
Meta's AI Agent Muse Is Reportedly Mixing Up Users' Conversations and Leaking Private Data
Author : Ravit Jain
Internal concerns are mounting at Meta over its AI agent Muse, with employees reporting that the application has been pulling in private data from other users' accounts and displaying it incorrectly within their own sessions. The issues, which surfaced in internal company communications, were detailed in a Reuters investigation that reviewed posts made by staff within Meta's internal systems.
According to the report, workers discovered that Muse was presenting them with information that had no connection to their own accounts or activity. Examples cited by employees included childcare-related requests they had never submitted, career guidance they had never sought, product orders they had no knowledge of, and bank account balances belonging to other people entirely.
Unexpected Financial Data Surfacing in Unrelated Chats
One particularly striking incident involved an employee who was using Muse for a routine conversation about weather and tide conditions. Without any apparent prompt, the AI agent appended an unsolicited message to the exchange, reading: "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57." The employee had no prior connection to those figures, raising immediate questions about where Muse was sourcing the financial data and whose accounts it had accessed.
Cross-App Data Access Raises Alarm
A separate incident involved a female employee who was using Muse on her personal mobile device. The AI agent issued her a reminder to reply to a message her sister had sent via Signal, the widely used encrypted messaging platform. The employee stated clearly that she had not authorized Muse to access any third-party applications, nor did she recall being presented with any such permission request.
"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 internal post, as cited in the report.
The incident is notable given Signal's reputation as one of the most privacy-focused messaging platforms available, with end-to-end encryption designed to prevent unauthorized access. The fact that Muse appeared to have read content from within the app - without explicit user consent - amplified concerns among staff about the breadth of data the agent was capable of accessing.
Meta Frames Issues as Part of Internal Testing Process
A spokesperson for Meta responded to the reports by characterizing the employee feedback as a standard and expected component of the product development cycle. The company maintained that internal testing is specifically designed to surface issues so that privacy and safety measures can be identified and implemented before broader public availability.
The spokesperson also drew a distinction between the version of Muse that employees are currently testing and the one that is accessible to the general public. According to the company, staff are frequently evaluating features and behaviors that have not yet been released externally, suggesting that the experiences described by employees may reflect functionality still in development rather than issues present in the consumer-facing product.
"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. "Importantly, employees have access to different versions of Muse than the public, so they're often reviewing features, functionality, and behavior not available elsewhere."
Muse's Rapid Growth Since Launch
Muse launched in September and has since recorded extraordinary growth figures. The app has surpassed 10 million downloads, placing it among the fastest-growing applications ever released. That rapid adoption makes the privacy concerns raised by internal testers all the more significant, as any data handling issues could potentially affect a large and growing user base.
The incidents highlight the broader challenges facing AI agent technology, particularly as these tools are increasingly designed to interact with multiple data sources, apps, and services simultaneously. The ability to pull information from across a user's digital ecosystem - while convenient in theory - introduces complex questions around data boundaries, consent, and the risk of cross-account contamination when systems fail to correctly isolate user data.
Meta has not indicated a timeline for addressing the specific behaviors described by employees, though the company's statement suggests that the feedback gathered during internal testing will inform how privacy protections are built into future versions of the product.
Source Credits : TechCrunch


