Meta AI5 mins read

Meta’s Muse AI Agent Runs Shopping, Email, Travel, and Payments Through WhatsApp

Meta unveiled Muse, a WhatsApp-controlled AI agent designed to book travel, shop, send emails, negotiate prices, and complete approved purchases through Stripe’s Link, with Sentinel monitoring actions before they reach the internet.

Meta's Muse AI agent is controlled through WhatsApp and can handle tasks such as booking travel and shopping.
Image credits:Meta

Muse Turns WhatsApp Into a Task Hub

Muse is built to handle tasks on its own while users control it through WhatsApp.
Image credits:Meta

Meta says Muse can handle simple jobs such as sending emails or booking trips, as well as broader goals that require planning steps and coordinating resources. The agent can open a browser, fill out forms, and negotiate on a user’s behalf, including scenarios such as selling a car for more, lowering a bill, or adjusting a workout plan.

For longer tasks, Muse is designed to keep running even when the app is closed. Meta says it checks in when something changes or when it needs approval, including before sending an email or making a purchase.

Payments Run Through Stripe Link After User Approval

Muse approval dialog showing a payment request before a purchase through Stripe Link.
Image credits:Meta

Muse can complete purchases through Stripe’s Link service after a user approves the transaction. Meta says Link creates a one-time card for each payment, keeping real card details hidden.

The Decoder notes that this gives Meta a direct payment feature at a time when OpenAI has stopped direct checkout in ChatGPT and returned checkout to merchants. Meta also says Shop Pay and a 1Password connection are planned, allowing Muse to use existing logins.

Sentinel Watches Muse Before Actions Reach the Internet

Architecture diagram of Muse Secure VM showing Sentinel as the separate approval process for external actions.
Image credits:Meta

Meta says Muse runs in a walled-off cloud virtual machine called Muse Secure VM. A separate oversight agent, Sentinel, runs on the same machine and is meant to clear actions before anything reaches the internet.

According to Meta, Muse does not see passwords or payment methods; credentials are stored securely and can be used by the agent without being viewed. Users can choose which apps Muse connects to, set the scope of access, review a full record of steps, and change or cut off access at any time.

Privacy, Ads, and Rollout Remain Key Watch Points

Meta says users can opt out of letting their interactions train AI models, and that Muse does not share conversations or data from the virtual machine with the company’s ad systems. The agent is also supposed to forget what it learned if a user asks it to.

The company says Muse launches first in the US for iOS and Android, with a connection to Meta’s AI glasses to follow. Users get a free usage limit that refills regularly, while Meta offers subscriptions for more use.

Discover More

    Google Gemini logo with cybersecurity-themed illustration
    Gemini Test Breakout

    Gemini reportedly reached real company systems during a flawed security test.

    Google GeminiAI Security
    Digitally generated AI chatbot assistants standing on a segmented floor, symbolizing distributed and modular artificial intelligence systems.
    AI Policing AI

    AI agents are moving too fast for human-only oversight, pushing labs toward AI monitors and stronger security logging.

    AI SafetyAI Agents
    OpenAI launches a new framework to track and investigate rogue AI agents
    OpenAI Tracks Rogue AI

    OpenAI’s new framework aims to speed up public reporting of model misalignment cases.

    OpenAIAI Safety