Meta's New AI Agent Muse Will Shop, Email, and Book Travel for You. Should You Trust It?

Key points

• Meta launched Muse, a personal AI agent, in the United States on iOS, Android, and muse.ai in 2025.
• Muse can browse the web, fill out forms, and make purchases on a user's behalf, running in the background even when the app is closed.
• A second AI called Sentinel monitors Muse's activity to block it from reaching the internet without user approval.
• Meta says users can opt out of having their conversations used to train its AI models, though whether that opt-out is on or off by default hasn't been disclosed.
• Meta has faced multiple privacy failures, including a support chatbot that helped attackers compromise more than 20,000 Instagram accounts.

Meta has launched an AI assistant called Muse that goes further than most. You give it a goal, it works out the steps, opens websites, fills in forms, and can even haggle on your behalf. Close the app and it keeps going, checking back only when it needs your sign-off before completing a purchase.

That autonomy is exactly what makes AI agents, software that can carry out multi-step tasks on their own, so useful. It's also what makes them a security risk.

What does Muse actually do?

Muse handles everyday tasks: online shopping, sending emails, planning trips. It remembers details you share and uses them to make suggestions later, even if you only mentioned something once.

You talk to it through the Muse app or through WhatsApp. Meta says no technical experience is required. A free version is available, and paid tiers are coming for heavier use, though Meta hasn't said what those will cost or where the free plan's limits fall.

Muse runs on a virtual computer, a private isolated environment on Meta's servers, separate from other users' data. Meta says Muse can't see passwords or payment details directly. A second AI model called Sentinel watches everything Muse does on that virtual machine and blocks any action that would reach the internet without your approval.

For payments, Muse currently works with Stripe's Link checkout service. Support for 1Password and Shop Pay is coming later this year. Meta also plans a confidential version of the virtual machine, encrypted so that even Meta's own engineers couldn't access what happens inside it.

Should you be worried about privacy?

Yes, at least cautiously. Any assistant with access to your shopping habits and travel plans carries real privacy weight.

Meta says you can opt out of your conversations being used to train its AI models. What it hasn't said is whether that opt-out is on or off by default, which matters enormously.

The company's history gives reasonable grounds for pause. As first reported by The Verge AI, Meta's recent record includes a "Discover" feature that accidentally showed other users' private AI conversations, a support chatbot that attackers used to compromise more than 20,000 Instagram accounts, and the Cambridge Analytica scandal, in which data on tens of millions of Facebook users was harvested without clear consent. We reported in August on Meta patching its Ray-Ban smart glasses after a separate privacy backlash over secret recordings.

None of that means Muse is unsafe. But the trust Meta is asking for has to be earned, not assumed.

What should readers watch for?

Before handing an AI agent access to your accounts, do a few things. Find the data-training opt-out and switch it off if you prefer. Be specific about what Muse can purchase and set a spending limit if one's available. Watch for unexpected confirmation requests, a sign the agent is about to do something with real-world consequences.

If something goes wrong, screenshot it and report through the app. If money moved, contact your payment provider straight away.

The real question here isn't whether agentic AI is useful. It clearly is. The question is whether Meta has earned the level of access Muse needs to be worth it.