· via TechCrunch
Meta's Muse AI agent reportedly mixed user chats and exposed bank balances
Internal Meta posts reviewed by Reuters show the Muse AI agent surfacing other users' bank balances, childcare requests and Signal messages, a serious privacy breach at one of the fastest-growing apps on record.

Employees report cross-account data leaks
Meta employees have raised concerns internally that Muse, the company's AI agent, has been mixing up conversations across different accounts and exposing private user data in the process, according to a Reuters report that examined internal company posts, as covered by TechCrunch. The incidents described go beyond ordinary glitches: employees said Muse surfaced childcare requests they never made, career advice they never asked for, product orders they never placed, and bank balances that belonged to other people.
Because the findings come from internal posts rather than a public disclosure, the full scope of the problem is unclear, including how many users are affected and whether leaked information reached anyone beyond the employees who reported it.
Bank balances and a Signal reminder
Two examples stand out in the reporting. In one, an employee told colleagues that during an otherwise ordinary exchange about weather and tide conditions, Muse sent an unsolicited extra message listing two bank balances: $2,722.58 at Capital One and $5,899.57 at Charles Schwab. Employees said such balances did not belong to them.
In the other, an employee using Muse on her personal phone received a prompt from the agent to reply to a message from her sister on Signal, the end-to-end encrypted messaging app. "This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this," she wrote, according to the report. The detail matters because it implies the agent may have drawn on data from outside its own app, access the user says she never granted.
A fast-growing app under internal scrutiny
Muse launched in September and has since racked up more than 10 million downloads, making it one of the fastest-growing apps on record, TechCrunch notes. That trajectory raises the stakes considerably: a data-handling defect in an agent deployed at that scale is a categorically different problem from the same defect in a limited beta.
Neither Reuters nor TechCrunch reports what actually caused the cross-contamination, whether a session-routing error, a failure in how the agent retrieves conversation context, or something in how it ingests data from other services. Meta did not immediately respond to TechCrunch's request for comment.
Why it matters
Muse is an agent product, built not just to answer questions but to act, remember and connect information across a user's digital life. That model only works if the agent keeps one person's context strictly separate from everyone else's. The reported incidents are failures of precisely that boundary, and they involve the most sensitive categories of data an assistant can touch: banking and private messaging.
The Signal episode also raises a consent question that goes beyond a software bug. The employee says the agent referenced another app without permission ever being granted; if that holds up, it points to a data-access problem rather than a simple mix-up between chats.
Finally, the story became public because employees flagged it internally first. Meta is likely to face questions from privacy regulators about how financial data and cross-app information ended up in other people's conversations, and about what the company knew about the issue before these reports surfaced.
- #meta
- #ai-agents
- #privacy
- #data-leak
- #chatbots