Meta Muse Faces Privacy Questions Over Messages and Profiles of Users’ Contacts

Meta’s personal AI agent Muse launched in the U.S. on September 8, 2026. Since then, two separate privacy concerns have emerged: a dispute over whether Muse accessed a columnist’s private messages without permission, and questions about instructions to build detailed profiles of people in users’ lives. The first account is contested; the second raises questions about the data an agent may infer about people who never chose to use it. 14112047

1. Did Muse access Jason Aten’s Mac messages without permission?

Inc. columnist Jason Aten said Muse surfaced information from his Mac Messages conversations even though he recalled declining access during setup. He also reported that when he asked how it knew about the conversations, Muse attributed the information to notification previews. 1132

Meta disputes that Muse could read Messages without the required permissions. The company says users must enable both macOS Full Disk Access and the Messages connector in Muse for the Mac app to access message content. 47

Those accounts leave an important point unresolved: Aten says the information appeared despite his refusal, while Meta says the required opt-ins make that impossible without permission. The reporting provided here does not independently establish how Muse obtained the information Aten saw. Muse’s explanation about notification previews also became part of the dispute: an executive at Meta said the explanation was incorrect. 114751

2. Do Muse’s instructions call for profiling people around the user?

A separate WIRED report describes internal Muse instructions directing the agent to maintain detailed profiles of people in a user’s life, with evidence-based updates made hourly. The report is based on internal files and instructions researchers were able to extract from the system. 20

That raises a distinct privacy question from the Messages dispute. A user may connect personal data to get help from an agent, but information about friends, family, colleagues, or other contacts can appear in that data too. Those people may not know their details are being collected or used to shape a profile. The report describes the instructions; it does not by itself establish what information Muse has collected about any particular person. 20

Why the disputes matter beyond one account

Personal agents are designed to act across services, which can require access to sensitive information. Reuters reported that Muse was designed to connect to apps and data spanning email, calendars, payments, health, shopping, and smart-home services. 1 WIRED also cited third-party estimates of more than 900,000 Muse downloads in a September 24 report, an indication of its early reach rather than a confirmed count of users. 23

Together, the disputes point to two different tests for an AI agent’s privacy promises: whether it accesses a user’s data only with clear permission, and how it handles information about people beyond that user. The message-access account remains contested, and the profile report makes the second question harder to ignore.