TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection
Tom Sander, Hongyan Chang, Tomáš Souček, Tuan Tran, Valeriu Lacatusu + 8 more
TLDR
TextSeal is a new LLM watermark using dual-key generation and multi-region localization for robust, distortion-free detection and distillation protection.
Key contributions
- Introduces TextSeal, a state-of-the-art LLM watermark with dual-key generation for output diversity.
- Features entropy-weighted scoring and multi-region localization for improved, robust detection.
- Zero inference overhead, supports serving optimizations like speculative decoding and multi-token prediction.
- "Radioactive" watermark transfers through distillation, protecting against unauthorized model use.
Why it matters
This paper introduces a highly effective and robust LLM watermark, TextSeal, that surpasses existing methods in detection strength and maintains output quality. Its unique ability to transfer through model distillation offers crucial protection against unauthorized use, addressing key challenges in LLM provenance and intellectual property.
Original Abstract
We introduce TextSeal, a state-of-the-art watermark for large language models. Building on Gumbel-max sampling, TextSeal introduces dual-key generation to restore output diversity, along with entropy-weighted scoring and multi-region localization for improved detection. It supports serving optimizations such as speculative decoding and multi-token prediction, and does not add any inference overhead. TextSeal strictly dominates baselines like SynthID-text in detection strength and is robust to dilution, maintaining confident localized detection even in heavily mixed human/AI documents. The scheme is theoretically distortion-free, and evaluation across reasoning benchmarks confirms that it preserves downstream performance; while a multilingual human evaluation (6000 A/B comparisons, 5 languages) shows no perceptible quality difference. Beyond its use for provenance detection, TextSeal is also ``radioactive'': its watermark signal transfers through model distillation, enabling detection of unauthorized use.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.