arXiv cs.CL· Jiahao Xu, Rui Hu, Olivera Kotevska, Zikai Zhang·· 3 天前AI 评分26
面向大语言模型的多数位感知水印方法
Majority Bit-Aware Watermarking For Large Language Models
AI 导读
研究者提出多数位感知编码（majority bit-aware encoding）新范式，将水印信号强度与绿名单规模解耦，即使使用大绿名单也能在生成文本中保留强水印信号。基于该范式给出 MajorMark 与面向长消息优化的 MajorMark⁺ 两种实现，在 SOTA LLM 上取得比既有基线更高的解码准确率和更优文本质量。该工作已被 NeurIPS 2026 接收，代码已开源。
正文 · 原文
Abstract:The growing deployment of Large Language Models (LLMs) has raised concerns about their misuse in generating harmful or deceptive content. To address this issue, watermarking methods have been proposed to embed identifiable multi-bit messages into generated text for misuse tracing. However, existing methods often suffer from a fundamental trade-off between text quality and decoding accuracy. In particular, they have to restrict the size of the preferred token set (i.e., green list) during encoding to maintain a detectable watermark signal for decoding, which inevitably degrades generation quality. To improve this trade-off, we propose a novel message encoding paradigm called \textit{majority bit-aware encoding}, which relaxes the watermark signal strength from the green list size. This strategy allows for a strong watermark signal to be preserved in generated texts even when using a large green list. We introduce two instantiations of this paradigm: MajorMark and MajorMark$^{+}$, where the latter is specifically optimized for long messages. Extensive experiments on state-of-the-art LLMs demonstrate that our methods achieve higher decoding accuracy and superior text quality compared to prior baselines. The code is at this https URL.
| Comments: | This work has been accepted for presentation at NeurIPS 2026 |
| Subjects: | Computation and Language (cs.CL); Cryptography and Security (cs.CR) |
| Cite as: | arXiv:2508.03829 [cs.CL] |
| (or arXiv:2508.03829v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2508.03829 arXiv-issued DOI via DataCite |
Submission history
From: Jiahao Xu [view email]
[v1]
Tue, 5 Aug 2025 18:19:00 UTC (712 KB)
[v2]
Fri, 8 May 2026 22:08:12 UTC (900 KB)
[v3]
Tue, 13 Oct 2026 21:51:12 UTC (899 KB)
来源：arXiv cs.CL · arxiv.org