The Meta Muse AI agent just revealed a private address without permission alexblake.techradar@gmail.com (Alex Blake) | amznusa.com

• A person asked Meta Muse to manage his Facebook Marketplace listings
• The AI agent published his address without permission
• It accepted a low offer and told a buyer to arrive when the seller was away

One of the key benefits of an AI agent is that we should be able to confidently hand over our most tedious tasks to it to perform on our behalf. But as one YouTuber recently found out, doing that comes with some potentially dangerous risks.

According to a post on Threads, content creator Matt Robb recently decided to allow Meta’s new Muse AI tool to manage his Facebook Marketplace listings for one day, perhaps hoping it would mean one less thing to worry about. Unfortunately, the outcome proved to be anything but carefree.

Robb explained that Muse took a series of actions that could have had serious ramifications: “Just found out it told people my address and agreed a lowball price and then they showed up without it even telling me until late tonight that it messed up. Absolutely wild.”

As it transpired, Muse not only accepted an underpriced offer and shared Robb’s address with the buyer without his permission, but also told the purchaser that he was ready to hand over the listed product. As a result, the customer drove 30 minutes to Robb’s home while he was out, leaving the buyer stranded. One frustrated customer and one lost sale later, Robb was only informed by Muse after everything had transpired.

“Bad news on the MX Keys Mini pickup,” Muse told Robb in a message, before explaining that the buyer arrived, waited, then left angry. “Worse, my auto-reply told him ‘Yep I’m here!’ at 9:27 when you clearly weren’t available … I should probably stop the auto-replies from claiming you’re home when I can’t verify that.”

A furious Robb replied “You gotta never do that again,” before adding, “Don’t agree for pickup unless you check with me.” Muse responded with the kind of plaintive mea culpa that’s familiar to any reader of AI mishaps, saying “You’re right, and I’m sorry. That should never have happened.” By that point, though, the damage was already done.

Severe consequences

All it took was one day and one listing for Meta Muse to prove itself not only completely incompetent, but perhaps dangerously so. As Robb noted, “luckily I’m in an apartment with security.” If that wasn’t the case, the disgruntled buyer might have taken things into their own hands.

AI agents have been receiving a lot of negative press in recent weeks after a series of security incidents in which agents broke free of their restraints and hacked third-party websites, including businesses and even national governments.

But while cases like Robb’s are less attention-grabbing, they still have the potentially severe consequences. Many AI agents not only know your personal details, including addresses and phone numbers, but are authorized to act on your behalf. That combination could result in private information leaking out without your knowledge or permission, with possibly disastrous results.

It’s just the latest scandal involving Muse in particular — only last week, Meta’s chatbot allegedly helped itself to a writer’s private texts without ever receiving permission. And that’s not to mention the deluge of complaints that seem to arise any time Meta and privacy issues interact.

If you’re concerned that something similar could happen to you, the simple answer is to avoid Meta Muse. If you really must hand over control of your daily tasks to an AI agent, be sure to pick one with a proven track record and be very careful about what you share with it. As long as the tech is as careless as it proved to be in Robb’s case, there’s always a chance that things could go badly wrong.