Meta Releases Muse, a Personal AI Agent With Privacy ‘Built Into It’

Details

Meta Superintelligence Labs — the unit Mark Zuckerberg formed roughly a year ago with aggressive researcher recruitment — released Muse on September 8, Meta's first public personal AI agent. Users can access it via a dedicated iOS/Android app, the website Muse.ai, or by direct messaging in WhatsApp. Meta AI glasses integration is on the roadmap.\n\nMuse's payment layer runs through Stripe's Link infrastructure, which issues single-use virtual card numbers each time the agent makes a purchase. This prevents real card data from being passed to third-party merchants — and Meta says Muse is the first AI agent platform covered by Link's purchase protection guarantee, including no-fee returns. Stripe's existing user base makes this integration operationally credible.\n\nThe privacy architecture uses a per-user 'Secure VM' — a virtual machine that isolates browsing sessions and untrusted web data from the agent's action-taking core. This is a substantive design choice, not just marketing language, though independent security audits have not been published.\n\nMuse enters a market where OpenClaw and Instinct already have established user bases. Meta arrives late but brings distribution advantages: WhatsApp's global reach, glasses hardware, and a free entry tier. The core trust problem remains: Meta's track record on user data handling will be a real adoption barrier regardless of architectural claims.