Reproduce “Attribute-based Undetectable Watermarking for Generative AI Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Miryam Mi-Ying Huang, Chung-Wei Lee, Max Raffel, Er-Cheng Tang
The authors introduce attribute-based watermarking that lets providers delegate AI-content detection by policy-defined category rather than issue a universal key. It binds watermark randomness to an output’s attributes using a constrained pseudorandom function, then gives detectors keys restricted to authorized categories. In tests, the Software key detected 98.0% of Software-prompt outputs, while the Sports key detected 17.65%; cross-category detections remained nonzero.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.