Token-Level Attribution of Multi-Mode LLM Watermarks: Optimal Rule and Belief Weighting
Abstract
A watermark verifier may hold many modes, i.e., secret keys, and a document may mix spans of different modes with unwatermarked text. Directly reporting the mode with the largest pivotal statistic (pivot) inflates the Type I error, and restoring the nominal level reduces power as candidate modes are added. We formulate token-level multi-mode attribution for the Gumbel-max watermark as maximizing correct-label power under a Type I error constraint given a mode prior. We show that the optimal rule reports the mode with the largest prior-weighted alternative density and rejects when this density exceeds one common threshold. Based on this rule, we propose the Belief-weighted Likelihood Attribution with Density Estimation (BLADE) detector. It estimates the mode prior from neighboring pivots and the alternative density from reference token probabilities. Under a conditional product model of the pivots, it keeps the exact Type I error. We further show that BLADE approaches the optimal rule as its prior and density estimates improve, with a sharp coefficient on the density error. On OPT-1.3B and other open-weight large language models, BLADE attains higher correct-label power than the compared detectors across candidate counts, token-substitution rates, generation temperatures and generator models.
Related
- Optimal Weighting of Training Data in Adaptive Coding
- Deep belief networks are exact
- From Type to Token: Petri Nets, Individual Token Philosophy, and Individualized Causal Effects
- Ada-TokenCom: Rate-Adaptive Token Communications via Large-Model-Driven Token Compression and Generation
- Almost Every Graph Is Reconstructible from Its Token Graphs
BibTeX
@misc{zhou2026tokenlevel,
title = {{Token-Level Attribution of Multi-Mode LLM Watermarks: Optimal Rule and Belief Weighting}},
author = {Rui Zhou and Licheng Zhao and Chao Shen and Wenqiang Pu},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.23601},
eprint = {2610.23601},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.23601},
}
RIS
TY - UNPB
TI - Token-Level Attribution of Multi-Mode LLM Watermarks: Optimal Rule and Belief Weighting
AU - Rui Zhou
AU - Licheng Zhao
AU - Chao Shen
AU - Wenqiang Pu
PY - 2026
DA - 2026/10/16
PB - arXiv
DO - 10.48550/arxiv.2610.23601
UR - https://arxiv.org/abs/2610.23601
AB - A watermark verifier may hold many modes, i.e., secret keys, and a document may mix spans of different modes with unwatermarked text. Directly reporting the mode with the largest pivotal statistic (pivot) inflates the Type I error, and restoring the nominal level reduces power as candidate modes are added. We formulate token-level multi-mode attribution for the Gumbel-max watermark as maximizing correct-label power under a Type I error constraint given a mode prior. We show that the optimal rule reports the mode with the largest prior-weighted alternative density and rejects when this density exceeds one common threshold. Based on this rule, we propose the Belief-weighted Likelihood Attribution with Density Estimation (BLADE) detector. It estimates the mode prior from neighboring pivots and the alternative density from reference token probabilities. Under a conditional product model of the pivots, it keeps the exact Type I error. We further show that BLADE approaches the optimal rule as its prior and density estimates improve, with a sharp coefficient on the density error. On OPT-1.3B and other open-weight large language models, BLADE attains higher correct-label power than the compared detectors across candidate counts, token-substitution rates, generation temperatures and generator models.
ER -