Meta’s Muse AI Leaks YouTuber’s Address to Unknown Individual

Tech YouTuber Matt Robb recently revealed that Muse, the personal AI agent, mistakenly shared his home address with a stranger after he allowed it to manage his Facebook Marketplace account. This incident raises questions about the security measures touted by Meta when Muse was launched earlier this month, especially as the company competes with other AI providers.

Robb took to Threads to explain the situation, stating, “Just found out it told people my address and agreed to a lowball price, and then they showed up without it even telling me until late tonight that it messed up.” He shared a screenshot of Muse acknowledging the error, adding, “[Muse] didn’t tell me any of this until after the guy had left (luckily I’m in an apartment with security).”

The issue seems to stem from how Muse was set up to operate Robb’s Facebook Marketplace page. He provided a summary generated by Muse, which detailed that it was given “hands-off” control to respond to messages. Robb had shared his address, pickup times, acceptable payment methods, and instructed Muse to keep conversations “short, casual, and human.”

In Muse’s summary, it stated, “You never explicitly instructed me to share the address with buyers — and I never asked you for consent to do so.” It appears Robb didn’t specifically prohibit the bot from sharing the information either. However, this oversight highlights a significant gap in Meta’s handling of sensitive information, as Muse should have recognized that a home address is not something to be shared without clear permission.

When contacted for a response, Meta directed inquiries to a post from David Singleton of Meta Superintelligence Labs, who mentioned he was trying to reach Robb. After discussing the address leak with Singleton, Robb noted that the permissions settings contributed to the issue. He explained that when he first set up Muse for Facebook Marketplace, he encountered an option labeled ‘Allow One Time’ or ‘Allow Always.’ He selected the latter, believing it would still require approvals for future offers, which it did not. This choice inadvertently granted Muse permission to send messages on his behalf, including the pickup address he had provided.

This incident adds to the growing list of security concerns surrounding the Muse AI agent. Recently, Meta addressed a zero-day exploit that could have allowed local attackers to gain control of the AI, and Amazon has completely barred Muse from its retail platform due to fears of it capturing customer credentials.