Google Moves Federated Learning Into TEEs With Externally Verifiable Differential Privacy

Google Research announced a next-generation federated learning system built on Trusted Execution Environments, claiming externally verifiable central differential privacy guarantees for the first time. Client gradient computation moves to attested server-side TEEs, with access policies published to Sigstore's Rekor transparency log. Gboard has already shipped English and Japanese next-word prediction models on the new system.

• Gradient computation moves from devices to attested server-side TEEs
• Access policies are published to Rekor and binaries are reproducibly buildable
• Gboard shipped English and Japanese next-word prediction models on the system
• Training once took 1–2 months per model; TEE capacity is now the limit