Authorship Attribution in the Era of LLMs: Problems, Methodologies, and Challenges

TL;DR

This survey maps four core AA challenges in the LLM era: human-authored text attribution, LLM-generated text detection, LLM-generated text attribution, and human-LLM co-authored text attribution. It traces a progression from stylometric features to ML classifiers, pre-trained LMs, and end-to-end LLM-based approaches, while detailing detectors, watermarking, and attribution techniques. The paper inventories datasets, benchmarks, and evaluation metrics, and highlights open issues such as cross-domain generalization, explainability, and ethical concerns. It then outlines future directions, including finer attribution granularity, standardized benchmarks, adversarial robustness, and interdisciplinary collaboration to advance robust, trustworthy attribution in practice.

Abstract

Accurate attribution of authorship is crucial for maintaining the integrity of digital content, improving forensic investigations, and mitigating the risks of misinformation and plagiarism. Addressing the imperative need for proper authorship attribution is essential to uphold the credibility and accountability of authentic authorship. The rapid advancements of Large Language Models (LLMs) have blurred the lines between human and machine authorship, posing significant challenges for traditional methods. We presents a comprehensive literature review that examines the latest research on authorship attribution in the era of LLMs. This survey systematically explores the landscape of this field by categorizing four representative problems: (1) Human-written Text Attribution; (2) LLM-generated Text Detection; (3) LLM-generated Text Attribution; and (4) Human-LLM Co-authored Text Attribution. We also discuss the challenges related to ensuring the generalization and explainability of authorship attribution methods. Generalization requires the ability to generalize across various domains, while explainability emphasizes providing transparent and understandable insights into the decisions made by these models. By evaluating the strengths and limitations of existing methods and benchmarks, we identify key open problems and future research directions in this field. This literature review serves a roadmap for researchers and practitioners interested in understanding the state of the art in this rapidly evolving field. Additional resources and a curated list of papers are available and regularly updated at https://llm-authorship.github.io

Figures (1)

• Figure 1: Four representative Problems in Authorship Attribution: (1) Human-written Text Attribution, which involves attributing an unknown text to its human authors; (2) LLM-generated Text Detection, which focuses on detecting whether a text has been generated by LLMs; (3) LLM-generated Text Attribution, aimed at identifying the specific LLM or human responsible for a given text; (4) Human-LLM Co-authored Text Attribution, which classifies a text as human-written, LLM-generated, or a combination of both. These problems become progressively more complex, as indicated by the arrows.