mlsanityData to 7 Oct 2026

Improved Unbiased Watermark for Large Language Models

arXiv 2502.11268 ↗First posted 16 Feb 2025cs.CLcited by 13

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

Abstract (arXiv)

As artificial intelligence surpasses human capabilities in text generation, the necessity to authenticate the origins of AI-generated content has become paramount. Unbiased watermarks offer a powerful solution by embedding statistical signals into language model-generated text without distorting the quality. In this paper, we introduce MCmark, a family of unbiased, Multi-Channel-based watermarks. MCmark works by partitioning the model's vocabulary into segments and promoting token probabilities within a selected segment based on a watermark key. We demonstrate that MCmark not only preserves the original distribution of the language model but also offers significant improvements in detectability and robustness over existing unbiased watermarks. Our experiments with widely-used language models demonstrate an improvement in detectability of over 10% using MCmark, compared to existing state-of-the-art unbiased watermarks. This advancement underscores MCmark's potential in enhancing the practical application of watermarking in AI-generated texts.

Key facts. Cited by 13 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 12 of its 19 references that are in the map. First posted on arXiv 16 Feb 2025.

Builds on

12 of 19 references.

Closest ancestors · 5

Same direction, newest first

  1. 2024A Watermark for Order-Agnostic Language Models17 Oct
  2. 2024Distortion-free Watermarks are not Truly Distortion-free under Watermark Key Collisions2 Jun
  3. 2023Robust Distortion-free Watermarks for Language Models28 Jul
  4. 2023Provable Robust Watermarking for AI-Generated Text30 Jun
  5. 2023Paraphrasing evades detectors of AI-generated text, but retrieval is an effective defense23 Mar

Other references · 7

  1. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  2. 2024The Llama 3 Herd of Modelscited by 14,192
  3. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  4. 2019BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehensioncited by 5,874
  5. 2024Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phonecited by 1,883
  6. 2017Get To The Point: Summarization with Pointer-Generator Networkscited by 1,281
  7. 2020Multilingual Denoising Pre-training for Neural Machine Translationcited by 1,028

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 4; 2026: 9.

13 papers.

Sort:

By area:

  1. 2025In-Context Watermarks for Large Language Modelscited by 9
  2. 2025Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoptioncited by 7
  3. 2025Leave No TRACE: Black-box Detection of Copyrighted Dataset Usage in Large Language Models via Watermarkingcited by 5
  4. 2026A Unified Framework for LLM Watermarkscited by 3
  5. 2026MC2Mark: Distortion-Free Multi-Bit Watermarking for Long Messagescited by 2
  6. 2025Can You Detect the Difference?cited by 1
  7. 2026Every Bit, Everywhere, All at Once: A Binomial Multibit LLM Watermarkcited by 1
  8. 2026Robust LLM Watermarking with Minimal Semantic Distortion for IP Protectioncited by 1
  9. 2026Tracing Provenance and Detecting Tampering with Complementary LLM Watermarkscited by 1
  10. 2026WorldCup Sampling for Multi-bit LLM Watermarkingcited by 0
  11. 2026Optimal Multi-bit Generative Watermarking Schemes Under Worst-Case False-Alarm Constraintscited by 0
  12. 2026Learning to Follow In-Context Watermark Instructions via Self-Distillationcited by 0
1 more
  1. 2026TTMark: Pairwise Distortion-Free Watermarking Beyond Single-Token Entropycited by 0