Advancing Private AI Compute with secure, server-side memory
Google's Private AI Compute team announces a persistent memory layer that lets cloud-based AI assistants keep context across a user's devices. Data would sit in encrypted cloud storage whose keys stay only on the user's devices, and Google plans to publish its server software for verification and share an independent security audit.
What the article says
On 23 September 2026, Google DeepMind described server-side persistent memory for its Private AI Compute platform, which handles complex tasks in hardware-isolated cloud enclaves. Until now, this technology and similar solutions were stateless: all context was erased at the end of each task, which limits continuous assistance.
The new layer stores useful information in a dedicated, encrypted space whose keys are held only on the user's devices; according to Google, neither third parties nor Google can access it. Data is decrypted temporarily inside the enclave, then re-encrypted. Google publishes a tamper-proof public ledger of the server software that devices can verify, and updates a technical brief with an independent audit, without naming the auditor. The page gives no schedule and names no products or regions, and its guarantees rest on Google's claims.
Read the official article AnnuIA profile
Summary written by AnnuIA in its own words from the official page of Google DeepMind. The content and rights belong to the publisher: please read the original article.