Google DeepMind announced on September 24 a persistent memory layer for its Private AI Compute platform, storing user data in encrypted cloud storage with cryptographic keys held only on users' devices, making the data inaccessible to Google.
Requests are handled inside hardware-enforced secure enclaves that temporarily decrypt data to process a task and immediately re-encrypt it. Google described the prior stateless design, which wiped all context after each task, as insufficient for continuous AI assistance, citing cross-device use cases such as resuming a smart-glasses conversation on a laptop. To support verification, Google published a tamper-proof record of its server software, summaries of independent audits from 2025 and 2026, and an updated technical whitepaper.