Google DeepMind adds secure server-side memory to Private AI Compute

On September 23, 2026, Google DeepMind published a technical update describing a secure, server-side memory layer for Private AI Compute. Google says the design is intended to let an AI assistant retain context over time and across devices, addressing a limitation of its earlier cloud processing, which cleared context when a task ended.

In Google’s description, personal information is stored in dedicated encrypted cloud storage, while the keys needed to unlock it are held on users’ personal devices. When a model needs information to handle a request, an authenticated, end-to-end encrypted channel connects the device to an isolated cloud environment. Data is temporarily decrypted in isolated memory, used to process the request, then updated with new context and encrypted again. Google says the architecture combines hardware-enforced secure enclaves, encrypted channels and per-user databases protected by device-derived keys.

The update builds on Private AI Compute, which Google introduced on November 11, 2025, as a cloud platform for Gemini-based AI experiences with privacy protections it compared to on-device processing. Google says the earlier system did not retain context after a task ended. The new server-side memory therefore addresses a central design trade-off: enabling cloud AI to carry personal context forward while maintaining data protections.

Google also says it is publishing a tamper-proof public record of its server software and an updated technical whitepaper, alongside a technical-methods update that includes results from an independent cybersecurity audit. Those materials give outside reviewers a basis to assess the system’s privacy claims. The protections described here are Google’s account of its design, not independent verification of its effectiveness.