RoSeMary introduces a breakthrough crypto-ML watermarking framework that protects intellectual property in LLM-generated code while maintaining usability.

• Addresses the unique challenges of watermarking code (vs. text) due to its structured, low-entropy nature
• Leverages zero-knowledge proofs for secure verification without revealing watermark signatures
• Provides robust protection against removal attacks while maintaining code functionality
• Establishes a balanced approach to the detectability-fidelity-robustness tradeoff

This research enables organizations to safely integrate AI-generated code into development workflows while protecting intellectual property rights and preventing misuse.

Original Paper: Robust and Secure Code Watermarking for Large Language Models via ML/Crypto Codesign