AI RESEARCH
Hidden in Plain Tokens: Simply Robust, Gradient-Free Watermark for Synthetic Audio
arXiv CS.LG
•
ArXi:2605.25967v1 Announce Type: new As policy catches up with the capabilities of generative AI, watermarking is central to content provenance efforts. Inference-time watermarks for autoregressive models are unfit for continuous modalities due to discretization inconsistencies. Existing methods overcome this by finetuning the modality tokenizers, nullifying the watermark's