Meta’s new AI assistant is asking testers to connect their bank accounts

Meta is quietly testing an AI agent capable of managing calendars, sorting emails and tracking spending directly through a person’s bank accounts. The assistant, called Muse, has moved into a closed alpha stage with a small group of testers, and the scope of access it is requesting has drawn immediate scrutiny.

Muse’s closed alpha and the waitlist system

According to a report from Testing Catalog, Muse has graduated from internal testing at Meta into a limited external phase. The company plans to onboard new users gradually through a waitlist or invite codes rather than opening the app broadly. The exact stage of testing is not entirely clear from available reporting, which at different points describes the rollout as both a closed alpha and a closed beta, suggesting Meta itself may be treating the labels loosely as it works out how broadly to expand access. An App Store listing for “Muse from Meta” already includes screenshots and a detailed feature description, pointing to a product that is further along in development than the closed-testing label might suggest. Those screenshots show features such as proactive task suggestions, a built-in approval step before the agent sends a message or spends money, and price tracking for tickets and reservations.

The permissions Muse is built to request

Muse is designed to connect with a wide range of everyday services rather than operating as a standalone chatbot, according to a report from Android Authority. Reported connectors include Gmail and Google Calendar, along with health and fitness apps such as Peloton and HealthEx. The more sensitive category involves financial access: Muse could link directly to a user’s card providers and bank accounts, track expenses, and flag unused subscriptions worth canceling. It is also possible the agent will rely on bank and card statements that arrive by email rather than a direct account connection, though the reporting leaves that distinction unresolved. Either approach would give the assistant visibility into some of the most sensitive data a person generates.

Why this follows a broader pattern at Meta

Muse is not Meta’s first attempt to compete in the AI assistant space this year. The company previously introduced Muse Spark as a challenger to Google’s Gemini and released a local AI model, referred to internally as Glimmer, last month, part of a steady cadence of AI product rollouts. Positioning Muse as an agent that manages a user’s calendar, inbox, tasks and finances places it in the same category as other emerging “agentic” assistants from major tech companies, which are increasingly designed to take action on a person’s behalf rather than simply answer questions. That trend has accelerated across the industry over the past year, with several large technology companies racing to ship assistants that can book appointments, manage subscriptions and handle routine correspondence without step-by-step human direction for each task.

The trust question Meta has to answer

Handing an AI system standing access to a bank account is a materially different request than granting it access to a calendar or an inbox, and Meta’s history with user privacy adds weight to that distinction. The company has faced repeated scrutiny over data practices in the past, and asking testers to link financial accounts to a new, unproven agent raises the stakes for how carefully that access is scoped and audited. The built-in approval step shown in the App Store screenshots, which reportedly requires user confirmation before Muse sends a message or completes a purchase, suggests Meta is aware of the sensitivity and has tried to build in a checkpoint before the agent can act autonomously with money.

What happens next for testers and the public rollout

Because Muse remains in a closed and invite-limited phase, most people will not encounter a request to connect a bank account through the app anytime soon. The bigger question is what Meta learns from this early testing group about how comfortable users are with an AI agent holding that level of access, and whether the eventual public version scales back the financial connectivity or leans further into it. How Muse handles safeguards during this alpha period is likely to shape both its design before a wider release and the level of scrutiny regulators and privacy advocates apply once it reaches more users.

For anyone who does receive an invite code during this testing window, the practical decision comes down to weighing convenience against exposure: an assistant that can genuinely track spending, cancel forgotten subscriptions and manage a calendar in one place is a meaningfully different proposition than a chatbot confined to answering questions, but it also means trusting a single company with a more complete picture of daily financial life than most people hand over voluntarily. Meta has not published a detailed account of how it secures the financial data Muse would access, what it retains, or how long testers’ connected accounts remain linked once the alpha period ends, leaving those specifics as open questions heading into any wider release.

This article was created with the assistance of AI and reviewed by an editor.

More from Morning Overview