Online LLM watermark detection via e-processes

2026-02-15

AI Summary

This study addresses the problem of online watermark detection in text generated by large language models, aiming to effectively distinguish AI-generated content from human-written text. The authors formulate this task as a sequential independence hypothesis test and propose a unified online detection framework based on e-processes, which allows for optional stopping at any time while maintaining statistical validity. They innovatively design an adaptive empirical e-process construction method that enhances detection power without compromising theoretical guarantees. Experimental results demonstrate that the proposed approach matches or exceeds the performance of existing methods in detection accuracy, while providing rigorous anytime-valid statistical error control.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-in-the-loop Machine Learning

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Watermarking for large language models (LLMs) has emerged as an effective tool for distinguishing AI-generated text from human-written content. Statistically, watermark schemes induce dependence between generated tokens and a pseudo-random sequence, reducing watermark detection to a hypothesis testing problem on independence. We develop a unified framework for LLM watermark detection based on e-processes, providing anytime-valid guarantees for online testing. We propose various methods to construct empirically adaptive e-processes that can enhance the detection power. In addition, theoretical results are established to characterize the power properties of the proposed procedures. Some experiments demonstrate that the proposed framework achieves competitive performance compared to existing watermark detection methods.

Problem

• LLM watermark detection
• online testing
• AI-generated text
• hypothesis testing
• e-processes

Innovation

• e-processes
• online watermark detection
• large language models
• anytime-valid inference
• statistical hypothesis testing