Watermarking for Large Language Models

The rapid advancement of Large Language Models (LLMs) in code generation has raised significant attribution and intellectual property concerns. Code watermarking offers a potential solution but faces unique challenges due to programming languages' strict syntactic constraints and semantic requirements. To address these challenges, we introduce ACW (AST-guided Code Watermarking), a novel adaptive framework that leverages Abstract Syntax Tree (AST) analysis during training to learn watermark embedding strategies. Our framework identifies substitutable code components and strategically biases token selections to embed watermarks. We also propose a novel sampling scheme that distributes tokens between green/red lists according to semantic context, ensuring statistical distinguishability while preserving code functionality. Extensive experiments demonstrate that ACW achieves a significant improvement in watermark detection accuracy compared to existing methods, with negligible impact on code functionality. This adaptive framework offers a promising solution for effective and practical code watermarking in the age of LLMs. Our code is available at: https://github.com/TimeLovercc/code-watermark . Recent research has explored techniques like entropy-based methods and the utilization of variable type information to embed watermarks while maintaining type safety [19, 9] . However, a significant limitation of these approaches lies in their detection phase, which often necessitates access to the 39th Conference on Neural Information Processing Systems (NeurIPS 2025). def calculate_discount(price, is_member): if is_member == True: return price * 0.8 return price Challenge 1 Position: • Must be exactly "*" • No alternative forms : Not able to plant code watermarks here.