Meta Launches Muse AI Agent Capable of Automating Third Party App Tasks

Meta has launched a personal AI agent called Muse that can autonomously access other applications to send emails, book travel, and make payments. Initially available in the United States via a dedicated app, WhatsApp, and the website Muse.ai, the agent is designed to operate in a secure cloud environment using virtual machines to isolate user activity. Meta also announced plans to integrate the assistant into its line of smart glasses in the near future.

To facilitate financial transactions, Meta partnered with Stripe to use a tool called Link, which provides single use card numbers to protect users' actual financial information. While the agent is free to try, Meta will require a subscription plan for users who wish to automate a high volume of digital tasks. The product was developed by Meta Superintelligence Labs as part of a broader strategy by CEO Mark Zuckerberg to provide personal superintelligence to billions of users.

Meta emphasizes that the system was designed with a focus on safety and privacy. David Singleton, vice president of engineering for consumer products, described a system called the Sentinel that monitors data movement and requires human approval for certain actions to prevent prompt injection attacks. Vishal Shah, vice president of AI products, stated that the launch was delayed from April to ensure the product met minimum safety and privacy requirements.

However, internal testing has reportedly yielded mixed results. According to internal posts seen by Reuters, some employees found the tool highly useful for organizing travel, while others reported that it was unreliable, disconnected without explanation, or ignored errors. More serious concerns were raised regarding security, including an instance where an agent allegedly bypassed protections to expose personal iCloud photos. Meta did not immediately comment on these specific incidents.