mlsanityData to 7 Oct 2026

Optimizing watermarks for large language models

arXiv 2312.17295 ↗First posted 28 Dec 2023cs.CR cs.AI cs.CLcited by 22

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

Abstract (arXiv)

With the rise of large language models (LLMs) and concerns about potential misuse, watermarks for generative LLMs have recently attracted much attention. An important aspect of such watermarks is the trade-off between their identifiability and their impact on the quality of the generated text. This paper introduces a systematic approach to this trade-off in terms of a multi-objective optimization problem. For a large class of robust, efficient watermarks, the associated Pareto optimal solutions are identified and shown to outperform the currently default watermark.

Key facts. Cited by 22 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 32 of its 50 references that are in the map. First posted on arXiv 28 Dec 2023.

Builds on

32 of 50 references.

Closest ancestors · 21

Same direction, newest first

  1. 2023On the Learnability of Watermarks for Language Models7 Dec
  2. 2023Mark My Words: Analyzing and Evaluating Language Model Watermarks1 Dec
  3. 2023Improving the Generation Quality of Watermarked Large Language Models via Word Importance Scoring16 Nov
  4. 2023WaterBench: Towards Holistic Evaluation of Watermarks for Large Language Models13 Nov
  5. 2023Watermarks in the Sand: Impossibility of Strong Watermarking for Generative Models7 Nov
  6. 2023Watermarking LLMs with Weight Quantization17 Oct
  7. 2023A Semantic Invariant Robust Watermark for Large Language Models10 Oct
  8. 2023Unbiased Watermark for Large Language Models22 Sep
  9. 2023An Unforgeable Publicly Verifiable Watermark for Large Language Models30 Jul
  10. 2023Robust Distortion-free Watermarks for Language Models28 Jul
  11. 2023Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy25 Jul
  12. 2023Provable Robust Watermarking for AI-Generated Text30 Jun
9 more
  1. 2023On the Reliability of Watermarks for Large Language Models7 Jun
  2. 2023Baselines for Identifying Watermarked Large Language Models29 May
  3. 2023Who Wrote this Code? Watermarking for Code Generation24 May
  4. 2023Watermarking Text Generated by Black-Box Language Models14 May
  5. 2023Evading Watermark based Detection of AI-Generated Content5 May
  6. 2023Robust Multi-bit Natural Language Watermarking through Invariant Features3 May
  7. 2023Paraphrasing evades detectors of AI-generated text, but retrieval is an effective defense23 Mar
  8. 2023Can AI-Generated Text be Reliably Detected?17 Mar
  9. 2023A Watermark for Large Language Models24 Jan

Other references · 11

  1. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  2. 2022OPT: Open Pre-trained Transformer Language Modelscited by 3,464
  3. 2015A Diversity-Promoting Objective Function for Neural Conversation Modelscited by 1,068
  4. 2015Teaching Machines to Read and Comprehendcited by 1,039
  5. 2020Multilingual Denoising Pre-training for Neural Machine Translationcited by 1,028
  6. 2023DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvaturecited by 705
  7. 2023Generative Language Models and Automated Influence Operations: Emerging Threats and Potential Mitigationscited by 157
  8. 2020Evaluating the Evaluation of Diversity in Natural Language Generationcited by 127
  9. 2020Adversarial Watermarking Transformer: Towards Tracing Text Provenance with Data Hidingcited by 121
  10. 2019Neural Linguistic Steganographycited by 73
  11. 2023Red Teaming Language Model Detectors with Language Modelscited by 55

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 2; 2025: 8; 2026: 12.

22 papers.

Sort:

By area:

  1. 2024Watermarking Techniques for Large Language Models: A Surveycited by 20
  2. 2025AdaDetectGPT: Adaptive Detection of LLM-Generated Text with Statistical Guaranteescited by 20
  3. 2024Theoretically Grounded Framework for LLM Watermarking: A Distribution-Adaptive Approachcited by 17
  4. 2026Learn-to-Distance: Distance Learning for Detecting LLM-Generated Textcited by 8
  5. 2025From Trade-off to Synergy: A Versatile Symbiotic Watermarking Framework for Large Language Modelscited by 5
  6. 2025Invisible Entropy: Towards Safe and Efficient Low-Entropy LLM Watermarkingcited by 5
  7. 2025BitMark: Watermarking Bitwise Autoregressive Image Generative Modelscited by 5
  8. 2025SciPostGen: Bridging the Gap between Scientific Papers and Poster Layoutscited by 4
  9. 2026PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant Attackscited by 4
  10. 2026A Unified Framework for LLM Watermarkscited by 3
  11. 2025SoK: Are Watermarks in LLMs Ready for Deployment?cited by 2
  12. 2026Feature-Augmented Transformers for Robust AI-Text Detection Across Domains and Generatorscited by 2
10 more
  1. 2025HeavyWater and SimplexWater: Distortion-Free LLM Watermarks for Low-Entropy Next-Token Predictionscited by 1
  2. 2026Robust LLM Watermarking with Minimal Semantic Distortion for IP Protectioncited by 1
  3. 2026SAMark: A Self-Anchored Text Watermarking with Paragraph-Level Paraphrase Robustnesscited by 1
  4. 2025Robust Spectral Watermark for Synthetic Tabular Datacited by 0
  5. 2026On Google's SynthID-Text LLM Watermarking System: Theoretical Analysis and Empirical Validationcited by 0
  6. 2026Optimal Multi-bit Generative Watermarking Schemes Under Worst-Case False-Alarm Constraintscited by 0
  7. 2026Trustworthy AI: Ensuring Reliability and Accountability from Models to Agentscited by 0
  8. 2026AliMark: Enhancing Robustness of Sentence-Level Watermarking Against Text Paraphrasingcited by 0
  9. 2026Signature filtering: a lightweight enhancement for statistical watermark detection in large language modelscited by 0
  10. 2026Beyond Heuristic Tuning: Power-Calibrated LLM Watermarkingcited by 0