Muse Users Are Feeding Friends' Data to AI Without Consent

Overview

Meta’s newly launched AI personal assistant, Muse, has surged to the top of the Apple App Store, attracting millions of downloads since its release last month. The app promises to streamline daily life by automating mundane tasks such as scheduling appointments, managing emails, and retrieving information from across a user’s digital ecosystem. However, this convenience comes with a significant privacy tradeoff that extends far beyond the individual user.

Critics and security experts warn that by granting Muse access to their contacts, chat logs, and private messages, users are inadvertently exposing their friends, family members, and acquaintances to mass surveillance. The core issue is not merely that Meta collects data, but that individuals are becoming willing nodes in a surveillance network they did not create and cannot control.

The Surveillance Paradox

The rise of AI agents represents a shift from passive data collection to active, autonomous automation. Unlike traditional apps that require explicit permission for each action, AI assistants like Muse are designed to act proactively on behalf of the user. This autonomy requires a comprehensive understanding of the user’s life, including their relationships and communications.

“These systems are advertised as autonomous helpers, genies, or servants who do what you need before you ask. These autonomous actions are applied not just to the person who installed Muse, but also to everyone that person communicates with.” — Meredith Whittaker, President of the Signal Foundation

This dynamic creates a paradox where the utility of the tool is directly proportional to the amount of private data it consumes. To function effectively, Muse must ingest sensitive information such as bank account details, passwords, and intimate chat histories. In doing so, it inevitably captures the private conversations and personal data of third parties who never consented to being processed by Meta’s algorithms.

The Illusion of Consent

Meta defends Muse’s data practices by emphasizing an opt-in framework. Daniel Roberts, a Meta spokesperson, stated that the assistant gathers information based on public data and what the user explicitly chooses to share. He argued that this contextual knowledge allows Muse to remember that a recent invoice sender is the same plumber hired years prior, or to recall a spouse’s favorite flowers.

However, the user interface is designed to encourage maximal data sharing, often obscuring the extent to which private information is exposed. Security experts describe these design choices as dark patterns—subtle manipulative techniques that guide users toward decisions that benefit the company rather than the individual. By streamlining the permission process and hiding complex privacy settings, the interface makes it easy for users to accidentally grant access to vast troves of personal history.

Real-World Consequences

The abstract risks of data privacy have already manifested in concrete, alarming incidents. Users have reported scenarios where Muse accessed data they believed were strictly off-limits. In one notable case, a tech reviewer found that Muse had divulged his home address to a stranger, effectively summoning an uninvited guest to his apartment. The reviewer had technically granted permission for this access, but did so unknowingly, failing to grasp the implications of the app’s broad reach.

Other users have described conflicts where Muse accessed millions of text messages, leading to disputes over whether the agent was malfunctioning or simply exploiting ambiguities in permission settings. Meta has disputed claims that Muse can access messages without permission, attributing such incidents to misconfigured app settings. On social media, Meta executive David Singleton clarified that on macOS, Muse requires full disk access to read messages, a threshold many users do not understand they have crossed.

These incidents highlight a critical gap between user intention and system reality. When an AI agent is granted access to a user’s communication logs, it gains the ability to act on behalf of that user using the context of those conversations. This means that private confessions, health details, and financial arrangements shared with friends can be repurposed by the AI without the explicit knowledge of the participants.

Third-Party Privacy Erosion

The most pressing ethical concern is the erosion of third-party privacy. Unlike previous iterations of social media surveillance, where users voluntarily posted content to a public feed, AI agents allow individuals to digitize and process private, one-on-one interactions. Your friends cannot easily detect when their private conversations are being ingested by your AI assistant, nor can they opt out of being remembered by it.

Eva Galperin, Director of Cybersecurity at the Electronic Frontier Foundation (EFF), noted that these outcomes are predictable given the design of current AI agents. She emphasized that the primary takeaway is a widespread lack of understanding regarding what data is actually being handed over. The design intentionally obscures the volume and sensitivity of the data being collected, leaving users unaware until it is too late.

The Path Forward

The introduction of autonomous AI agents marks a new chapter in the long-standing tension between convenience and privacy in Silicon Valley. The Faustian bargain of exchanging personal data for utility is no longer limited to the individual user; it now implicates everyone in their social circle. Without robust technical safeguards and clearer regulatory frameworks, the risk of accidental surveillance will continue to grow.

Existing privacy laws struggle to keep pace with the capabilities of luxury surveillance tools like Muse. Because the data is accessed through the device of a consenting user, traditional frameworks that rely on direct informed consent from all data subjects often fall short. As AI agents become more prevalent, the definition of privacy may need to evolve to account for the rights of those who are data, but not decision-makers, in the digital exchange.