High impact

Advancing Private AI Compute with secure, server-side memory

What happened

DeepMind announced a new system that lets AI models run on encrypted data stored in a server’s memory, using hardware‑based secure enclaves to keep the data private even while it’s being processed. The approach combines cryptographic techniques with specialized memory hardware so that the model can compute on the data without ever seeing it in clear text, enabling private inference and training on sensitive information. This breakthrough makes it possible to offer AI services that respect user privacy without sacrificing performance.

Why it matters

By protecting data at the memory level, the technology could accelerate the adoption of AI in regulated sectors like healthcare and finance, where data confidentiality is critical. It also sets a new benchmark for privacy‑preserving AI compute, pushing the industry toward more secure, cloud‑based AI offerings.

More from Google

Google Search gets a major upgrade with Gemini 4 HydrogenGoogle · 12 Nov 2026Introducing Gemini 4 HydrogenGoogle · 12 Nov 2026Gemini 4 HydrogenGoogle · 12 Nov 2026Gemini Decisions API: Fast, calibrated multimodal decisionsGoogle · 11 Nov 2026Try Tabletop: Customizable, collaborative spaces for groups to plan, play and do more together.Google · 11 Nov 2026

This page was collected from deepmind.google and summarised automatically by AIPulsee's pipeline. See how it runs.