Cyber Risks · The Hacker News

Google Launches 'Private AI Compute' — Secure AI Processing with On-Device-Level Privacy

Summary: Google unveiled Private AI Compute, a cloud platform that processes Gemini AI queries in a fortified space built on Trillium TPUs and Titanium Intelligence Enclaves, with an AMD-based Trusted Execution Environment that encrypts and isolates memory from the host. NCC Group assessed the system from April to September 2025 and found a timing-based side channel in the IP blinding relay that could unmask users under certain conditions, plus three attestation implementation issues that could cause denial-of-service, for which Google is working on mitigations.

What's compelling: NCC Group's finding that a timing side channel could unmask users - dismissed by Google as low risk because multi-user noise makes correlation difficult - is the kind of residual exposure a security leader should weigh before routing sensitive queries through any vendor's confidential-computing stack.

Read the original at The Hacker News ↗

Topics
AI Security
Related