META’S Muse AI agent is under renewed scrutiny after TIME magazine detailed its data handling, revealing that Muse maintains constantly updated dossiers on users and their connections. The service, launched last month, acts as a multi-task digital assistant—organising emails, handling shopping, and managing subscriptions—while collecting broad personal data.
Surfshark’s analysis of Muse disclosures in the App Store shows it enumerates 31 of 35 data types, including categories branded as “Sensitive Info” by Apple, though Meta says not every user will have all those details recorded. TIME reports that Muse’s instructions call for hourly refreshes of personal dossiers, potentially logging how people met their contacts and any tensions within groups. It also conducts a nightly review of recent conversations to guide future interactions.
Crucially, Muse stores these files in Meta’s cloud, with Meta claiming an encryption option that would obscure access even from the company itself—but that feature had not yet been implemented at the time of reporting. Meta asserts that conversations and data are not shared with advertising systems, and users can grant limited access or request forgetting learned details.
However, the article notes that Muse can still collect data about non-users it interacts with, and EPIC highlights the risk of “shadow profiles” for those who do not use the service. Reported real‑world instances include a Mac backdoor being discovered in Muse and a case where a Marketplace buyer was directed to a seller’s home without explicit consent.
The piece recommends auditing Muse permissions and considering minimising access to reduce potential harm, while stressing that some claims rely on investigative reporting and published analyses rather than direct Meta confirmation.