An Unforgeable Publicly Verifiable Watermark for Large Language Models

2023/07/30 by Aiwei Liu, Liu, Aiwei, Leyi Pan +11 · 10 citations
Computer Science · #68T50 #Computation and Language (cs.CL) #FOS: Computer and information sciences #Hate Speech and Cyberbullying Detection #I.2.7 #Internet Traffic Analysis and Secure E-voting #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2307.16230

openalex publication_date 2023/07/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Recently, text watermarking algorithms for large language models (LLMs) have been proposed to mitigate the potential harms of text generated by LLMs, including fake news and copyright issues. However, current watermark detection algorithms require the secret key used in the watermark generation process, making them susceptible to security breaches and counterfeiting during public detection. To address this limitation, we propose an unforgeable publicly verifiable watermark algorithm named UPV that uses two different neural networks for watermark generation and detection, instead of using the same key at both stages. Meanwhile, the token embedding parameters are shared between the generation and detection networks, which makes the detection network achieve a high accuracy very efficiently. Experiments demonstrate that our algorithm attains high detection accuracy and computational efficiency through neural networks. Subsequent analysis confirms the high complexity involved in forging the watermark from the detection network. Our code is available at \hrefhttps://github.com/THU-BPM/unforgeable_watermarkhttps://github.com/THU-BPM/unforgeable_watermark. Additionally, our algorithm could also be accessed through MarkLLM \citeppan2024markllm \footnotehttps://github.com/THU-BPM/MarkLLM.

Cited by

• Adaptive Testing for Segmenting Watermarked Texts From Language Models
• PVMark: Enabling Public Verifiability for LLM Watermarking Schemes
• Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoption
• Analyzing and Evaluating Unbiased Language Model Watermark
• LLM Watermark Evasion via Bias Inversion
• RLCracker: Evaluating the Worst-Case Vulnerability of LLM Watermarks with Adaptive RL Attacks
• MarkDiffusion: An Open-Source Toolkit for Generative Watermarking of Latent Diffusion Models
• Paladin: Defending LLM-enabled Phishing Emails with a New Trigger-Tag Paradigm
• SAEMark: Multi-bit LLM Watermarking with Inference-Time Scaling
• Majority Bit-Aware Watermarking For Large Language Models

Related

• A Watermark for Large Language Models
• Necessary and Sufficient Watermark for Large Language Models
• Signal Watermark on Large Language Models
• Provable Robust Watermarking for AI-Generated Text
• Token-Specific Watermarking with Enhanced Detectability and Semantic Coherence for Large Language Models