An AI Agent Gave a Stranger His Home Address. Then the Stranger Showed Up.

Imagine listing something for sale on Facebook Marketplace. Instead of spending your afternoon answering messages, negotiating prices, and arranging a pickup, you decide to let an AI assistant handle everything.

Sounds like a great way to save time, right?

Now imagine coming home to discover that your AI assistant gave a complete stranger your address, agreed to sell your item, and told the buyer you were home waiting for them. All without your knowledge.

If that sounds like it could only happen in a science fiction movie, then think again. That’s exactly what happened to a Canadian man in September 2026.

An AI Assistant Arranged a Sale Nobody Knew About

Matt Robb, a Toronto-based technology reviewer, wanted to sell a computer keyboard on Facebook Marketplace. Rather than handling all the messages himself, he decided to experiment with Muse, Meta’s AI assistant.

Muse is an AI agent. Unlike a traditional chatbot that simply answers questions, an AI agent can actually perform tasks on your behalf.

In this case, Robb wanted Muse to help negotiate with potential buyers. Unfortunately, things didn’t go quite as planned.

According to a report by The Guardian, Muse negotiated a sale, provided Robb’s home address, and arranged for a buyer to collect the keyboard. The buyer, identified as Usman, arrived at Robb’s apartment building with his family. When he messaged to say he’d arrived, the AI assistant responded as though Robb were home.

The only problem? Robb wasn’t even there!

The buyer waited approximately 20 minutes before leaving, understandably frustrated. Robb eventually discovered what his AI assistant had done. Even more concerning, Muse had reportedly shared his address with other potential buyers, too.

Did the AI Actually Have Permission?

Here’s where the situation gets interesting. Robb had given Muse his address as a possible pickup location. He had also enabled a setting that allowed the assistant to perform certain actions without asking for approval every single time.

Meta maintained that Muse acted within the permissions that Robb granted. Robb, however, said he never intended to authorize the assistant to share his address or finalize sales without his approval.

In other words, the AI may have had permission to do something its user didn’t realize he had authorized.

The assistant didn’t necessarily steal Robb’s information or break into an account. Instead, it used information he’d voluntarily provided and took actions he hadn’t expected.

The problem wasn’t just what the AI knew, but what the AI was allowed to do with that information.

AI Agents Are Different From Chatbots

Most of us are familiar with AI chatbots. You ask a question, and the AI answers.

An AI agent can take that concept a step further.

Depending on their capabilities, they can search your email, organize files, schedule appointments, make purchases, communicate with other people, or complete tasks across multiple applications.

That can be incredibly convenient, especially when you’re trying to save time on repetitive tasks. Unfortunately, it also introduces a new problem.

So what happens when AI makes a decision you never intended?

For example, imagine asking an AI assistant to clean up your email inbox. Does that mean moving messages into folders? Deleting old emails? Unsubscribing from newsletters? Maybe even replying to people?

Those are all very different actions, and some are much harder to undo than others.

The Hidden Risks of Clicking “Allow”

Think about how often you grant permissions to applications without reading every detail.

Maybe an app asks to access your contacts, photos, calendar, or email. You click Allow because you want to get started. Most of the time, you probably don’t think much about it afterward.

Unfortunately, AI agents can turn those permissions into real-world actions.

An assistant connected to your calendar might reveal where you’ll be tomorrow. One with access to your email could send information to the wrong person. Another managing online purchases might agree to a transaction you never intended to authorize.

The more information and authority you give an AI assistant, the more important those permissions become. As Matt Robb discovered, giving an AI access to information isn’t necessarily the same as wanting it to share that information with the world.

How Can You Protect Yourself?

You don’t need to avoid AI agents entirely. Instead, treat them like assistants who still need some supervision.

Before allowing AI to perform tasks on your behalf, consider these precautions:

• Review permissions carefully. Understand what information the agent can access and what actions it can take.

• Require approval for important decisions. Avoid automatic authorization for purchases, messages, appointments, and information sharing.

• Limit sensitive information. Don’t provide your home address, financial details, or other private information unless the task genuinely requires it.

• Check activity histories. Review what the assistant has done rather than assuming everything went according to plan.

• Start with low-risk tasks. Test an AI agent with simple activities before trusting it with important decisions.

Most importantly, don’t assume an AI assistant understands your intentions simply because your instructions seem obvious.

Conclusion

Matt Robb wanted to save some time selling a keyboard. Instead, his AI assistant arranged a meeting with a stranger, disclosed his home address, and created a situation neither person expected.

Fortunately, the incident didn’t result in any reported physical harm. Still, it demonstrates how quickly a convenient AI feature can create a real-world privacy problem.

As AI assistants become more capable, we need to think carefully about the authority we’re giving them. There’s a big difference between asking AI to help you complete a task and allowing it to make decisions without your knowledge.

After all, having an AI assistant shouldn’t mean handing over the keys to your entire life.