Skip to main content
Meta's AI agent, Muse, displays a user's bank balance during a weather chat, exposing privacy concerns.

Editorial illustration for Meta's AI Agent Muse Exposed Bank Balance in Weather Chat

Meta's AI Agent Muse Leaks Private User Data

Meta's AI Agent Muse Exposed Bank Balance in Weather Chat

• 4 min read

Meta employees are flagging a problem with the company's AI agent Muse: it keeps serving up private data that belongs to other people. According to a Reuters report based on internal company posts, workers testing the assistant found it surfacing childcare requests they never sent, career advice they never asked for, and product orders tied to accounts that weren't theirs. Bank balances showed up too, attached to conversations that had nothing to do with finances.

Muse launched in September and has already logged more than 10 million downloads, making it one of the fastest-growing apps Meta has put out. That kind of scale means any mixing of accounts isn't a small glitch, it's a privacy problem touching a lot of users at once. Employees posting internally described the behavior as random and unprompted, with the agent inserting financial details into chats about completely unrelated topics like weather forecasts.

The reports raise questions about how Muse pulls and stores context between sessions, and whether safeguards exist to keep one user's data from bleeding into another's conversation. Meta has not responded to requests for comment on the internal posts or what, if anything, the company is doing to fix the issue.

Meta employees have raised concerns that the company’s AI agent Muse has been mixing up conversations across different accounts and exposing private user data, according to a new Reuters report that looked at internal company posts.

Why this matters

A weather query returning someone else's Capital One and Schwab balances isn't a quirky bug, it's a session-management failure at a company that handles billions of accounts. If Meta's internal teams are flagging this through company posts rather than catching it in QA, that tells us something about how fast these agents are getting pushed toward users before the plumbing underneath them is solid. For developers building on agent frameworks, this is the cautionary tail: memory and context isolation between user sessions needs to be treated as a security boundary, not a feature to patch later.

For founders racing to ship AI agents with financial or personal data access, Muse is a preview of the liability conversation you'll be having with your legal team sooner than planned. Researchers should take note too, because this isn't a hallucination problem, it's an architecture problem, and those are harder to paper over with better prompting. Watch whether Meta disables Muse's cross-account features or just quietly rate-limits the bug reports.

That response will tell us more about their actual priorities than any statement will.

Common Questions Answered

What types of private data has Meta's AI agent Muse been exposing to users?

According to internal Meta employee reports, Muse has surfaced various sensitive information including bank balances from financial institutions like Capital One and Schwab, childcare requests, career advice, and product orders tied to accounts that weren't the users' own. These data breaches occurred during conversations unrelated to the exposed information, such as bank balances appearing in weather chat queries.

Why is Muse mixing up conversations and exposing data across different user accounts?

The article indicates that Muse is experiencing a session-management failure, which is causing the AI agent to confuse conversations across different accounts and retrieve information from the wrong users. This fundamental infrastructure problem suggests that the underlying systems managing user sessions and data separation are not functioning properly.

When did Meta launch the AI agent Muse and what concerns have employees raised?

Meta launched Muse in September, and internal company posts from Meta employees have flagged serious concerns about the assistant mixing up conversations and exposing private user data. Rather than these issues being caught during quality assurance testing, employees are raising concerns through internal channels, suggesting the agent may have been pushed toward users before the underlying systems were adequately tested.

What does this Muse data exposure incident reveal about the development pace of AI agents?

The incident demonstrates that AI agents are being deployed to users at a rapid pace before the foundational infrastructure and session-management systems are solid and properly tested. The fact that Meta's internal teams are discovering these critical privacy failures through employee posts rather than QA processes suggests a gap between development speed and security readiness in agent deployment.

LIVE19:17Meta's AI Agent Muse Exposed Bank Balance in Weather Chat