Who Wrote the Book? Detecting and Attributing LLM Ghostwriters

Anudeex Shetty, Qiongkai Xu, Olga Ohrimenko, Jey Han Lau
3/30/2026
cs.CL

No official code found. Best implementation: anudeex/TRACE (2 stars, Apache-2.0, last commit 2026-03-31).

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.

View on arXivView PDF

Code Implementations(1)

anudeex/TRACECommunity74%
Apache-2.0

Code for our paper - "Who Wrote the Book? Detecting and Attributing LLM Ghostwriters"

20PythonMar 21, 20266 months agoApache-2.0

Similar Papers

MultiGhostBench: A Multilingual Benchmark for Long-Form LLM-Generated Text Attribution under Distribution Shifts

86% match1 repo12026

ALMs: Authorial Language Models for Authorship Attribution

86% match2024

Forging LLM Authorship Fingerprints with Targeted Rewriting

86% match1 repo2026

Neural Authorship Attribution: Stylometric Analysis on Large Language Models

85% match2023

A Bayesian Approach to Harnessing the Power of LLMs in Authorship Attribution

84% match2024

Rank-Turbulence Delta and Interpretable Approaches to Stylometric Delta Metrics

84% match2026

Cite this paper

@article{shetty2026wrote,
  title  = {Who Wrote the Book? Detecting and Attributing LLM Ghostwriters},
  author = {Anudeex Shetty and Qiongkai Xu and Olga Ohrimenko and Jey Han Lau},
  year   = {2026},
  eprint = {2603.28054},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2603.28054v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2603.28054.svg)](https://codeofpaper.com/paper/2603.28054)

Discussion