AI & ComputeLOT 261119 · 3 min read
Meta's Muse Agent Leaked Bank Balances Across Accounts
Meta employees say the Muse AI agent mixed conversations across accounts, leaking bank balances and Signal messages. Muse has passed 10 million downloads since its September launch.
Key takeaways
- Reuters reviewed internal Meta posts describing Muse mixing conversations and exposing private data across accounts.
- Muse sent one employee bank balances of $2,722.58 at Capital One and $5,899.57 at Charles Schwab belonging to someone else.
- Muse reminded an employee to answer a sister's message on Signal, an app she never granted access to.
- Muse launched in September and has surpassed 10 million downloads.
- Meta says employees test different Muse versions than the public.

Meta employees report that the company's AI agent Muse has been mixing conversations across accounts and exposing private user data, according to a Reuters review of internal company posts. In one instance, the agent quoted two bank balances that did not belong to the person reading them.
The message arrived uninvited. During a chat about weather and tide information, Muse reportedly inserted: "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57." The balances were not the recipient's own. That is the sort of feature nobody requests.
What did employees actually see?
The internal posts describe a pattern rather than a single glitch. Employees said Muse surfaced:
- Childcare requests they never made
- Career advice they never asked for
- Product orders they never placed
- Bank account balances belonging to other people
One employee using Muse on her personal phone received a reminder to respond to a message from her sister on Signal, the end-to-end encrypted messaging app. She had not connected the agent to that app.
"This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this," the employee wrote in her post, according to the report. An AI assistant that reads encrypted messages without permission is not an assistant. It is a liability with a friendly interface.
How did Meta respond?
A Meta spokesperson addressed the reports in an email to TechCrunch. "Employee testing is core to the product development process and the entire point is to get feedback, so we can implement safety and privacy protections and improve products," the spokesperson said.
The statement added a qualification worth noting: "Importantly, employees have access to different versions of Muse than the public, so they're often reviewing features, functionality, and behavior not available elsewhere."
That framing cuts both ways. If employees see unreleased builds, the leaks may not affect public users. If the underlying data-handling flaws sit on the server side, version differences matter far less. Meta did not say which case applies.
Why does this matter for a 10-million-download app?
Muse launched in September and has recorded more than 10 million downloads, making it one of the fastest-growing apps on record. Scale converts a bug into a policy question. A cross-account data leak that touches one percent of ten million users is still a hundred thousand people reading strangers' finances.
The episode also lands at an awkward moment for AI agents generally. The industry pitch rests on granting assistants broad access to messages, calendars and payment tools. Reports of an agent pulling encrypted Signal conversations and third-party bank balances into the wrong chat give regulators and users a concrete reason to withhold exactly that access.
TechCrunch updated its article on November 19, 2026 to include Meta's comment. The company has not disclosed how many accounts the cross-contamination affected, whether public users encountered it, or what engineering changes followed the internal reports. Until Meta answers those questions, the fastest-growing app on record carries an asterisk attached to its privacy claims.
techcrunch.com (Source)
More from Nathan Brooks
Show full bio
Senior reporter covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,132 articles