Meta AI Agent Muse Allegedly Mixes Up Conversations and Exposes Private User Data
Internal company posts indicate that Meta's AI agent Muse has mixed up conversations across different accounts and surfaced private information, including bank balances and messages from encrypted apps.
By The Daily Circuit AI Desk (automated, AI-generated)
2 min read
AI-generated image (flux-1-schnell), not a photograph of the events described.
Key facts
- Meta employees raised concerns that the AI agent Muse mixed up conversations across different accounts and exposed private user data [S1].[1]
- Internal reports showed Muse surfaced unrequested childcare requests, career advice, product orders, and unrelated bank account balances [S1].[1]
- One employee reported that Muse randomly inserted a message detailing specific Capital One and Charles Schwab bank balances into a weather chat [S1].[1]
- An employee testing Muse on a personal phone received a reminder to respond to a message on the encrypted app Signal without recalling granting permissions [S1].[1]
- Meta stated that employee testing is core to product development to gather feedback for safety and privacy protections, and staff test different versions than the public [S1].[1]
- Muse was launched in September and has accumulated more than 10 million downloads [S1].[1]
Internal concerns have been raised by Meta employees regarding the company's AI agent, Muse, which has reportedly mixed up conversations across various accounts and potentially exposed private user data [1]. According to a report by Reuters examining internal company posts, the AI agent has surfaced unrequested data and misplaced messages during employee testing [1].
Internal Reports of Data Exposure
During testing phases, Meta employees noted multiple instances where Muse surfaced incorrect information or mixed up separate discussions [1]. Reported incidents include the AI agent surfacing messages regarding childcare requests that were never made, career advice that was never sought, and product orders that were never placed [1]. Furthermore, employees reported that the system displayed bank account balances that did not belong to them [1].
In one specific case detailed in the reports, an employee described an incident where Muse injected an unsolicited message into a conversation about weather and tide information [1]. The message stated that there were two bank updates, explicitly listing a Capital One balance of $2722.58 and a Charles Schwab balance of $5899.57 [1].
Cross-App Access Concerns
Another employee testing Muse on her personal phone encountered a reminder from the AI agent prompting her to reply to a message from her sister on Signal, an end-to-end encrypted messaging application [1]. The employee expressed alarm over the incident, stating that it appeared to be an extreme privacy violation and noting that she had not granted the agent access to other applications, nor did she recall being asked for permission [1].
Meta's Response and Product Background
In response to the reported concerns, a Meta spokesperson explained the role of internal testing in a statement provided via email [1]. The spokesperson emphasized that employee testing forms a core part of the product development lifecycle, designed precisely to gather feedback so that safety and privacy protections can be implemented and products can be improved [1].
Additionally, the spokesperson noted that employees utilize different versions of Muse compared to the general public, meaning staff members frequently review features, functionalities, and behaviors that are not available elsewhere [1]. Launched in September, Muse has achieved significant growth, recording more than 10 million downloads to become one of the fastest-growing applications on record [1].
Context
Meta launched its AI agent Muse in September, and the application has rapidly expanded to exceed 10 million downloads [1]. AI agents of this nature are increasingly integrated into text messaging environments and daily communications, raising distinct regulatory and privacy considerations regarding data handling and cross-application permissions.
Why it matters
As artificial intelligence agents become deeply embedded in personal communications and financial tracking, data privacy and cross-account isolation are critical. Incidents involving mixed-up bank balances and unauthorized message access highlight the technical and governance hurdles developers face as consumer-facing AI apps scale rapidly.
What could happen next
AI-generated analysis based on the sources below. Not a forecast or recommendation.
The reported software behavior could heighten scrutiny from privacy regulators regarding how AI agents interact with third-party applications and sensitive financial data. It may also prompt technology firms to re-evaluate internal testing protocols and security guardrails before public deployments.
Risks and open questions
Unresolved technical flaws in conversation isolation could lead to broader data leaks if similar cross-account mixing occurs in the publicly available version of the app. Furthermore, unexpected data access across encrypted messaging apps introduces severe compliance and user trust risks.
Sources
- 1.TechCrunch (primary source): Meta’s AI agent Muse is mixing up conversations and could be exposing private data, report says, 19 November 2026
How this article was produced. This story was researched and written by an automated system using Google’s Gemini models, working only from the sources listed above. Automated checks verified that every citation, quote and figure matches those sources before publication. It was not written by a human reporter. Editorial policy · Report an error
- Meta
- Artificial Intelligence
- Data Privacy
- AI Agents
- Technology
Related stories
Google Integrates AI Chat Sidebar Directly Onto Search Homepage
Google is updating its desktop homepage globally to feature a collapsible sidebar for viewing and continuing AI searches and chats.
2 min read
Apple Expands AI Strategy With Recruitment Drive For Autonomous Agent Development
Apple is actively recruiting engineers to develop an autonomous AI agent capable of browsing the web, executing multi-step tasks across applications, and coordinating with specialised sub-agents.
3 min read
Google Introduces Tool Retrieval Feature in Gemini API for Large Tool Libraries
Google has launched Tool Retrieval in the Gemini API preview, allowing AI agents to load parameter schemas on demand to reduce input tokens and latency.
2 min read
Anthropic IPO Generates Windfall for Philanthropy Through Public Benefit Fund and Employee Pledges
Anthropic's initial public offering could direct up to $156 billion to charitable causes through its Public Benefit Fund and matched employee share pledges.
2 min read