Internal concerns have emerged regarding Meta’s AI agent, Muse, which is allegedly mixing up data across different user accounts and leaking private information, according to a recent investigation by Reuters. The report cites internal company posts where employees described instances where the AI generated messages related to requests and data that did not belong to the specific user interacting with the system.
Staff members reported that Muse displayed personal details such as bank account balances, unmade product orders, childcare requests, and unsolicited career advice. In one notable case, an employee using the agent for weather and tide information received an unsolicited message displaying specific financial data, including a Capital One balance of $2,722.58 and a Charles Schwab balance of $5,899.57.
Another user on a personal device reported that Muse prompted them to reply to a message from a family member on Signal, an end-to-end encrypted messaging platform. The employee questioned the breach of privacy, stating that they had not granted the AI access to other applications or recalled being prompted for such permissions. “This seems like an extreme privacy violation,” the employee wrote.
Despite these reported safety mishaps, Muse has experienced rapid growth since its launch in September 2026, becoming one of the fastest-growing apps on record with over 10 million downloads. Meta did not immediately respond to requests for comment from TechCrunch regarding the findings.
Cross-contamination sounds like a classic vector space embedding issue. Yikes.
My mom keeps telling me to use it for recipes, but now I feel guilty recommending it.
Probably just a bad dev branch running in production. They will patch it quickly enough.
Ten million downloads in a few months? Most people just ignore the privacy terms anyway.
Of course it accessed Signal. Meta would actually do that. Total trust deficit with these folks.
Seeing someone else’s bank balance pop up on a weather app is terrifying. Where are the safeguards?