Efficient Online LLM Watermark Detection via Rao-Blackwellized E-Processes
It addresses the need for reliable, real-time detection of AI-generated text in streaming settings, which is crucial for applications like content moderation and plagiarism detection.
The paper develops an online watermark detection framework for LLMs using Rao-Blackwellized e-processes, enabling anytime-valid inference with recursive token-level updates. The method achieves rigorous Type I error control under optional stopping and demonstrates efficient detection in simulations and real text experiments.
As large language models (LLMs) are increasingly deployed, reliable and efficient mechanisms for distinguishing AI-generated text from human-written content have become essential. Statistical watermarking has emerged as a promising solution, yet most existing methods are typically fixed-horizon procedures, precluding valid early stopping in streaming generation. In this paper, we develop an efficient online watermark detection framework with anytime-valid inference based on Rao-Blackwellized e-processes, enabling recursive token-level evidence updates without storing the full history. In particular, we instantiate the framework for the Gumbel-max watermark and reduce the original token-level dependence testing problem to a pivot-induced sequential testing problem with an explicit null distribution. Theoretically, we prove anytime-valid Type I error control under arbitrary optional stopping and establish positive asymptotic log-growth under watermarking, implying consistency of the proposed stopping rules. Simulations and experiments on real LLM-generated text demonstrate efficient online detection with rigorous anytime-valid guarantees.