Latent Trajectory Discrimination for AI-Generated Text Detection

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Gianluca Bonifazi, Christopher Buratti, Michele Marchetti, Federica Parlapiano, Giulia Quaglieri, Davide Traini, Domenico Ursino, Luca Virgili
• Venue: arXiv: cs.CL
• Published: 2026-07-16
• Collected: 2026-07-17
• URL: http://arxiv.org/abs/2607.14967v1
• PDF: https://arxiv.org/pdf/2607.14967v1
• DOI: UNKNOWN

Plain-Language Brief

The paper presents a new approach for detecting AI-generated text called Geometric Trajectory and Contrastive Learning (GTCL). Instead of treating documents as static objects, GTCL models how textual representations evolve across the sequence to distinguish between latent generation trajectories. The method outperforms existing detection baselines on three different benchmarks.

What the Document Says

• Most existing methods for AI-generated text detection (AIGTD) consider documents as static objects, using aggregate statistics or globally compressed embeddings.
• This paper proposes a new approach called Geometric Trajectory and Contrastive Learning (GTCL), which treats AIGTD as the problem of distinguishing between latent generation trajectories.
• GTCL models how textual representations evolve across a sequence by segmenting documents into ordered local units, encoding each unit in an embedding space, and constructing a structured and sequence-level representation.
• The framework applies contrastive learning to these trajectories to learn geometric regularities associated with the autoregressive generation.
• Evaluations on three different benchmarks show that GTCL outperforms detection baselines consistently across models and domains.
• These results imply that explicitly modeling sequential dynamics provides robust discriminative signals for AIGTD.

Key Concepts Mentioned

• AI-generated text detection
• latent space
• autoregressive generation
• Geometric Trajectory and Contrastive Learning (GTCL)

Claims Made by the Document

• Existing methods for AIGTD overlook the dynamic nature of autoregressive generation.
• GTCL outperforms detection baselines consistently across models and domains.

Evidence Used

• The paper discusses existing approaches to AIGTD that treat documents as static objects. It then presents GTCL and its results.

Methods / Approach

• Geometric Trajectory and Contrastive Learning (GTCL)

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

The study's performance is evaluated on three specific benchmarks. It's unclear how generalizable the results are to other datasets or scenarios.

Open Questions

• What are the potential implications of this research for improving AI-generated text detection beyond the specific methods and benchmarks used in this study?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The paper presents a new approach for detecting AI-generated text. It focuses on the dynamic nature of autoregressive generation and proposes a framework called GTCL that models how textual representations evolve across a sequence. The results show that this approach outperforms existing detection baselines.

Raw Abstract / Excerpt

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.