Lost in Overlap: Exploring Logit-based Watermark Collision in LLMs

TL;DR

This paper investigates the issue of watermark collision in logit-based watermarking methods for LLMs, revealing its threat to copyright protection and proposing it as a new attack strategy.

Contribution

It introduces watermark collision as a novel attack concept that undermines logit-based watermarking across various LLM tasks and applications.

Findings

Watermark collision significantly reduces watermark detection accuracy.

It poses a universal threat to all logit-based watermark algorithms.

Watermark collision impacts downstream applications and content integrity.

Abstract

The proliferation of large language models (LLMs) in generating content raises concerns about text copyright. Watermarking methods, particularly logit-based approaches, embed imperceptible identifiers into text to address these challenges. However, the widespread usage of watermarking across diverse LLMs has led to an inevitable issue known as watermark collision during common tasks, such as paraphrasing or translation. In this paper, we introduce watermark collision as a novel and general philosophy for watermark attacks, aimed at enhancing attack performance on top of any other attacking methods. We also provide a comprehensive demonstration that watermark collision poses a threat to all logit-based watermark algorithms, impacting not only specific attack scenarios but also downstream applications.