Meta Extends Private Processing Confidential Computing to AI Glasses

Summary

Meta is extending its Private Processing confidential-computing system to AI glasses, allowing cloud-based AI models to handle demanding features while Meta says it cannot access the user data involved. Basic actions such as calls and text replies can run on the glasses, but larger models, persistent context and proactive assistance require cloud processing. Meta first introduced Private Processing for WhatsApp and the Meta AI app in 2025, and says the glasses implementation is intended for tasks including streaming transcription, contextual search and long-term recall. The design uses trusted execution environments, whose encrypted memory keys remain inside security hardware, plus remote attestation and a public transparency ledger to verify the hardware and software before data is sent. Meta lists five requirements: isolation from hosts and operators, fail-closed behavior, public verifiability, non-targetability and encrypted storage controlled by a user-provided key. Anonymous credentials and an OHTTP relay operated by Fastly or Cloudflare are intended to prevent authentication and network operators from linking requests to accounts, while TEE-based storage hides access patterns that an external encrypted database could expose. Because operators cannot inspect TEE memory or model inputs and outputs, monitoring is limited to aggregate signals such as utilization, latency and hardware failures. Meta says independent firms including NCC Group audit the design, and its security program will provide researchers with binaries and documentation. The company describes the system as groundwork for more agentic, multimodal glasses experiences. Separately, Meta announced that its Muse personal AI agent is coming to the glasses, along with Ray-Ban Meta Gen 3 from $449 and Ray-Ban Meta Audio pre-orders from $349.