Latent Trajectory Discrimination for AI-Generated Text Detection

2026-07-16

AI Summary

This work addresses a key limitation in existing AI-generated text detection methods, which treat documents as static entities and overlook the dynamic evolution of latent representations during autoregressive generation. To overcome this, the paper reframes detection as the task of distinguishing latent generation trajectories, introducing a novel dynamic perspective that captures the geometric regularities underlying text generation. The authors propose the Geometric Trajectory and Contrastive Learning (GTCL) framework, which segments documents into ordered local units, constructs sequence-level trajectory representations in embedding space, and leverages contrastive learning to extract discriminative features. Experimental results demonstrate that GTCL significantly outperforms current state-of-the-art methods across three benchmark datasets, confirming that modeling generation dynamics yields robust, cross-model, and cross-domain detection signals.

Abstract

Most existing approaches to AI-Generated Text Detection (AIGTD) treat documents as static objects and base their decisions on aggregate statistics or globally compressed embeddings. However, this perspective overlooks the inherently dynamic nature of autoregressive generation, where content evolves progressively through the latent space. In this paper, we reformulate AIGTD as the problem of distinguishing between latent generation trajectories. Instead of relying on static representations, we model how textual representations evolve across the sequence. To this end, we propose Geometric Trajectory and Contrastive Learning (GTCL), a framework that segments the document into ordered local units, encodes each unit in an embedding space, and constructs a structured and sequence-level representation. GTCL then applies contrastive learning to these trajectories to learn geometric regularities associated with the autoregressive generation. Evaluations performed on three different benchmarks and several approaches show that GTCL outperforms detection baselines consistently, which implies that explicitly modeling sequential dynamics provides robust discriminative signals across models and domains. These results suggest that modeling trajectory differences could improve detection and open up a dynamic direction that has been underexplored in previous AIGTD literature.

Problem

• AI-Generated Text Detection
• latent trajectory
• autoregressive generation
• sequential dynamics
• text detection

Innovation

• latent trajectory
• autoregressive generation
• contrastive learning
• geometric regularity
• dynamic representation