Muse, Meta’s New Personal AI Agent, Needs You to Trust It

image via WIRED

September 8, 2026, 8:12 PM

• •Muse is rolling out on iOS, Android, the web at Muse.ai, and through WhatsApp, with AI-glasses support planned.
• •Meta says the agent can autonomously handle digital tasks including email, travel booking, online browsing, and commerce actions.
• •Purchases are processed through Stripe Link using single-use card numbers rather than exposing users’ real payment details.
• •Muse runs each user session in a Secure VM intended to isolate web data and integrations from action-taking components.
• •Meta says a future Confidential VM will use trusted execution environments, local user-managed keys, audits by outside security firms, and published binaries and transparency logs.
• •Meta has expanded its bug bounty coverage to Muse, with rewards up to $300,000 for valid vulnerabilities.

Meta has launched Muse, a personal AI agent available through a dedicated app, website, WhatsApp, and eventually its AI glasses. The company says Muse can automate tasks like sending emails, booking travel, selling a car, and making purchases using Stripe infrastructure that avoids sharing a user’s real card details. Meta is pitching security and privacy as Muse’s main differentiators, using a Secure VM architecture, a monitoring system called Sentinel, and human approval prompts meant to resist prompt-injection attacks. The company also plans a more private Confidential VM model with user-managed keys and external audits, while acknowledging that current policy protections are stronger than the technical limits of today’s system.

Read original article