Meta’s Muse draws scrutiny over friends’ private data
Meta’s Muse is designed to act on users’ behalf by drawing on extensive personal information. Privacy advocates warn that the data may also belong to people who never agreed to share it.
Radwan AbdulhadiEditor and technology writer
An assistant built around extensive access
Meta’s Muse, an AI personal assistant released last month, is drawing scrutiny over how much information users may provide about other people in their lives. The app is designed to help automate everyday tasks, but its usefulness depends on access to a broad range of personal context, including contacts, chat logs, photos, passwords and financial accounts. Further details are available from The Verge in the original source material.
According to The Verge, Muse had been downloaded millions of times and reached No. 1 on Apple’s App Store after its release. The app presents itself as an autonomous helper that can perform tasks for users, while encouraging them to share the information needed to understand their habits, relationships and ongoing obligations.
That arrangement creates a privacy problem beyond the individual who installs the software. Messages, contact details and other records can contain information about friends, relatives, colleagues and people encountered years earlier, including individuals who have never interacted with Muse or agreed to provide data to it.
Meta says sharing is controlled by users
Meta spokesperson Daniel Roberts told The Verge that an assistant needs context about both its user and the people they communicate with in order to be useful. He said Muse gathers that context from public information and from material users have chosen to share, allowing it to connect details such as an invoice sender’s identity or a spouse’s preference for particular flowers.
Roberts pointed to Muse’s opt-in process, along with privacy settings and other controls, as ways for users to make decisions about what the assistant can access. Meta has also said that the system remembers only information users choose to share.
The company’s position places responsibility for the assistant’s data access largely on the person who installs it. That may be straightforward when the information belongs solely to the user, but it is less clear when the same account contains private conversations, relationship details or identifying information about other people.
Permission settings have already caused confusion
The Verge reports that some users have described Muse behaving in ways they did not expect. One reported incident involved the assistant appearing to collect millions of text messages that the user believed were unavailable to it. Meta disputed that Muse can access text messages without permission and attributed the episode to misconfigured application permissions.
Meta executive David Singleton said on Threads that, on a Mac, Muse can access messages only when a user grants the application full disk access. That explanation still highlights the difficulty of understanding what broad operating-system permissions may allow an AI agent to do once they are enabled.
In another case, Muse reportedly provided a technology reviewer’s home address and summoned a stranger to his apartment. The reviewer had authorized access to the relevant information but, according to the report, did not realize that he had done so.
Privacy advocates warn about dark patterns
Eva Galperin, director of cybersecurity at the Electronic Frontier Foundation, said the reported incidents were predictable because AI agents are designed to encourage users to hand over large amounts of information. She argued that people often do not understand the kind of data they are providing or the actions that may follow.
Meredith Whittaker, president of the Signal Foundation, similarly told The Verge that autonomous systems affect not only the person who installs them, but also everyone whose information appears in that person’s emails, messages, contacts and photos. In her view, the consequences extend to people who have no practical way to consent or object.
The criticism centers on interface design as much as on the underlying technology. If permission prompts and settings make broad access appear necessary for useful features, users may approve more than they intended before they understand the consequences. Galperin described this pattern as a way of steering people toward excessive disclosure.
Muse profiles people around the user
Security researchers who examined Muse’s operating instructions reportedly found an internal process that builds and updates profiles of people in a user’s life on an hourly basis. The profiles are intended to support personalized recommendations by assessing relationships and the context surrounding them.
The reported process tracks factors such as how close people are, what their relationships are based on, how they interact and what their current needs may be. That approach gives Muse a detailed picture of a user’s social world, but it also turns private interactions into material for automated analysis.
Whittaker said the idea that most users are making a fully informed choice about this kind of profiling is unrealistic. She compared the situation with long-standing privacy policies and terms of service that place the burden on consumers while technology companies benefit from collecting more information.
Consent remains difficult for everyone else
The broader concern is that people cannot easily opt out when someone else decides to provide their information to an AI assistant. A friend, relative or business contact may share conversations and identifying details without notifying the people mentioned, leaving them unable to determine what was collected or how it might be used.
The issue is part of a wider debate over consumer devices and services that gather information about people nearby or connected to the user. Earlier concerns about wearables and other connected products focused on recording or observing bystanders; AI agents add the ability to organize that information into persistent profiles and use it to take action.
Meta’s Muse therefore presents a challenge that extends beyond whether its controls technically include an opt-in step. The central question is whether users can understand the scope of the information they are authorizing, particularly when the data describes people who were never given a meaningful choice.
Sources
- The VergeFriends don’t let friends feed their data to AI agents
Continue chronologically
Related entity coverage
Related stories
Meta Launches Muse AI Agent Amid Data Privacy Concerns
Meta’s new AI assistant, Muse, is gaining popularity for automating personal tasks, but critics warn the app prioritizes aggressive data collection.
Meta Defends Adult Content Generation on Muse AI Agent
Meta defends allowing its Muse AI agent to generate explicit adult content and sexting, drawing criticism from consumer advocates and privacy experts.
Meta’s Muse AI Agent Stirs Big Hopes But Low Business Impact
Meta’s new Muse AI agent has sparked intense corporate discussion, but major companies report minimal transaction volume and low early disruption.
Meta Launches Muse AI Agent on iPad After Mobile Debut
Meta has launched its Muse AI assistant on the iPad just a month after its mobile debut, expanding integrations and partner connectors.
