Meta Launches Personal AI Agent Muse Amid Internal Privacy and Reliability Concerns
Meta officially launched Muse, an autonomous personal AI agent designed to execute tasks like sending emails and booking travel, available initially in the US via a standalone app and WhatsApp. The tool offers a free basic tier alongside paid subscription tiers at $20 and $100 per month, with plans to expand to Meta's smart glasses. Muse represents Mark Zuckerberg's push into "personal superintelligence" as Meta attempts to diversify revenue beyond digital advertising and monetize its massive AI infrastructure investments. However, security risks inherent to autonomous agents accessing sensitive user data across third-party apps present major technical and privacy challenges. Each Muse agent runs on an isolated cloud virtual machine to execute background tasks continuously and is modeled after the open-source OpenClaw project. Internal testing revealed significant issues, including a safety bypass that exposed personal iCloud photos, unexpected app crashes, and frequent forced logouts experienced by CTO Andrew Bosworth.
## BACKGROUND
AI agents are autonomous software programs capable of using large language models to reason, plan, and execute multi-step workflows on behalf of users across various digital services. Meta modeled Muse on OpenClaw, a popular open-source AI agent framework created by Peter Steinberger that allows language models to interact directly with local applications and messaging channels. Running these agents on dedicated virtual machines isolates their operations while enabling background execution without active user supervision.