Google introduces secure cloud memory for private AI assistance

Google has unveiled a new technical update to its Private AI Compute platform, introducing secure, server-side memory to enable persistent, cross-device AI assistance while maintaining strict privacy standards. The update allows for a secure digital vault in the cloud, where encrypted data is stored and decrypted only in isolated, authenticated environments. Cryptographic keys remain on the user’s device, ensuring data remains inaccessible to third parties, including Google. This architecture combines hardware-enforced secure enclaves, encrypted channels, and device-derived encryption keys to protect user data. The development addresses the challenge of providing continuous AI assistance across devices without compromising privacy, as traditional stateless systems wipe context after each task. Google also published a tamper-proof public record of its server software and shared results from an independent cybersecurity audit to bolster transparency. The project was co-developed by Google DeepMind, Platforms & Devices, Core, and Cloud teams, with executive sponsorship from Four Flynn, Jay Yagnik, and David Kleidermacher. The new persistent memory layer functions like a secure enclave, temporarily decrypting data to handle requests and immediately re-encrypting it. This enables seamless experiences such as resuming complex conversations between devices or accessing previously viewed content across platforms. Google invites the privacy community to review its technical brief, system architecture, and verification protocols, emphasizing that privacy is central to the design of its AI assistance systems.