Meta extended Private Processing to its AI glasses. The system runs AI models inside confidential virtual machines in the cloud. Even Meta cannot access user data in these environments. Glasses verify server trust via hardware-signed remote attestation before sending data. Data remains encrypted in use, in transit and at rest. User-provided keys control access to stored data. The infrastructure spans multiple regions and is fault-tolerant. Debugging relies on aggregate signals to avoid exposing user data. Binary transparency ledgers allow external verification of deployed code. The design prevents targeting individual sessions or storage.
What changed
AI glasses processed personal data in the cloud with potential exposure to Meta and host systems.
What it unlocks
Running stateful, deeply personal AI workloads for glasses in the cloud without Meta or host systems accessing the data.
- 5 engineering requirements for Private Processing
What you need to act on it
- Meta AI Glasses
- Private Processing infrastructure
Sources