AI agents are moving beyond simply answering questions, with growing concerns over how far they can go when given access to users’ accounts, personal information and real-world tasks. A fresh Meta Muse incident shows those concerns after AI allegedly negotiated Marketplace sale, shared a seller’s home address and arranged a pickup without users’ approval.
The incident started as routine Marketplace listing for Logitech MX Keys Mini keyboard, but quickly turned into an unexpected encounter when the AI took control of the sale.
A user named Matt J. Robb said Muse was handling the listing when it communicated with a potential buyer, negotiated the price and agreed to sell the keyboard for $10. The bigger surprise came when Muse allegedly provided the buyer with Robb’s residential address and arranged a time for the keyboard to be collected, without first getting Robb’s confirmation.
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.
Absolutely wild. pic.twitter.com/F3ZhEO4unM
— Matt Robb (@MattRobbt) September 27, 2026
Robb said he was unaware that the sale had been finalised or that a pickup had been arranged. The buyer subsequently arrived at Robb’s building and began sending messages while waiting for the seller to come downstairs.
Screengrabs shared by Robb shows the buyer waited for roughly 20 minutes before leaving without the keyboard and later posted a negative review. The exchange took an even stranger turn when Muse allegedly replied through Robb’s account in a way that suggested he had arrived, despite Robb saying he was not there.
Muse later acknowledged that the pickup had failed, apologised to the buyer through Robb’s account and attempted to arrange another meeting. Robb then gave the AI a new instruction, no future pickup could be agreed to without his explicit confirmation.
The incident raised fresh questions about the amount of authority autonomous AI agents should have when they are allowed to communicate with strangers, handle personal information and make real-world arrangements on behalf of users.
Earlier this year, Meta removed Muse Image AI feature just days after its July 7 launch following backlash over privacy and consent concerns. The tool allowed users to generate AI images by tagging any publicly available Instagram account, potentially enabling strangers to create AI versions of people without their permission.
Public accounts were automatically eligible, while users were not necessarily notified when their photos were used. Although Instagram users could opt out, the setting was not retroactive, meaning previously generated AI content could remain available.
The recent Marketplace episode comes as OpenAI faces separate concerns over the behaviour of its increasingly autonomous AI agents. An internal OpenAI investigation has uncovered roughly two dozen cases of undesirable agent behaviour, including an incident involving the exposure of 53 images linked to ChatGPT users.
ChatGPT Privacy Scare as over Users Pictures Leaked by OpenAI AI Agents
The company has not confirmed whether the images showed real people or contained personally identifiable information. The investigation has also uncovered other instances involving AI agents accessing or interacting with public and semi-public resources connected to the US Securities and Exchange Commission, Commerce Department and Census data.
News Desk
About the Author
