mlsanityData to 14 Oct 2026

Addressing Tokenization Inconsistency in Steganography and Watermarking Based on Large Language Models

arXiv 2508.20718 ↗First posted 28 Aug 2025cs.CLcited by 4

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

Abstract (arXiv)

Large language models have significantly enhanced the capacities and efficiency of text generation. On the one hand, they have improved the quality of text-based steganography. On the other hand, they have also underscored the importance of watermarking as a safeguard against malicious misuse. In this study, we focus on tokenization inconsistency (TI) between Alice and Bob in steganography and watermarking, where TI can undermine robustness. Our investigation reveals that the problematic tokens responsible for TI exhibit two key characteristics: infrequency and temporariness. Based on these findings, we propose two tailored solutions for TI elimination: a stepwise verification method for steganography and a post-hoc rollback method for watermarking. Experiments show that (1) compared to traditional disambiguation methods in steganography, directly addressing TI leads to improvements in fluency, imperceptibility, and anti-steganalysis capacity; (2) for watermarking, addressing TI enhances detectability and robustness against attacks.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Placed in Watermarking generated content and models, one of 12 directions in LLM safety and security. Builds on 23 of its 42 references that are in the map. First posted on arXiv 28 Aug 2025.

Builds on

23 of 42 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024An Entropy-based Text Watermarking Detection Method20 Mar
  2. 2024GumbelSoft: Diversified Language Model Watermarking via the GumbelMax-trick20 Feb
  3. 2023Robust Distortion-free Watermarks for Language Models28 Jul
  4. 2023Provable Robust Watermarking for AI-Generated Text30 Jun

Other references · 19

  1. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  2. 2020Language Models are Few-Shot Learnerscited by 32,419
  3. 2023GPT-4 Technical Reportcited by 18,396
  4. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,599
  5. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  6. 2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,224
  7. 2015Neural Machine Translation of Rare Words with Subword Unitscited by 2,918
  8. 2018SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processingcited by 1,989
  9. 2024Continual Pre-Training for Cross-Lingual LLM Adaptation: Enhancing Japanese Language Capabilitiescited by 111
  10. 2024Coercing LLMs to do and reveal (almost) anythingcited by 89
  11. 2019Neural Linguistic Steganographycited by 78
  12. 2021The Threat of Offensive AI to Organizationscited by 38
7 more
  1. 2019Towards Near-imperceptible Steganographic Textcited by 32
  2. 2024Tokenization Matters! Degrading Large Language Models through Challenging Their Tokenizationcited by 32
  3. 2024Glitch Tokens in Large Language Models: Categorization Taxonomy and Effective Detectioncited by 29
  4. 2020Near-imperceptible Neural Linguistic Steganography via Self-Adjusting Arithmetic Codingcited by 17
  5. 2022Addressing Segmentation Ambiguity in Neural Linguistic Steganographycited by 7
  6. 2022Tokenization Consistency Matters for Generative Models on Extractive NLP Taskscited by 7
  7. 2024GlitchMiner: Mining Glitch Tokens in Large Language Models via Gradient-based Discrete Optimizationcited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026Hide and Seek in Embedding Space: Geometry-based Steganography and Detection in Large Language Modelscited by 1
  2. 2026HiTMS: A High-Throughput Multi-Stream Linguistic Steganography Frameworkcited by 1
  3. 2026Text Steganography with Dynamic Codebook and Multimodal Large Language Modelcited by 0
  4. 2026Tokenization-Invariant Watermarking: A Byte-Local Method for LLM Textcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Can Watermarks Survive Translation? On the Cross-lingual Consistency of Text Watermark for Large Language Models
  2. 2025Evaluating the Robustness and Accuracy of Text Watermarking Under Real-World Cross-Lingual Manipulations
  3. 2025Character-Level Perturbations Disrupt LLM Watermarks
  4. 2023On the Learnability of Watermarks for Language Models
  5. 2024Provably Robust Watermarks for Open-Source Language Models
  6. 2023On the Reliability of Watermarks for Large Language Models
  7. 2026Language Model Fingerprinting Requires Rethinking Watermark Teachers
  8. 2023A Semantic Invariant Robust Watermark for Large Language Models