Google Unveils Secure AI Compute System for Cloud-Based Data Processing

Google has unveiled Private AI Compute, a system that allows its most powerful AI models to handle personal data in the cloud while maintaining strict security controls. This means that users don't have to worry about their sensitive information being accessed by others, including Google itself. The system uses a layered security architecture to process sensitive information, which is similar to on-device computation but with the power of the cloud.

The security architecture relies on custom Tensor Processing Units and AMD-based Trusted Execution Environments, wrapped inside what Google calls Titanium Intelligence Enclaves. This creates a 'bank vault within a bank vault' scenario, where data being processed is isolated from the broader cloud infrastructure. Three key mechanisms keep things locked down: remote attestation verifies the computing environment's integrity before data arrives, end-to-end encryption protects information in transit and at rest, and ephemeral processing ensures that data is deleted after each request completes.

NCC Group, an independent security consultancy, was brought in to validate the system. The assessment covered the architecture, cryptography, and underlying code from spring through fall 2025. The first features leveraging Private AI Compute are rolling out on Pixel 10 devices, including Magic Cue, a suggestion engine that surfaces contextually relevant actions, and the Recorder app, which gets an upgrade with expanded transcription capabilities.