A Reuters report citing internal Meta posts claims the AI agent Muse has cross-referenced conversations between different user accounts and exposed private financial data. Employees described instances where the system surfaced messages regarding childcare requests, career advice, and product orders that the user never initiated. One staff member reported Muse sending a chat message containing specific bank balances from Capital One and Charles Schwab. Another employee noted the agent prompted a reply to a Signal message, an app that does not grant third-party access. The employee stated this appeared to be an extreme privacy violation without prior consent. Muse launched in September and has recorded over ten million downloads. Meta did not immediately respond to a request for comment.
The core issue concerns how the system aggregates data across separate user sessions and integrates with external applications without explicit permission. This behaviour suggests a failure in isolation protocols rather than a change to customer-facing product features. The report highlights specific technical missteps in how the agent handles private information.
- Financial details from Capital One and Charles Schwab appeared in unrelated chats.
- The agent accessed Signal without user authorisation.
- Internal logs show messages about childcare and career advice surfaced incorrectly.



