LLM Watermarking Using Mixtures and Statistical-to-Computational Gaps

Pedro Abdalla, Roman Vershynin
5/2/2025
cs.CRcs.LG

No confident code implementation found for this paper.

Abstract

Given a text, can we determine whether it was generated by a large language model (LLM) or by a human? A widely studied approach to this problem is watermarking. We propose an undetectable and elementary watermarking scheme in the closed setting. Also, in the harder open setting, where the adversary has access to most of the model, we propose an unremovable watermarking scheme.

View on arXivView PDF

Code Implementations

No confident code match yet

We couldn't find an author-owned or strongly-evidenced community implementation for this paper. Any repos shown below are weak matches — verify before relying on them.

No code implementations found yet.

Know of an implementation? Let us know in the comments below!

Similar Papers

Learnable Linguistic Watermarks for Tracing Model Extraction Attacks on Large Language Models

89% match2024

Can Watermarked LLMs be Identified by Users via Crafted Prompts?

88% match1 repo282026

Yet Another Watermark for Large Language Models

88% match2025

Theoretically Grounded Framework for LLM Watermarking: A Distribution-Adaptive Approach

88% match1 repo72024

Breaking Distortion-free Watermarks in Large Language Models

88% match2025

On the Reliability of Watermarks for Large Language Models

88% match1 repo6792026

Cite this paper

@article{abdalla2025watermarking,
  title  = {LLM Watermarking Using Mixtures and Statistical-to-Computational Gaps},
  author = {Pedro Abdalla and Roman Vershynin},
  year   = {2025},
  eprint = {2505.01484},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2505.01484v2}
}

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/2505.01484.svg)](https://codeofpaper.com/paper/2505.01484)

Discussion