Something odd happened this week. A journalist tried out Meta’s new AI agent, Muse. He says it read his private messages — without his permission. Meta strongly denies it. The clash has kicked off a real debate about Meta Muse privacy. How much of our private life can an AI agent actually see?
So what really happened? Let us look at the facts, the claims, and what you can learn from this story.
What Is Meta Muse?
Meta launched Muse on September 8 in the United States. It is a personal AI agent. It works across your apps and takes actions for you.
Once you allow it, Muse can look at your calendar, your messages, even your files. It books meetings, drafts emails, and handles small jobs for you. Meta insists it designed Muse with safety and privacy in mind.
Agents like Muse take action. That is why they ask for deeper access to your data. Muse is not yet available in India. Meta has said it will come to the Ray-Ban Meta AI glasses in the coming months. Want background on Muse? Read our earlier post on the Meta Muse AI agent for small business.
What Happened: The Journalist’s Test
Technology columnist Jason Aten wanted to test how safe Muse really is. He installed the agent on his iPhone and his Mac mini.
On the Mac, he kept a setting called Full Disk Access turned off. This setting should block apps from reading private data such as messages and calendars. Aten believed his messages were safe.
But then something strange happened. Muse started to refer to his private conversations. Aten says the agent synced his Messages database. That database held around 187,000 lines of messages.
When Aten asked Muse how it got this data, the agent gave a surprising answer. It said it had only seen notification previews, not his actual messages.
One important note: this dispute is about Apple’s Messages app on a Mac. It is not about WhatsApp.
What Meta Says
Meta disagrees with the journalist’s claim. The company says Muse cannot read your Messages unless you do two things. First, you must turn on Full Disk Access. Second, you must turn on the Messages connector inside Muse.
Meta also says this process needs several steps. One of them is a manual confirmation in macOS Settings. In Meta’s view, the agent cannot see your messages by accident.
Meta keeps saying Muse was designed to protect user privacy. Even so, nobody has explained how the agent saw those messages in this test. His story and Meta’s story do not match. Right now, we simply do not know what really happened.
Why This Meta Muse Privacy Story Matters for Freelancers and Students
Why should you care? Because AI agents are not like normal apps. A normal app does one job. An agent does jobs for you — it writes, books, clicks, and sends. And to do all that, it needs deep access to your stuff.
In one test covered in the news report, Muse asked for a long list of permissions. The list included contacts, calendars, email, messages, notes, WhatsApp, the microphone — even full disk access.
Now picture your own phone. Freelancer? Then your chats likely hold client talks, payment details, and unfinished work. Student? Then your messages hold private talks with friends and family — things you would never want a stranger to read.
Give an agent the wrong permission, and all of that sits within its reach. More access means more risk. That is the real lesson of this story. For more tools that respect your workflow, see our guide to the best AI tools for freelancers.
Notification Previews Can Leak Your Messages
One detail in this story deserves your full attention. Muse told the journalist it had only seen notification previews.
Notification previews show part of your message on your lock screen. Any app that can see these previews can read bits of your private chats. This is a quiet risk that many people miss.
The fix is easy. Turn off message previews for your sensitive apps. One change in settings — and your lock screen stops showing private chats to anyone who glances at it.
How to Stay Safe When Using AI Agents
Do not panic — just take charge of your settings. Here is what actually helps.
Only hand over the permissions the agent truly needs. Most people tap “Allow” on everything without reading. Do not be that person. Start small, and add more later only if you must.
Every few weeks, open your phone settings and look at what each app can see. Old apps you forgot about may still read your messages. Switch those off.
Hide lock-screen previews for banking apps, work chats, anything personal. It takes ten seconds.
And keep your most private accounts away from AI tools. Banking, medical stuff, work chats — if the agent does not need them, do not connect them.
The Bigger Picture
This story is not over. Meta says it did not happen. The journalist says it did. Maybe we will learn more in the coming days. Maybe not.
Either way, the takeaway stays the same. AI agents are useful, but they run on your data. Before you say yes to every permission popup, pause for one second and ask: does it really need this?
Your data is yours. Keep it that way.