Google said in a blog post that it launched Private AI Compute, combining Gemini models in the cloud with privacy assurances comparable to on-device processing.
In the post, Google describes a secured environment that isolates user data, built on a single Google stack with custom TPUs and Titanium Intelligence Enclaves, using remote attestation and encryption to connect devices to a hardware-secured “sealed” cloud.
A technical brief says user data processed on Private AI Compute is kept within a protected execution environment, with bi-directional attestation between trusted nodes and ALTS/Noise encryption; an external auditor validated the initial system design.
Google lists early product uses: Magic Cue gets more timely suggestions on Pixel 10, and Recorder expands summarization language support, according to Pixel Help pages linked from the post.
Google’s brief adds a roadmap for more transparency and attestation mechanisms in future releases.
