Initially available in the United States through a dedicated app and WhatsApp, Muse can connect to services including email, calendars, shopping, payments, health and smart-home devices. It can continue working in the background after a user closes the app.
Meta says users remain in control of which services Muse can access and what it is allowed to do. The system can also ask for approval before sensitive actions such as sending an email or making a purchase.
The company is offering a free version of Muse, along with subscription plans costing $20 and $100 a month for users who want heavier usage.
Built to act, not just answer
Muse runs inside its own virtual machine, which Meta describes as a dedicated computer in the cloud. The setup allows the agent to open a browser, fill out forms and carry out tasks on a user’s behalf.
Read also : Netflix, Spotify, Google and OpenAI start collecting Morocco’s 20% VAT
It can also work on longer-term requests. Meta says users can give Muse a goal and have it develop a plan, coordinate resources and continue working until it needs input or approval.
For online purchases, Muse can use Stripe’s Link wallet, which generates a one-time card so users’ actual payment details are not exposed to the agent. Meta also says support for Shop Pay and 1Password is coming.
The company has built a separate system called Sentinel to monitor Muse’s actions. According to Meta, Muse cannot access passwords or payment details directly, while users can review an audit trail of what the agent has done and change or revoke its permissions.
Meta also says users can opt out of having their interactions with Muse used to train its AI models. An encrypted version, called Muse Confidential VM, is planned for later this year.
Safety concerns remain
The launch comes after Meta delayed the product in April to improve its security.
Vishal Shah, Meta’s vice president of AI products, told Reuters the company believed the additional work had brought Muse to the minimum safety standard needed for release. He acknowledged that mistakes could still happen.
Internal testing seen by Reuters, however, showed that the system was still producing problems.
One Meta employee reported that Muse had become useful enough during a three-week honeymoon in Indonesia to feel like a “third participant.” Others reported more serious issues, including an incident in which Muse exposed private iCloud photos after being asked to identify toys in images.
Read also : UN sets November 27 as International Day to Eliminate Child Marriage
Another employee said the agent had repeatedly failed while monitoring websites for items that sell out quickly, including stopping page refreshes and disabling monitoring without explanation. Meta did not respond to Reuters’ questions about those specific incidents.
The problems highlight the difficulty of giving AI systems access to real personal accounts. An agent that can send messages, access private information and spend money has far more room to cause damage than a chatbot that only generates text.
Meta is betting that its security controls can make that kind of access practical for everyday users.
Muse is currently available in the US on iOS, Android and the web. Meta plans to bring it to its smart glasses as well.
