META Muse, the general‑purpose AI assistant from Meta, can unintentionally expose sensitive user information when conversations are shared via public links. Researchers from Claroc found that Muse-generated share URLs are easy to discover because the conversations’ titles are embedded in the links, coupled with long, unguessable strings.
By enumerating around 6,800 Muse share links, they uncovered a range of documents and data that users had intended to keep private, including a candidate’s notes for an executive-search call, a national marketing company’s internal revenue review marked for the CEO, a small‑business loan application, a bank fraud alert with partial account data, a real‑estate investor’s internal performance report with addresses and financials, and a workplace staff vacation calendar.
The discovery hinges on Muse’s URL design, which uses the domain cf.metaaiusercontent[.]com and appends the conversation title directly in the path, making the links searchable via passive DNS databases. Claroc describes these as GhostShares—pages that open for people who were never given the link but can find it through discovery. The report notes that Meta marketed Muse as privacy‑ and security‑focused, yet the public nature of the URL structure undermines that claim.
The piece also recalls a prior security issue: a zero‑day disclosed by Patrick Wardle that could let other apps gain control of a Muse account, with Meta downplaying the severity, arguing exploitation required an already compromised device.
In response, readers are advised to be cautious when sharing Muse chats, limit the amount of personal information discussed, and delete share links promptly. The article underscores that the ease of link discovery elevates the risk to both individuals and organisations.