Muse messes up, makes Facebook deal and shares YouTuber's address with strangers, then apologises

Meta’s Muse AI agent is going viral. Anybody who has checked on tech news and social media recently would know Mark Zuckerberg and Co., are finally onto something big. Muse is topping app store charts and winning hearts for its simplicity and ease of use. As usage soars, users are finding out use cases that best suit their needs. They are also finding potential issues. It is AI after all, and AI is prone to make mistakes from time to time.Two cases of Muse messing up are making headlines right now. One of them involves tech YouTuber Matt Robb who claims Muse successfully negotiated a sales deal with someone on Facebook Marketplace, then shared his address with them for product pickup as well, without taking explicit approval from Robb. To be clear, Muse is an AI agent, so technically it is designed to work autonomously on a user’s behalf. We do not know the extent of permissions Muse was given but it is safe to assume that it had access to his Facebook Marketplace account at the very least.According to Robb, who has shared his experience on Instagram Threads, he was using Muse to help manage his Facebook Marketplace listing for a Logitech MX Keys Mini keyboard. However, without his knowledge or approval, the AI agent apparently negotiated a lower price with a buyer, and even shared his pickup location and arranged a collection. Robb said he only learnt about the arrangement after the buyer had already turned up at his building.The buyer reportedly arrived at around 9.15 pm, waited for the seller and eventually left, giving him a negative rating. “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,” Robb posted on Threads. The screenshots shared by Robb also show Muse acknowledging that it had sent an automated message to the buyer saying, “Yep I’m here!” at 9.27 pm, even though Robb was unavailable. The assistant later apologised and offered to try the transaction again another day.Robb then instructed the AI not to agree to any future pickups without checking with him first. Muse acknowledged the mistake and said it would seek explicit approval before confirming a specific pickup.Another user claims Muse got his name wrong Meanwhile, another Threads user, named Zain Manji, has shared screenshots claiming that the AI agent unexpectedly addressed him as “Arjun” while helping him manage his weekly grocery order.When Manji questioned the AI agent, Muse initially blamed an error in an earlier instruction. It later claimed to have checked its records and found no previous mention of the name Arjun, apart from the correction it had made after the mistake.Both incidents are now going viral on social media, with users questioning how much autonomy AI agents should be given. The complaints have also put Meta’s Muse under scrutiny, raising questions about whether AI agents can follow instructions, remember past conversations and explain their actions reliably.Meta describes Muse as a personal AI agent capable of browsing websites, completing multi-step tasks and working in the background. The company also says it offers permission controls for connected services and requires approval for certain sensitive actions.Meanwhile, Meta’s Muse is also facing separate security concerns. According to a report by The Information, a security researcher has discovered a vulnerability that could potentially allow an attacker to access a user’s dedicated virtual machine, where data such as emails and files may be stored. The researcher identified the flaw through Meta’s bug bounty programme, which allows security researchers to report vulnerabilities to the company.Following the discovery, Meta reportedly added a clearer safety warning to Muse. According to an internal report reviewed by the publication, Meta classified the vulnerability as SEV-2, meaning it was considered a serious security issue. However, there is no confirmation that anyone accessed or stole users’ data. The incident meanwhile added to the concerns around Muse as users are questioning about how safely the AI agent can handle personal information and connected accounts.- Ends