HATS: High-Accuracy Triple-Set Watermarking for Large Language Models

Zhiqing Hu, Chenxu Zhao, Jiazhong Lu, Xiaolei Liu

Abstract

Misuse of LLM-generated text can be curbed by watermarking techniques that embed implicit signals into the output. We propose a watermark that partitions the vocabulary at each decoding step into three sets (Green/Yellow/Red) with fixed ratios and restricts sampling to the Green and Yellow sets. At detection time, we replay the same partitions, compute Green-enrichment and Red-depletion statistics, convert them to one-sided z-scores, and aggregate their p-values via Fisher's method to decide whether a passage is watermarked. We implement generation, detection, and testing on Llama 2 7B, and evaluate true-positive rate, false-positive rate, and text quality. Results show that the triple-partition scheme achieves high detection accuracy at fixed FPR while preserving readability.
BibTeX
@misc{hu2025hats,
  author = {Zhiqing Hu and Chenxu Zhao and Jiazhong Lu and Xiaolei Liu},
  title = {{HATS: High-Accuracy Triple-Set Watermarking for Large Language Models}},
  year = {2025},
  eprint = {2512.19378},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.19378},
}
Figure