Breaking the Generator Barrier: Disentangled Representation for Generalizable AI-Text Detection

2026-04-15

AI Summary

This work addresses the limited generalizability of existing AI-generated text detectors, which often rely on artifacts specific to particular language models and fail to generalize to unseen ones. To overcome this, the authors propose a progressive disentanglement framework that, for the first time, integrates semantic-minimal latent encoding, perturbation regularization, and discriminative representation alignment. This approach effectively decouples semantic content from generator-specific artifacts, substantially enhancing cross-model generalization. Evaluated on the MAGE benchmark encompassing 20 large language models, the method achieves up to a 24.2% improvement in accuracy and a 26.2% gain in F1 score over prior approaches. Moreover, its performance consistently improves as the diversity of training generators increases, demonstrating robust scalability and adaptability.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

As large language models (LLMs) generate text that increasingly resembles human writing, the subtle cues that distinguish AI-generated content from human-written content become increasingly challenging to capture. Reliance on generator-specific artifacts is inherently unstable, since new models emerge rapidly and reduce the robustness of such shortcuts. This generalizes unseen generators as a central and challenging problem for AI-text detection. To tackle this challenge, we propose a progressively structured framework that disentangles AI-detection semantics from generator-aware artifacts. This is achieved through a compact latent encoding that encourages semantic minimality, followed by perturbation-based regularization to reduce residual entanglement, and finally a discriminative adaptation stage that aligns representations with task objectives. Experiments on MAGE benchmark, covering 20 representative LLMs across 7 categories, demonstrate consistent improvements over state-of-the-art methods, achieving up to 24.2% accuracy gain and 26.2% F1 improvement. Notably, performance continues to improve as the diversity of training generators increases, confirming strong scalability and generalization in open-set scenarios. Our source code will be publicly available at https://github.com/PuXiao06/DRGD.

Problem

• AI-text detection
• generalizable detection
• unseen generators
• disentangled representation
• large language models

Innovation

• disentangled representation
• generalizable AI-text detection
• generator-agnostic
• perturbation-based regularization
• latent encoding