Saved in:
Bibliographic Details
Main Authors: Wu, Junxi, Huang, Kailin, Hu, Dongjian, Chen, Bin, Wu, Hao, Xia, Shu-Tao, Zou, Changliang
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.16923
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910143411650560
author Wu, Junxi
Huang, Kailin
Hu, Dongjian
Chen, Bin
Wu, Hao
Xia, Shu-Tao
Zou, Changliang
author_facet Wu, Junxi
Huang, Kailin
Hu, Dongjian
Chen, Bin
Wu, Hao
Xia, Shu-Tao
Zou, Changliang
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2604_16923
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Alignment Imprint: Zero-Shot AI-Generated Text Detection via Provable Preference Discrepancy
Wu, Junxi
Huang, Kailin
Hu, Dongjian
Chen, Bin
Wu, Hao
Xia, Shu-Tao
Zou, Changliang
Artificial Intelligence
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.
title Alignment Imprint: Zero-Shot AI-Generated Text Detection via Provable Preference Discrepancy
topic Artificial Intelligence
url https://arxiv.org/abs/2604.16923