Agents & Tools23 September 2026, 16:00 UTC1 min read

Advancing Private AI Compute with secure, server-side memory

The new capability stores information in an encrypted cloud vault while keeping the cryptographic keys only on users’ personal devices, so the data remains inaccessible to anyone else, including Google.

2 sources primary source

Key points

Private AI Compute now supports persistent, cross‑device memory stored in encrypted cloud vaults.
Encryption keys stay on users’ personal devices, preventing even Google from accessing the data.
Data is decrypted only inside a secure enclave via an end‑to‑end encrypted channel for each request.

When an AI model needs to retrieve context, an authenticated end‑to‑end encrypted channel links the device to an isolated cloud environment called a secure enclave. The enclave temporarily decrypts the data in isolated memory, processes the request, saves any new context, and re‑encrypts it immediately. This design aims to combine long‑term continuity for AI assistants with the strict privacy standards normally limited to on‑device processing.

Full story from Google DeepMind · by Google Private AI Compute Team primary sourceOpen source ↗