Ward: Provable RAG Dataset Inference via LLM Watermarks

Nikola Jovanović, Robin Staab, Maximilian Baader, Martin Vechev

Keywords: llm, watermarks, dataset inference, rag

TL;DR: We formalize RAG Dataset Inference, introduce a suitable dataset and baselines, and propose Ward, a rigorous method based on LLM watermarks.

Abstract: RAG enables LLMs to easily incorporate external data, raising concerns for data owners regarding unauthorized usage of their content. The challenge of detecting such unauthorized usage remains underexplored, with datasets and methods from adjacent fields being ill-suited for its study. We take several steps to bridge this gap. First, we formalize this problem as (black-box) RAG Dataset Inference (RAG-DI). We then introduce a novel dataset designed for realistic benchmarking of RAG-DI methods, alongside a set of baselines. Finally, we propose Ward, a method for RAG-DI based on LLM watermarks that equips data owners with rigorous statistical guarantees regarding their dataset's misuse in RAG corpora. Ward consistently outperforms all baselines, achieving higher accuracy, superior query efficiency and robustness. Our work provides a foundation for future studies of RAG-DI and highlights LLM watermarks as a promising approach to this problem.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 13947