Triospect: A Three-Dimensional Framework for Robust Statistical AI-Generated Text Detection Against Diverse Attacks

2026-06-29

AI Summary

Current AI-generated text detectors suffer from insufficient robustness against adversarial attacks. This work proposes Triospect, a novel three-dimensional detection framework that, for the first time, jointly models content semantics, expressive style, and statistical features to establish a multidimensional, fusion-based paradigm for robust detection. Evaluated on the Humanize-16K and RAID datasets, the proposed method significantly outperforms strong baseline models, achieving absolute improvements of 22.3% and 9.1% in AUROC, and 13% and 22% in TPR@FPR=0.1, respectively. These gains demonstrate its enhanced resilience against a wide range of adversarial attacks.

Abstract

Existing AI-generated text detectors are vulnerable to attacks that manipulate textual characteristics. In this study, we propose a novel Triospect Detection Framework by using additional perspectives of content (core ideas) and expression (stylistic elements) within a given text. Experiments on two benchmarks involving 17 attacks, 12 domains, and 17 source models demonstrate that Triospect is robust against these attacks. It improves the strong baseline by a significant margin of 22.3% (AUROC) and 13% (TPR01) on the Humanize-16K after-attack subset, and by 9.1% (AUROC) and 22% (TPR01) on the adversarial RAID. This framework marks a pioneering effort in statistical methods to enhance detection reliability against attacks. We release our data and code at https://github.com/baoguangsheng/triospect.

Problem

• AI-generated text detection
• adversarial attacks
• textual robustness
• statistical detection
• detector vulnerability

Innovation

• Triospect
• AI-generated text detection
• robustness
• statistical framework
• adversarial attacks