Meta AI Agent Muse Faces Privacy Scrutiny Over Data Exposure

Meta is confronting serious questions regarding the data privacy safeguards of its artificial intelligence assistant, Muse. Internal reports indicate that the virtual helper displayed unfamiliar conversations, unexpected bank account details, and private messaging app data during testing phases. The situation emerged following an investigation published on November 18 by Reuters, detailing how internal testers flagged the system for mixing up user data across accounts.

The privacy issues surfaced during internal evaluations of the technology, which has achieved rapid adoption by crossing more than 10 million downloads since its debut in September 2026. As artificial intelligence tools integrate deeper into personal devices and daily routines, the ability of these systems to maintain data boundaries becomes a central concern for consumers, regulators, and technology companies alike.

How Internal Testing Revealed Financial Data Glitches

Reuters reviewed internal documentation where Meta staff members detailed instances of receiving answers regarding topics they never discussed with the assistant. These unexpected outputs spanned childcare requests, career advice, product orders, and financial summaries that testers confirmed did not belong to them. In one specific case, an employee asked the assistant about local weather and tide conditions. Instead of providing a meteorological forecast, the software displayed a message containing two specific bank balances.

The system listed a Capital One balance of $2,722.58 alongside a Charles Schwab balance of $5,899.57. The worker stated they never shared financial credentials or banking history with the platform, sparking fears that the software might be leaking sensitive records belonging to another individual. However, investigators could not independently verify whether these financial figures represented genuine accounts or were simply fabricated by the artificial intelligence model through algorithmic hallucination.

Cross-App Access Involving Encrypted Messaging

Another disturbing finding involved unauthorized data retrieval from third-party applications designed for secure communication. A staff member using the software on her personal smartphone received an unexpected reminder to reply to a message from her sister sent via Signal. Signal is widely recognized as an end-to-end encrypted messaging platform built specifically to shield private communications from external surveillance.

The employee confirmed she never knowingly granted the assistant permission to read external software data, prompting urgent questions regarding how the tool acquired the notification content. Subsequent testing conducted by Reuters alongside researchers at the AI Now Institute successfully replicated instances of the assistant reading Signal message contents. Technical analysis revealed that certain operating system privileges, such as Android notification access and macOS accessibility settings, permitted the software to ingest data from separate applications without using official platform integration bridges. Similar testing performed on an iPhone running iOS did not trigger the same cross-application access behavior.

Meta Defends Development Process and Device Permissions

In response to the mounting privacy allegations, representatives for Meta emphasized that staff evaluations serve a critical function in uncovering software flaws prior to public release. A company spokesperson explained to TechCrunch that employees frequently evaluate early builds containing experimental features and capabilities that remain unavailable to general consumers. Furthermore, the corporation characterized several of the documented episodes as isolated occurrences, suggesting that the unexpected financial figures might be synthetic outputs generated by the neural network rather than stolen from real banking databases.

The company stressed that consumers retain ultimate control over connected services and must explicitly grant relevant device permissions. Management confirmed that engineering teams are actively developing patches to resolve stability and security bugs linked to both the core artificial intelligence model and the auxiliary software responsible for executing multi-step tasks.

Rapid Growth and Broader Industry Implications

Despite these emerging security hurdles, the software continues to expand its footprint as a central pillar of the parent company's commercial artificial intelligence strategy. Designed to automate routine chores and execute actions across disparate digital services, the product achieved massive popularity shortly after its September rollout. Nevertheless, the documented glitches have catalyzed a fierce debate regarding the appropriate boundaries for autonomous digital assistants operating across multiple personal applications.

External stakeholders have reacted sharply to the revelations. Signal publicly criticized the deployment methodology, issuing strong calls for enhanced privacy safeguards across the technology sector. Concurrently, major platform operators have begun reviewing their permission architectures. Apple and Google have explored methods to grant users finer control over sensitive device permissions, with Apple already moving to restrict one of the specific macOS accessibility pathways identified during the testing phase.

Although current findings do not definitively prove that genuine financial records belonging to outside users were leaked, the events underscore the inherent risks of deploying autonomous tools capable of interacting with private messages, personal accounts, and external software environments. As regulatory scrutiny intensifies, technology developers face mounting pressure to ensure that productivity enhancements do not compromise fundamental digital privacy.