AI Multibit Robust Watermark
Recovering content identifiers after text rewriting.
Research Intern, University of California, Santa Barbara · June 2026–present
Advisor: Prof. Yuheng Bu · Status: ARR under submission
This project develops robust multi-bit text watermarking for content attribution. Embedded identifiers can be recovered after paraphrasing without access to the original text or watermark-specific training.
The evaluated system achieved 99.83% watermark-bit injection success and 99.70% exact recovery of 32-bit identifiers in 10,000 resampled trials under the evaluated rewriting protocol.
On natural text, identifiers were recovered from all 20 evaluated Wikipedia and arXiv documents after rewriting at a target code rate of 0.2. These results support reliable identifier recovery within the evaluated setting.