Toward Stronger Code Watermarking: A Grammar-Driven Approach to Optimizing the Trade-off Between Quality and Detectability

2026-07-11

AI Summary

This work addresses the inherent tension between code quality and watermark detectability in existing logits-based text watermarking methods for code generation, a conflict exacerbated by the low entropy of source code. To resolve this, the authors propose Grammar-Driven Watermarking (GDW), the first approach that integrates syntactic structure into watermark design. GDW employs a grammar-guided three-level masking mechanism to guarantee syntactic validity, applies strong bias to content-bearing tokens and weak bias to syntax-critical tokens through structure-aware modulation, and introduces a role-aware weighted detection statistic aligned with the generation process. Experiments demonstrate that GDW consistently outperforms state-of-the-art methods across multiple programming languages, models, and decoding strategies, achieving superior watermark detectability without compromising code quality and exhibiting robustness against variable renaming attacks.

Abstract

With the rapid development of Large Language Models (LLMs), text watermarking has emerged as a crucial technique for identifying machine-generated content. However, directly applying existing logits-based watermarking methods to code generation remains challenging, since the low-entropy nature of code exacerbates the trade-off between code quality and watermark detectability. In this paper, we propose a novel code watermarking approach called Grammar-Driven Watermark (GDW) for LLMs. GDW preserves syntactic validity through a grammar-guided three-level masking mechanism and injects watermark signals via structural role-aware modulation, assigning a stronger bias to content-bearing tokens while applying a more conservative bias to syntax-critical tokens. Aligning with the generation process, we further design a role-aware weighted detection statistic to improve detectability. Experiments across multiple programming languages, models, and decoding strategies show that GDW establishes a stronger quality-detectability trade-off frontier than existing methods, while maintaining robustness against variable-renaming attacks.

Problem

• code watermarking
• quality-detectability trade-off
• low-entropy code
• LLM-generated code
• watermark detectability

Innovation

• code watermarking
• grammar-guided masking
• role-aware modulation
• quality-detectability trade-off
• LLM-generated code