Who Wrote the Book? Detecting and Attributing LLM Ghostwriters

TL;DR

This paper introduces GhostWriteBench, a large dataset for LLM authorship attribution, and proposes TRACE, an interpretable fingerprinting method that outperforms existing techniques, especially in out-of-distribution scenarios.

Contribution

The paper presents GhostWriteBench dataset and TRACE, a novel lightweight fingerprinting method for LLM attribution that is effective across various models and conditions.

Findings

TRACE achieves state-of-the-art performance.

TRACE remains robust in out-of-distribution settings.

Works well with limited training data.

Abstract

In this paper, we introduce GhostWriteBench, a dataset for LLM authorship attribution. It comprises long-form texts (50K+ words per book) generated by frontier LLMs, and is designed to test generalisation across multiple out-of-distribution (OOD) dimensions, including domain and unseen LLM author. We also propose TRACE -- a novel fingerprinting method that is interpretable and lightweight -- that works for both open- and closed-source models. TRACE creates the fingerprint by capturing token-level transition patterns (e.g., word rank) estimated by another lightweight language model. Experiments on GhostWriteBench demonstrate that TRACE achieves state-of-the-art performance, remains robust in OOD settings, and works well in limited training data scenarios.