LM-Watermark

Survey of watermarking for (large) language models

---

White-Box Watermark

• 2021 Protect, show, attend and tell- Empowering image captioning models with ownership protection (Jian Han Lim) Universiti Malaya

  • paper: https://www.sciencedirect.com/science/article/pii/S0031320321004659 (Pattern Recognition 2021) Crack it in Chinese
  • code: https://github.com/jianhanlim/ipr-imagecaptioning

• 2022 An Embarrassingly Simple Approach for Intellectual Property Rights Protection on Recurrent Neural Networks (Zhi Qin Tan) Universiti Malaya

  • paper: https://aclanthology.org/2022.aacl-main.8.pdf (AACL-IJCNLP 2022)
  • code: https://github.com/zhiqin1998/RecurrentIPR

• 2023 An Effective Framework for Intellectual Property Protection of NLG Models (Mingjie Li) Shanghai University

  • paper: https://www.mdpi.com/2073-8994/15/6/1287 (Symmetry 2023)
  • code:

Black-Box Watermark

NLU

• 2021 Robust Black-box Watermarking for Deep Neural Network using Inverse Document Frequency (Mohammad Mehdi Yadollahi) University of New Brunswick

  • paper: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9730156 (DASC-PICom-CBDCom-CyberSciTech 2021) Crack it in Chinese
  • code:

• 2022 TextBack: Watermarking Text Classifiers using Backdooring (Nandish Chattopadhyay) Nandish Chattopadhyay

  • paper: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9996658 (25th Euromicro Conference on Digital System Design 2022) Crack it in Chinese
  • code:

• 2023 PLMmark: A Secure and Robust Black-Box Watermarking Framework for Pre-trained Language Models (Peixuan Li) Shanghai Jiao Tong University

  • paper: https://ojs.aaai.org/index.php/AAAI/article/view/26750 (AAAI 2023) Crack it in Chinese
  • code:

• 2023 Are You Copying My Model? Protecting the Copyright of Large Language Models for EaaS via Backdoor Watermark (Peng) USTC

  • paper: https://aclanthology.org/2023.acl-long.423.pdf (ACL 2023)
  • code:

NLG

• 2023 GPTs Don’t Keep Secrets: Searching for Backdoor Watermark Triggers in Autoregressive Language Models (Evan Lucas) Michigan Technological University
  • paper: https://aclanthology.org/2023.trustnlp-1.21.pdf (TrustNLP 2023) Crack it in Chinese
  • code: https://github.com/evan-person/findingBackdoorWatermarks

Non-Box Watermark

Decoding-Based

• 2023 A Watermark for Large Language Models (John Kirchenbauer) University of Maryland

  • paper: https://proceedings.mlr.press/v202/kirchenbauer23a/kirchenbauer23a.pdf (ICML 2023)
  • code: https://github.com/jwkirchenbauer/lm-watermarking

• 2023 Provable Robust Watermarking for AI-Generated Text (Xuandong Zhao) UC Santa Barbara

  • paper: https://openreview.net/pdf?id=Bwz0fy9Hc9 (ICML Workshop 2023)
  • code: https://github.com/XuandongZhao/GPTWatermark

Editing-Based

• 2011 Watermarking the Outputs of Structured Prediction with an Application in Statistical Machine Translation （Ashish Venugopal）Google

  • paper: https://aclanthology.org/D11-1126.pdf (EMNLP 2011) Crack it in Chinese
  • code:

• 2022 Distillation-Resistant Watermarking for Model Protection in NLP (Xuandong Zhao) UC Santa Barbara

  • paper: https://aclanthology.org/2022.findings-emnlp.370.pdf (EMNLP Findings 2022)
  • code: https://github.com/XuandongZhao/DRW

• 2022 Protecting Intellectual Property of Language Generation APIs with Lexical Watermark (Xuanli He) Monash University

  • paper: https://ojs.aaai.org/index.php/AAAI/article/view/21321 (AAAI 2022) Crack it in Chinese
  • code: https://github.com/xlhex/NLG_api_watermark

• 2022 CATER: Intellectual Property Protection on Text Generation APIs via Conditional Watermarks (Xuanli He) University College London

  • paper: https://openreview.net/pdf?id=L7P3IvsoUXY (NeurIPS 2022) Crack it in Chinese
  • code: https://github.com/xlhex/cater_neurips

• 2023 Protecting Language Generation Models via Invisible Watermarking (Xuandong Zhao) UC Santa Barbara

  • paper: http://proceedings.mlr.press/v202/zhao23i/zhao23i.pdf (ICML 2023)
  • code: https://github.com/XuandongZhao/Ginsew

• 2023 A novel watermarking framework for intellectual property protection of NLG APIs (Mingjie Li) Shanghai University

  • paper: https://www.sciencedirect.com/science/article/pii/S0925231223008238 (NeuroComputing 2023) Crack it in Chinese
  • code:

• 2023 COSYWA: Enhancing Semantic Integrity in Watermarking Natural Language Generation (Junjie Fang) Xiamen University

  • paper: https://dl.acm.org/doi/abs/10.1007/978-3-031-44693-1_55 (NLPCC 2023) Crack it in Chinese
  • code:

---

arXiv

• 20221118 DeepHider: A Covert NLP Watermarking Framework Based on Multi-task Learning (Dai) Hainan University

  • paper: https://arxiv.org/ftp/arxiv/papers/2208/2208.04676.pdf (arXiv)
  • code:

• 20230210 Watermarking Pre-trained Language Models with Backdooring (Chenxi Gu) Fudan University

  • paper: https://arxiv.org/pdf/2210.07543.pdf (arXiv)
  • code:

• 20230309 DeepTextMark: Deep Learning based Text Watermarking for Detection of Large Language Model Generated Text (Travis Munyer) University of Nebraska Omaha

  • paper: https://arxiv.org/pdf/2305.05773.pdf （arXiv）
  • code:

• 20230514 Watermarking Text Generated by Black-Box Language Models (Yang) USTC

  • paper: https://arxiv.org/pdf/2305.08883.pdf (arXiv)
  • code: https://github.com/Kiode/Text_Watermark_Language_Models

• 20230522 Watermarking Text Data on Large Language Models for Dataset Copyright Protection (Liu) Lehigh University

  • paper: https://arxiv.org/pdf/2305.13257.pdf
  • code:

• 20230524 Who Wrote this Code? Watermarking for Code Generation (Taehyun Lee) Seoul National University

  • paper: https://arxiv.org/pdf/2305.15060.pdf (arXiv)
  • code:

• 20230525 Undetectable Watermarks for Language Models (Miranda Christ) Columbia University

  • paper: https://arxiv.org/pdf/2306.09194.pdf (arXiv)
  • code:

• 20230529 Baselines for Identifying Watermarked Large Language Models (Leonard Tang) Harvard University

  • paper: https://arxiv.org/pdf/2305.18456.pdf (arXiv)
  • code:

• 20230630 On the Reliability of Watermarks for Large Language Models (John Kirchenbauer) University of Maryland

  • paper: https://arxiv.org/pdf/2306.04634.pdf (arXiv)
  • code: https://github.com/jwkirchenbauer/lm-watermarking

• 20230725 Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy (Fu) University of California, Riverside

  • paper: https://arxiv.org/pdf/2307.13808.pdf (arXiv)
  • code:

• 20320726 Three Bricks to Consolidate Watermarks for Large Language Models (Pierre Fernandez) Centre Inria de l’Universite de Rennes

  • paper: https://pierrefdz.github.io/assets/publis/threebricks/paper.pdf (arXiv)
  • code:

• 20230728 Robust Distortion-free Watermarks for Language Models (Rohith Kuditipudi) Stanford University

  • paper: https://arxiv.org/pdf/2307.15593.pdf (arXiv)
  • code: https://github.com/jthickstun/watermark

• 20230729 Towards Codable Text Watermarking for Large Language Models (Liang) WeChat AI,

  • paper: https://arxiv.org/pdf/2307.15992.pdf (arXiv)
  • code: https://github.com/lancopku/codable-watermarking-for-llm

• 20230801 Advancing Beyond Identification- Multi-bit Watermark for Language Models (Yoo) Seoul National University

  • paper: http://arxiv.org/pdf/2308.00221 (arXiv)
  • code:

• 20230802 A Private Watermark for Large Language Models (Liu) Tsinghua University

  • paper: https://arxiv.org/pdf/2307.16230.pdf (arXiv)
  • code: https://github.com/THU-BPM/private_watermark

• 20230822 Evading Watermark based Detection of AI-Generated Content (Jiang) Duke University

  • paper: https://arxiv.org/pdf/2305.03807.pdf (arXiv)
  • code: https://github.com/zhengyuan-jiang/WEvade