Meta has launched Muse, a personal artificial intelligence agent designed to go beyond answering questions and autonomously carry out tasks on behalf of users. The agent can access connected applications to perform activities such as sending emails, booking travel, selling items and making purchases, marking a significant expansion of Meta’s efforts to bring agentic AI into everyday digital activities.
The company said Muse is initially available only in the United States, where users can access it through a dedicated Muse application or via WhatsApp. Meta also plans to bring the agent to its smart glasses soon, although it has not provided a specific timeline for that expansion. A basic version of Muse is available for free, while subscription plans are priced at $20 per month and $100 per month for users with heavier usage requirements.
Muse has been developed as part of Meta CEO Mark Zuckerberg’s vision of “personal superintelligence”, with the company seeking to create AI systems capable of taking on more work for users rather than simply responding to prompts. Internally known as Hatch, Muse can work across several categories of applications, including email, calendars, payments, health, shopping and smart-home services. Users decide which applications Muse can connect to and can revoke access whenever they choose.
Unlike conventional AI assistants that generally require users to provide instructions for every step, Muse is designed to continue working independently after receiving a task or broader objective. It can open a browser, fill out forms and negotiate on a user’s behalf. For longer-running tasks, it can continue working in the background even after the user closes the application and return when something changes or when it requires approval for an action.
The agent is powered by Muse Spark, which Meta describes as its most capable model to date for real-world agentic work. The company has also built Muse around a dedicated Muse Secure VM, a cloud-based virtual machine that houses the agent and a user’s connected data and credentials. A separate system called Sentinel monitors Muse’s planned actions and can require user permission before certain actions are completed.
Payments are another important part of Muse’s functionality. Meta said users can complete purchases through Link, built by Stripe, while Link’s wallet for agents can generate a one-time-use card so that users’ actual card details remain hidden during online transactions. Meta said Shop Pay and 1Password support are also coming. The company said Muse is the first AI agent covered by Link’s purchase protections, which include coverage for eligible issues such as damaged or lost items, price drops, no-fee returns and a return guarantee.
Privacy and security have been a major focus of the launch because Muse can work with sensitive personal information. Meta said users can control which applications the agent accesses and the level of permissions provided. The company also said Muse does not have visibility into users’ passwords or payment methods, while connected credentials are stored securely. Users receive an audit trail of actions taken and can change permissions or disconnect services at any time. They can also opt out of having their interactions used to train Meta’s AI models.
However, the launch comes as Meta continues to face questions about the reliability and safety of autonomous AI systems. Reuters reported that Meta employees testing Muse had recently raised concerns about security and reliability. Internal tests reportedly included an incident in which the agent was able to expose personal iCloud photos after being prompted to identify toys in images from a child’s birthday party. Employees also reported login problems and failures while using Muse to monitor items that could sell out quickly. Meta did not comment specifically on those incidents.
Meta said it had originally delayed the release in April to strengthen the product’s security. Vishal Shah, Vice President of AI Products at Meta, told Reuters that the additional work helped the company reach the minimum safety threshold required to make Muse available to users. Meta has also said it will introduce Muse Confidential VM later this year, with the entire virtual machine, including user data and conversations, encrypted using a key held only by the user.
The launch represents another step in Meta’s broader push to turn AI from a conversational technology into an active digital assistant capable of completing tasks. For Meta, Muse is also part of a larger strategy around AI infrastructure and new revenue opportunities as the company invests heavily in AI chips and related infrastructure. The company expects those investments to exceed $130 billion this year, according to the Reuters report.
With Muse, Meta is attempting to make autonomous AI accessible to mainstream consumers by combining task execution, application access, persistent background activity and personalised assistance within a single system. The initial US rollout will give the company an opportunity to assess how consumers respond to an AI agent that is capable of taking real-world actions while handling increasingly sensitive personal information.
