New method improves efficiency of locating AI-generated text watermarks

Query-efficient scan-and-search watermark localization

Cryptography and SecurityArtificial Intelligence

Summary

Detecting whether parts of a text come from AI is important for accountability. The authors found that existing methods to locate hidden AI watermarks in text are slow because they check every word closely. They designed a smarter way to spend less effort by first scanning broadly and then focusing only on suspicious areas. Their method, called BELL, can accurately find watermarked sections using fewer checks and controls the rate of false alarms. Tests showed BELL works better than previous methods across different AI text models and datasets.

What this means in practice

Authors

Liyuan Wang, Xinsong Ma, Ziye Ma

Abstract

Recently, the detection of AI-generated content has been of particular importance in various fields of our society. For accountability purposes, many large language models (LLMs) have implicitly watermarked their outputs. Localizing such watermarks requires a piece of text to undergo an additional verification process. Existing localization methods typically rely on dense token-level scoring, making verification costly even when an LLM-generated watermark only sparsely appears. To tackle this problem, we design a new adaptive allocation strategy that first maintains broad document coverage and then progressively concentrates the detection budget on regions that exhibit stronger watermark evidence or are more likely to contain watermarked content. We then transform the statistical signals from the selected regions into e-values and employ the proposed token-weighted e-BH procedure to determine the localized watermark regions. The resulting framework is called BELL (Budgeted E-evidence Localization for LLM Watermarks). Theoretically, BELL can control the token-weighted false discovery rate (FDR) without independence assumptions across regions. Finally, extensive experimental results demonstrate that BELL consistently outperforms competing methods across multiple LLMs, datasets, and watermarking schemes.