## Source summary NIST NCCoE announced a privacy-preserving federated learning testbed and a soybean genomic dataset. The environment supports comparisons of centralized and federated training, with or without differential privacy. Federated learning keeps data at participating sites while training a shared model; differential privacy introduces noise to reduce individual-record exposure, potentially affecting accuracy. ## TABS Company's business perspective When planning cross-company AI, assess access rights, exchanged training information and model performance rather than treating local data retention alone as proof of safety. Specify both acceptable accuracy and privacy objectives in a PoC, and review approvals and data-use conditions before introducing sensitive data. Source: NIST, “Privacy-Preserving Federated Learning (PPFL) Testbed and Curated Dataset”, 2026-10-29. Original: https://www.nist.gov/news-events/news/2026/10/privacy-preserving-federated-learning-ppfl-testbed-and-curated-dataset Reuse basis: NIST public-information policy (https://www.nist.gov/copyrights-disclaimers); only institutional explanatory text without separate copyright restrictions is summarized. Unofficial AI-assisted English translation of TABS Company's Korean summary and adaptation. No endorsement or sponsorship by the source institution is implied.