This research introduces a novel robust binary code (RBC) watermarking framework that efficiently identifies AI-generated content while preserving text quality.

• Embeds statistical signals using error-correcting codes that humans cannot detect
• Achieves improved detection rates compared to existing methods
• Maintains robustness against text modifications and tampering attempts
• Introduces zero distortion to the original probability distributions

As AI-generated content becomes increasingly realistic, this watermarking approach provides essential security capabilities for content authentication and mitigating potential misuse of language models.

Watermarking Language Models with Error Correcting Codes