce Log-likelihood Alignment Preference Discrepancy (LAPD), a standardized information-weighted statistic based on alignment imprint. We provide statistical guarantee that alignment-based statistics dominate Fast-DetectGPT in performance. We also theoretically show that LAPD strictly improves the unweighted alignment scores when the aligned and base models are close in distribution. Extensive experiments show that LAPD achieves an improvement 45.82% relative to the strongest existing baselines, yielding large and consistent gains across all settings.