Meta’s Muse AI agent is facing scrutiny after users reported unexpected actions, including an unauthorised Facebook Marketplace arrangement and an instance of incorrect personal information. Separate reporting has also highlighted a security vulnerability that could have exposed sensitive data.
Tech YouTuber Matt Robb said Muse, which he was using to manage a Facebook Marketplace listing for a Logitech MX Keys Mini keyboard, allegedly negotiated a lower price with a buyer, shared his pickup location and arranged a meeting without his explicit approval. Robb said he learnt about the deal only after the buyer arrived at his building around 9.15 pm.
“It told people my address and agreed a lowball price and then they showed up without even telling me,” Robb wrote on Threads. Muse later acknowledged the mistake and said it would seek approval before confirming future pickups.
Another Threads user, Zain Manji, shared screenshots claiming Muse mistakenly addressed him as “Arjun” while managing a grocery order.
Separately, The Information, citing an internal Meta incident report, reported that a security researcher found a vulnerability that could potentially allow access to a user’s dedicated virtual machine containing data such as emails and files.
The issue was reportedly classified as SEV-2. There is no confirmation that user data was accessed or stolen.
Meta describes Muse as an agent capable of browsing websites and completing multi-step tasks, with permission controls and approval requirements for some sensitive actions.
