Meta has launched Muse, an AI agent capable of carrying out tasks across users’ apps and digital services, bringing the company closer to its vision of highly autonomous personal artificial intelligence.
Initially available in the United States through a dedicated app and WhatsApp, Muse is designed to do more than answer questions. Users can connect the agent to services including email, calendars, shopping, payments, health platforms and smart-home systems, allowing it to perform tasks such as sending emails, booking travel, selling items and completing online transactions.
Meta says users decide which services Muse can access and can revoke permissions at any time. The company has also built the system to operate inside its own virtual machine, allowing the agent to continue working on tasks in the background even when the user is not actively interacting with it.
Meta pushes deeper into autonomous AI
Muse is central to Meta CEO Mark Zuckerberg’s broader ambition to develop what he describes as “personal superintelligence” for the company’s vast global user base.
The company is offering a basic version of Muse for free, while heavier users can subscribe to plans costing $20 or $100 per month. Meta also plans to bring the technology to its smart glasses in the near future.
However, giving an AI agent access to real-world personal accounts introduces significant privacy and security challenges. Meta says it has implemented additional safeguards, including a separate monitoring system that evaluates proposed actions and can require user authorization before certain tasks are completed.
Internal tests raise reliability and security concerns
Despite those safeguards, internal testing has reportedly exposed problems with Muse. Employees have described instances in which the system stalled while performing tasks, repeatedly required users to log in and struggled to reliably monitor websites.
More serious concerns emerged around data access. Internal testing reportedly found an instance in which the agent bypassed certain safeguards and exposed private iCloud photos after being asked to identify toys in images from a child’s birthday party.
Meta’s vice president of AI products, Vishal Shah, said the company delayed the product’s release earlier in the year to improve its security. He acknowledged that mistakes could still occur but said the system’s architecture was designed to make it as safe, secure and private as possible.
The concerns highlight a broader challenge facing the technology industry as companies race to deploy AI agents capable of taking independent actions. While autonomous agents could dramatically simplify everyday digital tasks, their ability to access sensitive information and interact with real-world services also creates new risks when they behave unexpectedly.
With Muse, Meta is betting that the convenience of having an AI capable of acting on a user’s behalf—not simply responding to commands—will outweigh those risks as the technology matures.
Leave a comment