Research2026-05-14
Watermarking Should Be Treated as a Monitoring Primitive
Source: Arxiv CS.AI
arXiv:2605.13095v1 Announce Type: cross Abstract: Watermarking is widely proposed for provenance, attribution, and safety monitoring in generative models, yet is typically evaluated only under adversaries who attempt to evade detection or induce false positives at the level of individual samples....
arxivpapers