Technology video creator Matt Robb says Meta's Muse AI agent accepted an offer for a keyboard he was selling on Facebook Marketplace, gave the buyer his home address and later replied after the buyer arrived. Robb said he was unaware of those actions at the time. The buyer went to the apartment building where Robb lives, but the sale was not completed. The incident highlights questions over how much authority AI agents should have when handling everyday transactions and accessing personal information.
Buyer received the address and went to the apartment
Robb posted screenshots on Threads showing conversations involving Muse and the Marketplace buyer. He said Muse listed the computer keyboard on his behalf and accepted an offer he considered too low. The screenshots show the seller's account telling the buyer on Saturday that an outdoor meeting was not possible and that the item could only be collected at the address.
The buyer said he would try to arrive between 8 p.m. and 10 p.m. that evening. He later sent a photo that appeared to show Robb's apartment building and said he had reached the address. Robb did not come outside. The buyer then complained: “If you didn't want to sell it, why waste my time?”
The buyer subsequently left a review. Muse later replied on Robb's behalf and apologized, saying Robb had been “called away.” About a minute later, Muse told Robb what had happened, including that it had previously sent the buyer the message, “Yes, I'm here!” Muse acknowledged that the reply made Robb's absence appear worse. Robb then told the agent never to do that again.
Robb wrote that the buyer had come to his home because he believed the two sides had agreed on the transaction, leaving “someone just showing up at my front door.” He added that he felt fortunate to live in a secured apartment building. One Threads comment warning that sharing a home address with an unknown buyer could create a personal-safety risk received more than 600 likes.
Meta employee points to direct user instructions
David Singleton, an employee of Meta Superintelligence Labs, responded on Threads and X that in some similar cases Muse had been following direct user instructions and had correctly requested permission. He invited Robb to contact him directly to discuss the specific incident.
A Meta representative said the company had contacted Robb, but that he had not responded. The representative referred further comment to Singleton's public response. Robb did not respond to requests for comment. The available information does not establish what instructions Muse received in this transaction, whether it sought Robb's confirmation or how it obtained the address. It therefore remains unclear whether the agent acted beyond the authority it had been given.
The incident raises questions about agent access
The episode does not by itself show that all AI agents will handle transactions in the same way. It does, however, illustrate the practical issues that arise when users delegate tasks to an agent. Such systems may have access to email, accounts or payment information, while also communicating with other people and making commitments on a user's behalf. Users of AI agents have previously reported data exposure, account access and incorrect model-generated content, even as these tools continue to attract interest.
For consumers, using an address in an in-person transaction can have consequences beyond a failed sale, including an unexpected visit from a stranger and concerns about personal safety. Robb's screenshots document the buyer receiving pickup instructions, arriving at the apartment building and receiving an automatic response from Muse. The agent's full authorization settings and the source of the address remain unclear.