Meta Launches Muse, an AI Agent That Can Act for You

Meta’s chatbot becomes an agent

Meta announced Muse on Tuesday 8 September, describing it as a personal AI agent that can take actions on a user’s behalf rather than merely answer questions. The service can send emails, book travel, shop, complete online forms, manage schedules and work on longer-running goals, according to Meta.

Muse is initially available only in the United States, through dedicated iOS and Android apps, the web and WhatsApp. Meta has said it is coming soon to its AI glasses, but has not provided a date. There is no confirmed UK launch or UK pricing.

It can keep working after you leave

Meta says Muse can continue working after the app is closed, return when circumstances change and make proactive suggestions based on previous conversations, goals and remembered preferences. Its stated use cases range from arranging travel and follow-up emails to exercise plans and setting up a business.

“Once a person shares a goal with Muse, it helps them develop a personalized plan and coordinate their time and resources, then advances the work on its own. It can open a browser, fill out forms, and negotiate on their behalf.” — Meta

The agent runs on Muse Spark, Meta’s model for agentic work. Meta’s design documentation says Muse has its own cloud computer, file system, terminal and browser, allowing it to run multiple tasks at once.

Approval checks remain central

Muse operates in an isolated Linux-based cloud environment called Muse Secure VM, Meta says. A separate system, Sentinel, can allow, block or request approval for proposed actions involving websites and connected services.

Meta says Muse does not see users’ actual passwords or payment details, and that credentials are stored separately and inserted when required. Users can choose connected apps, set read-only or write permissions, review an audit trail, disconnect services and tell the agent to forget learned information. They can also opt out of having interactions used to train Meta’s AI models.

Shopping support at launch includes Stripe Link, which generates a one-time-use card number rather than revealing the usual card details. Meta says Shop Pay and 1Password integration are planned, but were not available at launch.

Meta acknowledges the risks

Meta says purchases and similarly sensitive actions need user approval. However, the company also acknowledges Muse can make mistakes and remains susceptible to prompt-injection attacks, where malicious instructions on a website can attempt to influence an AI system’s behaviour.

It has opened a bug-bounty programme offering rewards of up to $300,000 for valid reports, including up to $130,000 for a successful prompt-injection attack affecting one user. Meta also plans a Muse Confidential VM later in 2026, which it says would encrypt the whole virtual machine with a key held only by the user. That system is not yet available.

Pricing and the UK picture

Meta says Muse will be free for most needs, with paid plans for heavier use. TechCrunch reports US plans called Power at $20 per month and Maximum at $100 per month, while the free tier is usage-metered. UK prices, VAT treatment and billing arrangements have not been confirmed.

For now, Muse illustrates where consumer AI is heading: from drafting messages and answering prompts towards acting across real accounts and services. That may make routine administration less tedious, but it also means users will need to decide how much access an AI agent should receive before it is trusted with meaningful work.