Google Unveils Next-Gen FL System Powered by TEEs for Enhanced Privacy

Google has unveiled its next-generation Federated Learning (FL) system, designed to provide stronger privacy guarantees and improved accuracy. The new system leverages Trusted Execution Environments (TEEs) to ensure that data anonymization is fully verifiable and auditable.

The previous FL systems had limitations in terms of transparency and control, allowing server operators to access device data without clear visibility. However, the new TEE-based system ensures that only metrics and differentially private model weights are visible to workload operators.

Devices participating in the new system can see the full set of server workloads that may access their data, with access policies published to a public transparency log called Rekor. External auditors can track these logs to monitor potential future server-side workloads.

Gboard has already adopted the new TEE-based FL system and is experiencing substantially faster compute times compared to the previous version.