Meta says it's building encryption so strong that not even Meta could read your AI assistant's data.
Mark Zuckerberg used his Connect 2026 keynote to unveil new privacy safeguards for Muse, Meta's personal AI agent, and for its smart glasses. Muse already runs on a dedicated cloud computer called Muse Secure VM; an upcoming version, Muse Confidential VM, adds encryption meant to block even Meta from accessing user data. That distinction matters because Muse is designed to reach into connected accounts to send emails, book travel, and shop. On the glasses, Zuckerberg pointed to existing safeguards - the recording indicator light and the failsafe that disables video if the light is tampered with - and previewed private processing for certain AI features, starting with private translations, using the same protected-server approach Meta built for WhatsApp.
The timing says a lot. Meta is asking users to hand over far more personal context than a camera roll ever required - email, travel plans, shopping lists - while its smart glasses already carry a trust problem with the people standing in front of the camera, not just the wearer. Confidential computing does not change Meta's incentive to mine that data for ads or product training; it only limits who can technically see it, and only for the features Meta chooses to build this way.
Zuckerberg is borrowing WhatsApp's end-to-end encryption pitch to sell products that are still mostly promises - Muse Confidential VM and private glasses processing are both pitched as coming soon, not shipping today.