CATMark: A Context-Aware Thresholding Framework for Robust Cross-Task Watermarking in Large Language Models

2025/09/27 by Zhang, Yu, Liu, Shuliang, Yang, Xu +1
#Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Cryptography and Security (cs.CR) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2510.02342

Abstract

Watermarking algorithms for Large Language Models (LLMs) effectively identify machine-generated content by embedding and detecting hidden statistical features in text. However, such embedding leads to a decline in text quality, especially in low-entropy scenarios where performance needs improvement. Existing methods that rely on entropy thresholds often require significant computational resources for tuning and demonstrate poor adaptability to unknown or cross-task generation scenarios. We propose Context-Aware Threshold watermarking (\myalgo), a novel framework that dynamically adjusts watermarking intensity based on real-time semantic context. \myalgo partitions text generation into semantic states using logits clustering, establishing context-aware entropy thresholds that preserve fidelity in structured content while embedding robust watermarks. Crucially, it requires no pre-defined thresholds or task-specific tuning. Experiments show \myalgo improves text quality in cross-tasks without sacrificing detection accuracy.

Citations

• Towards Watermarking of Open-Source LLMs
• ESPERANTO: Evaluating Synthesized Phrases to Enhance Robustness in AI Detection for Text Origination
• Qwen2.5-Coder Technical Report
• Watermarking Techniques for Large Language Models: A Survey
• PostMark: A Robust Blackbox Watermark for Large Language Models
• Watermarking Generative Tabular Data
• An Entropy-based Text Watermarking Detection Method
• ClaimVer: Explainable Claim-Level Verification and Evidence Attribution of Text Through Knowledge Graphs
• Authorship and the Politics and Ethics of LLM Watermarks
• DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence
• Adaptive Text Watermark for Large Language Models
• A Survey of Text Watermarking in the Era of Large Language Models
• Towards Optimal Statistical Watermarking
• WatME: Towards Lossless Watermarking Through Lexical Redundancy
• WaterBench: Towards Holistic Evaluation of Watermarks for Large Language Models
• REMARK-LLM: A Robust and Efficient Watermarking Framework for Generative Large Language Models
• Unbiased Watermark for Large Language Models
• Advancing Beyond Identification: Multi-bit Watermark for Large Language Models
• Large Language Models
• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
• Undetectable Watermarks for Language Models
• Who Wrote this Code? Watermarking for Code Generation
• A Watermark for Large Language Models
• Program Synthesis with Large Language Models
• Evaluating Large Language Models Trained on Code
• Measuring Mathematical Problem Solving With the MATH Dataset

Related

• From Trade-off to Synergy: A Versatile Symbiotic Watermarking Framework for Large Language Models
• Entropy-Guided Watermarking for LLMs: A Test-Time Framework for Robust and Traceable Text Generation
• Adaptive Text Watermark for Large Language Models
• Learning to Watermark: A Selective Watermarking Framework for Large Language Models via Multi-Objective Optimization
• BiMarker: Enhancing Text Watermark Detection for Large Language Models with Bipolar Watermarks