WaterMax: breaking the LLM watermark detectability-robustness-quality trade-off

2024/03/06 by Eva Giboulot, Giboulot, Eva, Teddy Furon +1 · 16 citations
Computer Science · #Computation and Language (cs.CL) #Cryptography and Data Security #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · pdf · doi:10.48550/arxiv.2403.04808

openalex publication_date 2024/03/06 · openalex created_date 2024/03/13 · openalex updated_date 2026/07/28

Abstract

Watermarking is a technical means to dissuade malfeasant usage of Large Language Models. This paper proposes a novel watermarking scheme, so-called WaterMax, that enjoys high detectability while sustaining the quality of the generated text of the original LLM. Its new design leaves the LLM untouched (no modification of the weights, logits, temperature, or sampling technique). WaterMax balances robustness and complexity contrary to the watermarking techniques of the literature inherently provoking a trade-off between quality and robustness. Its performance is both theoretically proven and experimentally validated. It outperforms all the SotA techniques under the most complete benchmark suite. Code available at https://github.com/eva-giboulot/WaterMax.

Cited by

• How Good is Post-Hoc Watermarking With Language Model Rephrasing?
• Optimal Watermark Generation under Type I and Type II Errors
• WaterSearch: A Quality-Aware Search-based Watermarking Framework for Large Language Models
• WaterMod: Modular Token-Rank Partitioning for Probability-Balanced LLM Watermarking
• Watermarking Discrete Diffusion Language Models
• δ-STEAL: LLM Stealing Attack with Local Differential Privacy
• Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoption
• SimKey: A Semantically Aware Key Module for Watermarking Language Models
• A Linguistics-Aware LLM Watermarking via Syntactic Predictability
• Detecting Post-generation Edits to Watermarked LLM Outputs via Combinatorial Watermarking
• On the Empirical Power of Goodness-of-Fit Tests in Watermark Detection
• AdaDetectGPT: Adaptive Detection of LLM-Generated Text with Statistical Guarantees
• PMark: Towards Robust and Distortion-free Semantic-level Watermarking with Channel Constraints
• Improving Detection of Watermarked Language Models
• Majority Bit-Aware Watermarking For Large Language Models
• Enhancing LLM Watermark Resilience Against Both Scrubbing and Spoofing Attacks

Related

• Learning to Watermark: A Selective Watermarking Framework for Large Language Models via Multi-Objective Optimization
• Theoretically Grounded Framework for LLM Watermarking: A Distribution-Adaptive Approach
• WaterPool: A Watermark Mitigating Trade-offs among Imperceptibility, Efficacy and Robustness
• Can Watermarked LLMs be Identified by Users via Crafted Prompts?
• Entropy-Guided Watermarking for LLMs: A Test-Time Framework for Robust and Traceable Text Generation