Exons-Detect: Identifying and Amplifying Exonic Tokens via Hidden-State Discrepancy for Robust AI-Generated Text Detection

TL;DR

Exons-Detect is a training-free, exon-aware method that enhances AI-generated text detection by focusing on token importance through hidden-state discrepancies, achieving state-of-the-art robustness and accuracy.

Contribution

It introduces a novel exon-aware token reweighting approach based on hidden-state discrepancies, improving detection robustness without training.

Findings

Achieves state-of-the-art detection performance.

Demonstrates strong robustness to adversarial attacks.

Improves AUROC by 2.2% over prior baselines.

Abstract

The rapid advancement of large language models has increasingly blurred the boundary between human-written and AI-generated text, raising societal risks such as misinformation dissemination, authorship ambiguity, and threats to intellectual property rights. These concerns highlight the urgent need for effective and reliable detection methods. While existing training-free approaches often achieve strong performance by aggregating token-level signals into a global score, they typically assume uniform token contributions, making them less robust under short sequences or localized token modifications. To address these limitations, we propose Exons-Detect, a training-free method for AI-generated text detection based on an exon-aware token reweighting perspective. Exons-Detect identifies and amplifies informative exonic tokens by measuring hidden-state discrepancy under a dual-model setting, and computes an interpretable translation score from the resulting importance-weighted token sequence. Empirical evaluations demonstrate that Exons-Detect achieves state-of-the-art detection performance and exhibits strong robustness to adversarial attacks and varying input lengths. In particular, it attains a 2.2\% relative improvement in average AUROC over the strongest prior baseline on DetectRL.