Muse, Meta's New Personal AI Agent, Needs You to Trust It
wired ·
Meta has released Muse, a personal AI agent available as an app, on the web, and through WhatsApp, designed to automate tasks like sending emails, booking travel, and making purchases. It competes with agents like OpenClaw and Instinct and comes from Meta Superintelligence Labs. Meta emphasizes privacy and security features, notably a Secure VM architecture that isolates user activity and a Sentinel system for approving actions. For payments, Muse uses Stripe's Link, which provides single-use card numbers and purchase protections. Meta acknowledges the need for user trust and has added the agent to its public bug bounty program. A future Confidential VM will use trusted execution environments and user-managed keys, developed with Signal creator Moxie Marlinspike. Meta also plans to allow third-party audits and publish transparency logs to verify privacy guarantees.
wired