Google DeepMind shared a technical update on its Private AI Compute architecture, introducing persistent, server-side memory that resolves a longstanding dilemma in modern AI: how to give an assistant long-term continuity across devices while upholding the strict privacy standards typically limited to on-device processing.
The new persistent memory layer functions like a secure digital vault in the cloud. Information needed to assist users is sealed within dedicated, encrypted storage, while the cryptographic keys required to unlock it are held exclusively on personal devices — ensuring data is inaccessible to anyone else, even Google.
When an AI model needs to access information, an authenticated, end-to-end encrypted channel connects the user's device to a protected, isolated environment in the cloud. That 'secure enclave' temporarily decrypts data in isolated memory to handle the request, saves any new context, and immediately encrypts it — keeping information private as if it never left the device.
The architecture combines hardware-enforced secure enclaves, encrypted channels, and per-user databases shielded by device-derived encryption keys. This evolution meets the computing needs of the AI era: local on-device processing has been the gold standard for privacy, but frontier AI models require far more computing power than any one device can provide.
Until now, Private AI Compute and similar industry solutions were strictly 'stateless,' wiping all context when a task ended. Workarounds like having AI save personal fact lists weren't enough to support rich, continuous experiences. The new capability enables seamless assistance — imagine pulling up assembly instructions on a laptop previously viewed through smart glasses, or resuming complex conversations between mobile and web.
Alongside the technical whitepaper, DeepMind is publishing a tamper-proof public record of server software and results of an independent cybersecurity audit, inviting the privacy community to verify protections.