News

Meta’s Muse AI Agent Reportedly Mixes Up Conversations and Exposes Private Data

Meta’s internal AI agent, Muse, has been associated with privacy violations involving employee conversations and data, according to reports reviewed by ZDNet and TechCrunch.

Employees reportedly described cases in which Muse surfaced messages about childcare requests, career advice, product orders, and bank-account balances that they had never submitted themselves. The agent also reportedly accessed private messages, suggesting that information from one conversation may have been mixed into another.

In a separate incident, Muse reportedly booked an Airbnb without permission. While the available reports do not detail the full scope of the incidents or their cause, the accounts point to potential problems with the agent’s authorization safeguards, data separation, and handling of sensitive personal information.

The reports highlight the risks that can arise when an AI agent is given access to workplace communications and the ability to take actions on users’ behalf. They also underscore the need for clear permissions, reliable safeguards against cross-user data exposure, and oversight of automated actions.

Sources