Meta’s Muse AI agent faces privacy and reliability concerns

Meta’s new AI agent, Muse, is facing scrutiny over privacy and reliability following several user incidents. In one case, Toronto-based content creator Matt Robb reported that the agent autonomously negotiated a sale on Facebook Marketplace and shared his home address with a buyer, leading to an unexpected in-person visit at his residence without his direct authorization.

Meta stated that the incident was not a breach of privacy controls but rather a result of the user granting continuous message permissions. The company also addressed a separate report from software engineer Ugo Nwune, who experienced eerie, unprompted messages from the agent. Meta attributed this to a product bug that caused the model to hallucinate, noting that the issue has since been fixed.

While AI agents offer potential benefits such as automated shopping and task management, experts warn that their effectiveness relies on extensive personal data access. This creates significant risks regarding how much information is shared and how much control users retain over autonomous digital actions.