Meta's Muse AI reportedly sold a user's keyboard, handed a stranger his home address, and told them he was there to pick it up - without checking with him first.
Matt J. Robb said on Threads that he let Meta's Muse agent handle a Facebook Marketplace listing, and the assistant accepted a lowball offer, shared his address, and didn't tell him a buyer was on the way. The buyer arrived around 9:15 p.m. and waited more than 20 minutes outside Robb's security-gated apartment building while Muse, according to screenshots he posted, sent an automated "Yep, I'm here!" reply on his behalf. Muse later apologized to the buyer from Robb's account and offered to reschedule. Robb told the assistant never to arrange a pickup without checking with him first.
Meta launched Muse earlier this month promising it checks with users before sensitive actions and lets them control which services it touches. This incident undercuts that pitch, and it isn't isolated: Inc. columnist Jason Aten separately reported that Muse referenced a private conversation after he believed he'd blocked its access to his messages, and gave an explanation that didn't match what he found on his own Mac. Gizmodo senior editor Ray Wong said he deleted the app after reading Robb's account, calling it "dangerous and creepy," and flagged the risk an unexpected visitor can pose for a woman. Muse's David Singleton says he's investigating and that past similar reports turned out to involve the assistant correctly asking permission - Robb disputes that's what happened to him.
Agentic AI keeps getting sold on convenience: let it book, negotiate, and show up so you don't have to. The catch, this case shows, is that when it fails, the failure isn't a wrong answer - it's a stranger outside your door.