Google adds private cloud memory for AI assistants

Google DeepMind says Private AI Compute will gain persistent, server-side memory for AI assistants. The idea is to let an assistant remember useful context across devices while keeping the privacy model closer to on-device processing.

Under the design, personal context is stored in encrypted cloud storage, but the keys needed to unlock it stay on the user’s devices. When an AI model needs that context, the device opens an authenticated encrypted channel to a hardware-protected cloud enclave. The enclave temporarily decrypts the data in isolated memory, handles the request, saves new context and encrypts it again.

The practical change is that cloud-scale assistants could become more continuous without relying only on local hardware or simple preference lists. Google gives examples such as resuming work across mobile, web and smart-glasses surfaces.

Google is also publishing an updated technical brief, a public record of server software and independent-audit material so devices and outside reviewers can verify the Private AI Compute stack. The important question now is adoption: the architecture is promising, but users will only feel the benefit once it is tied to real assistant features.