mlsanityData to 7 Oct 2026

TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection

arXiv 2605.12456 ↗First posted 12 May 2026cs.CR cs.CL cs.LGcited by 2

On the map: LLM safety and security → Watermarking generated content and models

Abstract (arXiv)

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.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Watermarking generated content and models, one of 12 directions in LLM safety and security. Builds on 36 of its 64 references that are in the map. First posted on arXiv 12 May 2026.

Builds on

36 of 64 references.

Closest ancestors · 23

Same direction, newest first

  1. 2025How Good is Post-Hoc Watermarking With Language Model Rephrasing?18 Dec
  2. 2025Leave No TRACE: Black-box Detection of Copyrighted Dataset Usage in Large Language Models via Watermarking3 Oct
  3. 2025MorphMark: Flexible Adaptive Watermarking for Large Language Models14 May
  4. 2024Robust Multi-bit Text Watermark with LLM-based Paraphrasers4 Dec
  5. 2024Waterfall: Framework for Robust and Scalable Text Watermarking and Provenance for LLMs5 Jul
  6. 2024MarkLLM: An Open-Source Toolkit for LLM Watermarking16 May
  7. 2024WaterMax: breaking the LLM watermark detectability-robustness-quality trade-off6 Mar
  8. 2024Watermarking Makes Language Models Radioactive22 Feb
  9. 2024k-SemStamp: A Clustering-Based Semantic Watermark for Detection of Machine-Generated Text17 Feb
  10. 2024Adaptive Text Watermark for Large Language Models25 Jan
  11. 2023On the Learnability of Watermarks for Language Models7 Dec
  12. 2023Mark My Words: Analyzing and Evaluating Language Model Watermarks1 Dec
11 more
  1. 2023REMARK-LLM: A Robust and Efficient Watermarking Framework for Generative Large Language Models18 Oct
  2. 2023A Semantic Invariant Robust Watermark for Large Language Models10 Oct
  3. 2023Advancing Beyond Identification: Multi-bit Watermark for Large Language Models1 Aug
  4. 2023Robust Distortion-free Watermarks for Language Models28 Jul
  5. 2023Three Bricks to Consolidate Watermarks for Large Language Models26 Jul
  6. 2023Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy25 Jul
  7. 2023On the Reliability of Watermarks for Large Language Models7 Jun
  8. 2023Undetectable Watermarks for Language Models25 May
  9. 2023Who Wrote this Code? Watermarking for Code Generation24 May
  10. 2023Robust Multi-bit Natural Language Watermarking through Invariant Features3 May
  11. 2023A Watermark for Large Language Models24 Jan

Other references · 13

  1. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  2. 2024Qwen2.5 Technical Reportcited by 4,713
  3. 2019The Curious Case of Neural Text Degenerationcited by 2,881
  4. 2022Fast Inference from Transformers via Speculative Decodingcited by 1,615
  5. 2025s1: Simple test-time scalingcited by 1,311
  6. 2019ELI5: Long Form Question Answeringcited by 567
  7. 2024Better & Faster Large Language Models via Multi-token Predictioncited by 344
  8. 2020Adversarial Watermarking Transformer: Towards Tracing Text Provenance with Data Hidingcited by 121
  9. 2020Radioactive data: tracing through trainingcited by 67
  10. 2024CaLMQA: Exploring culturally specific long-form question answering across 23 languagescited by 31
  11. 2021Frustratingly Easy Edit-based Linguistic Steganography with a Masked Language Modelcited by 28
  12. 2024Ward: Provable RAG Dataset Inference via LLM Watermarkscited by 15
1 more
  1. 2025Detecting Benchmark Contamination Through Watermarkingcited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Retrospective Distillation Attribution via Normalized Response Similaritycited by 0
  2. 2026Could LLM Watermark Detection be Public?cited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025StealthInk: A Multi-bit and Stealthy Watermark for Large Language Models
  2. 2026Improving the Trade-off Between Watermark Strength and Speculative Sampling Efficiency for Language Models
  3. 2026WaterMoE: Expert-Routing-based Watermarking for High Fidelity and Efficiency
  4. 2025Watermarking Discrete Diffusion Language Models
  5. 2026Language Model Fingerprinting Requires Rethinking Watermark Teachers
  6. 2026LLM Output Detectability and Task Performance Can be Jointly Optimized
  7. 2025Watermarking Diffusion Language Models
  8. 2026Global Sketch-Based Watermarking for Diffusion Language Models