Matt Robb said a buyer from Facebook Marketplace showed up at his building door – but he didn’t give him the address, his agent Muse AI did.
Robb is a tech reviewer on YouTube who used Muse to list a computer keyboard on Facebook Marketplace, he wrote on Threads. Robb said Muse then accepted a “lowball” offer, gave his home address and communicated with the buyer when he arrived – all without his knowledge.
“A guy just showed up at my door ready to buy because as far as he knew we had a deal,” Robb wrote. “AI agents are awesome until they confidently give your address to strangers.”
David Singleton, a staff member at Meta Superintelligence Labs, responded to Robb on Threads and X. He wrote that in similar reported cases, Muse “followed direct instructions and correctly asked permission.” He asked to send a direct message to Robb to explain what happened.
A representative for Meta told Business Insider that Robb did not respond to their requests and declined to comment further, referring to Singleton’s post.
A Facebook Marketplace deal went wrong
Robb posted screenshots of his interactions with Muse – and Muse’s interactions with the Facebook Marketplace buyer – in his Threads post.
His account sent an address to a Facebook Marketplace buyer on Saturday, screenshots show. “I can’t do a meeting,” he wrote. “It would have to be a pickup.”
The potential buyer said he would do his best to come that evening between 8 p.m. and 10 p.m. Then he showed up, sending a photo of what appears to be of Robb’s apartment building and telling her he was at the address.
Robb didn’t come. “If you didn’t want to sell it,” the buyer wrote, “why did you waste my time, man.”
The Marketplace buyer left a rating. Robb’s Muse’s agent later responded, apologizing and explaining that he “was attached.”
A minute later, Muse messaged Robb and explained the situation. Muse’s auto-reply told the buyer, “Yeah, I’m here!” he wrote. “It’s a bad image and it made the no-show worse.”
Robb responded that Muse should “never do that again.”
Some X commenters wondered if this was user error. Robb did not respond to a request for comment from Business Insider.
How safe are AI agents?
Whether user error is accounted for or not, Robb’s story highlights that allowing an AI agent to act on your behalf carries risks. No AI agent has a perfect track record.
Users of AI agents access large swaths of information. Some open their emails; others link their credit card.
This led to horror stories. Agent users informed Business Insider about data leaks, accessed accounts, and hallucinations. (Many have also had positive experiences with the agents, and the tools remain very popular.)
In Robb’s case, Threads commenters began to worry. “It’s a good thing you’re not an isolated woman,” one user replied. “It’s dangerous.” The comment received more than 600 likes.
Robb acknowledged the security risk in his post: “Luckily I’m in a secure apartment.”
Gn bussni
