Two papers accepted to NeurIPS 2025 as spotlights: one on the empirical evaluation of goodness-of-fit tests for watermark detection, and the other on mitigating the privacy–utility trade-off in decentralized federated learning.