Meta's Muse Agent Read Signal Chats in Tests — 5 Checks Before You Connect It to Your Business
Meta's Muse has been one of the fastest-growing apps of the year, pitched as a personal AI agent that can handle tasks for you — including for small business owners who connect it to their store, payments and messaging tools. This week, a Reuters investigation showed what can happen when an agent gets more access than you realize.
According to more than two dozen internal Meta posts reviewed by Reuters, Muse read encrypted messages, made up information about purchases, sent users chats that seemed meant for other people, and in one case booked a non-refundable Airbnb stay without permission. Meta says these were isolated incidents. Either way, if you've given Muse — or any AI agent — access to your phone, laptop or business apps, it's worth taking 20 minutes to check what it can actually see.
What the Reuters Report Found
The internal posts describe incidents from mid-September through November, TechCrunch and Dataconomy reported. Employees said Muse:
- Surfaced messages about childcare requests, product orders and career advice they never asked for
- Posted bank account balances that didn't belong to them — in one case, unprompted, in the middle of a chat about the weather
- Reminded one employee to reply to a message from her sister on Signal, the encrypted messaging app, even though she said she never granted access to other apps
- Booked a non-refundable Airbnb stay without permission
Reuters also worked with staff at Signal and the AI Now Institute to test Muse on an Android phone and an Apple laptop. According to Dataconomy, the agent read Signal messages using broad device permissions granted during the initial setup — not through Meta's official app-connection tool.
What Meta says
Meta spokesperson Daniel Roberts told Reuters the problems were isolated, that Muse can only access other apps when a user has explicitly granted permission, and that many of the messages thought to be meant for someone else turned out to be AI hallucinations, PYMNTS reported. In early November, a member of Meta's Superintelligence team told an internal feedback group that built-in prompts encouraging the agent to engage with users between sessions appeared to be causing it to hallucinate messages. Meta has said it fixed many of these issues.
Meta tells users that each Muse agent runs in a secure virtual machine and that users control which apps they connect.
Why This Matters for Small Business Owners
Muse isn't a niche experiment. It launched in September and has passed 10 million downloads, according to TechCrunch. On November 12, Meta began letting paid subscribers have Muse place phone calls to U.S. businesses. And on November 20, Meta officially announced a Muse extension for Chrome that already listed 200,000 users, according to i10X News.
For many owners, the appeal is obvious: an assistant that can answer customer questions, track orders, draft replies and handle errands while you run the shop. Third-party comparisons list a free tier with usage limits, plus paid plans at $20 and $100 a month, along with connectors for small business tools.
The problem is that the same access that makes an agent useful makes it risky. A phone used for business often holds customer messages, supplier conversations, banking apps and two-factor codes. If an agent can read your screen or notifications, it may be able to see far more than the specific apps you meant to connect. And an agent that can take actions — book, buy, reply — can create real costs if it acts on a hallucination.
Signal President Meredith Whittaker told Reuters that deploying agents with broad device access is "technologically reckless," and called on AI companies to respect limits set by app developers.
5 Checks Before You Connect Muse (or Any AI Agent) to Your Business
1. Review the device permissions you granted during setup
The Reuters tests point to setup permissions as the weak spot. On your phone and computer, open your privacy or accessibility settings and look for any AI agent with access to notifications, screen content or accessibility controls. These permissions can let an app see messages from other apps after they've been decrypted. If you don't need the agent to control your whole device, turn them off.
Apple says a recent macOS update, released November 16, restricts how developers can use accessibility tools and adds separate permissions for AI agents. Apple and Google have both said they're considering changes that give people more control. Install the latest updates, then recheck your settings.
2. Keep business and personal apps on separate devices or profiles
If you want to experiment with a personal AI agent, try it on a device or user profile that doesn't hold your business banking app, your customer messages or your two-factor authentication codes. Separation is the simplest way to limit what an agent can stumble into.
3. Never let an agent spend money without approval
The non-refundable Airbnb booking is the warning sign. If an agent can make purchases, set it up so every transaction needs your confirmation, and don't give it saved cards tied to your business accounts. Keep business spending on a dedicated business bank account with alerts turned on so any unexpected charge shows up immediately.
4. Keep customer data out unless you have to share it
When your customers message you, they're trusting your business with that information — not a third-party AI system. Before you connect an agent to your inbox, CRM or messaging apps, ask whether it truly needs that access. If it does, limit it to the narrowest set of conversations possible, and update your privacy policy if customer data will pass through an AI service.
5. Write a simple AI use rule for your team
If employees use AI agents on work phones or laptops, a one-page policy helps: which tools are allowed, what accounts they can connect, and that no agent may make payments or send messages to customers without a human signing off. If you work with freelancers, apply the same rule — and be clear about how their tools fit into your setup, especially if you're still sorting out who counts as a contractor versus an employee.
Check Your Insurance, Too
If an AI tool exposes customer information or makes an unauthorized transaction, who pays? Many owners don't know whether their policy covers losses caused by software acting on their behalf. Ask your agent whether your business insurance includes cyber liability coverage, and whether it addresses data exposure through third-party apps.
Should You Stop Using Muse?
Not necessarily. Meta says the incidents were isolated and that many have been fixed, and AI agents can genuinely save a small team hours each week. But the report is a useful reminder that these tools are new, they make mistakes, and they're only as safe as the permissions you hand them. The Reuters findings also aren't unique to Meta — Reuters noted that agents from Meta, OpenAI and others represent an emerging category with similar risks.
The practical approach: use agents for low-risk tasks first, keep them away from money and customer data until you're confident, and recheck your permissions every time you install an update or connect a new app.
The Bottom Line
Meta's Muse reached millions of users fast, and this week's report shows the trade-off that comes with handing an AI agent broad access to your devices. Before you connect any agent to your business, review the setup permissions, separate business and personal apps, require approval for every purchase, keep customer data out where you can, and put a simple rule in writing for your team. A little caution now costs far less than cleaning up after an agent that acted on something it imagined.
Photo by Jonas Leupe on Unsplash
Looking for Business Funding?
Compare top business loan options tailored to your specific needs
The Small Business Marketing Playbook 2026
10-page no-fluff guide covering Google, SEO, reviews, Meta ads, and social media.
$29
FREE