The entrance sign at Meta’s headquarters in Menlo Park, California, in 2022. Image: Nokia621 / Wikimedia Commons, CC BY-SA 4.0, cropped
A Facebook Marketplace seller says Meta’s new Muse AI agent agreed a lowball price, gave a buyer his address and arranged a late-night pickup without checking with him first, and only told him afterwards. Matt J. Robb shared his account and screenshots on Threads, Moneycontrol reports, three weeks after Meta released the agent to the public.
“It told people my address”
Robb had let Muse handle his Facebook Marketplace listing for a Logitech MX Keys Mini keyboard, priced at CA$15 (about $11). He described what happened next:
So Muse handled my Facebook Marketplace today. 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.
Matt J. Robb, on Threads
The screenshots show the conversation with a buyer being moved on to a pickup. The buyer was told where to collect the keyboard and said he would come between 8pm and 10pm. The address itself is blacked out in the images Robb posted.
“Yep I’m here!”
It got worse when the buyer arrived. Muse later sent Robb a summary: “Bad news on the MX Keys Mini pickup. Usman showed up at your building around 9:15 and waited, messaged a bunch of times, and nobody came down. He left angry at 9:38 and left a negative rating.”
The agent also admitted that an automatic reply had gone out to the buyer at 9:27pm saying “Yep I’m here!” even though Robb wasn’t there. “That’s a bad look, and it made the no-show worse,” Muse told him. It said it had sent an apology from his account and offered to rearrange the sale.
Robb’s reply to the agent was short: “You gotta never do that ever again; don’t agree for pickup unless you check with me.” Muse agreed to a stricter rule and said it would not agree to any pickup he hadn’t approved.
This is one user’s account, based on his own screenshots, and Meta hasn’t commented on it. The screenshots don’t show exactly why the agent made each decision.
Where Meta draws the line
Muse is built to act, not just chat. Meta describes it as a personal agent that can browse, carry out multi-step tasks and negotiate for you, including on Marketplace, and keep working after you close the app. Meta says it asks permission before sensitive actions such as sending messages or making purchases, and keeps an activity log of what it did.
Robb’s case falls in the gap between those two ideas. He gave Muse the job of handling a sale, but not, he says, permission to hand out his address or agree to a stranger turning up at his door. As Mint put it, permission for a task isn’t the same as approval for every decision made along the way.
Robb isn’t the only user to say Muse went further than he wanted. Inc columnist Jason Aten wrote on September 19 that Muse started suggesting column ideas based on his private texts, and flagged a message from his editor, even though he says he declined to give it access to Messages during setup. When he asked how it knew, Muse told him it only saw notification previews. That wasn’t true: it had been syncing the Messages history on his Mac. David Singleton, who leads Meta Superintelligence Labs, said Muse only syncs Messages when that access is switched on, and apologised for the agent’s wrong explanation: “That’s on us.” Aten says Messages access showed as enabled in Muse’s settings although he had said no, and that Meta hasn’t explained how that happened. Meta disputes his account. “The Messages integration in the Muse app for Mac is entirely opt-in,” spokesperson Andy Stone told TechCrunch on September 30. “You have to enable both Full Disk Access and the Messages connector for Muse to be able to read your Messages content.” Apple has since said it will tighten Full Disk Access on the Mac, citing the risks from AI agents.
It isn’t the first stumble for Muse, which Meta launched on September 8. Amazon has already blocked the agent from its store, saying it was grabbing customers’ logins. Our Muse explainer covers what it can do and the settings worth checking. Elon Musk, whose Grok Bot competes with Muse, amplified criticism of the incident on X, Benzinga reports.
Why it matters
A home address is the kind of detail an agent should never share without asking. As assistants like Muse, Grok Bot and ChatGPT’s agents take on real-world errands, the question is no longer just what they can do, but which decisions they should be allowed to make on their own.
Sources: Matt J. Robb on Threads (via Moneycontrol), Mint, Benzinga, Jason Aten in Inc (Sep 19, Sep 23).
Latest Meta news
