SoftBank Corp. Develops Synthetic Data Generation Pipeline to Enable Secure Training for its Large Telecom Model

SoftBank Corp. announced it built a synthetic data generation pipeline for its Large Telecom Model (LTM)*1, a generative AI foundation model for the telecommunications industry. This pipeline enables secure and highly accurate training while protecting confidential information related to network operations. By utilizing NVIDIA NeMo Safe Synthesizer, a synthetic data generation tool that supports Differential Privacy*2, SoftBank is successfully balancing the preservation of data quality while protecting confidential information, allowing the data to be safely used for LTM training.

*1: For more details on the “Large Telecom Model” (LTM), please refer to the press release dated March 19, 2025, titled “SoftBank Corp. Develops a Foundational Large Telecom Model (LTM).”

*2: Differential Privacy is a data protection technology that mathematically guarantees the output results will not significantly change depending on the presence or absence of specific individual data, achieved by adding strictly calculated noise during the data training or analysis process.

Read the full press release here.
https://www.softbank.jp/en/corp/news/press/sbkk/2026/20260317_01/