Meta’s Muse AI has a share-link privacy problem
Researchers say Meta’s new AI agent made thousands of shared chats discoverable, including business and personal details.
Continuing story · 1 update
WIRED AI reports Meta has paused Muse’s sharing feature while it reviews Claroc’s findings and considers changes to how the links can be discovered.
Meta’s new general-purpose AI agent Muse can expose sensitive information when users share conversations through web links, Ars Technica AI reports, citing researchers from security firm Claroc. Claroc said it enumerated 6,800 Muse share links and found material that appeared private, including executive-search notes, revenue review docs, loan application details, bank-fraud alert items, real-estate figures and staff vacation calendars. Ars says the weak point is URL design: Muse share links use cf.metaaiusercontent.com and include the conversation title, making them discoverable through passive DNS searches after a recipient opens a link.
Why it matters: For DMV founders, nonprofits, artists and immigrant-owned businesses, the warning is practical: don’t paste sensitive finance, staff or client info into AI tools and then assume a “share” link is private. Treat AI chat links like public pages unless the product proves otherwise.
Sources
This post summarizes reporting by the outlets above; read them for the full story. Fact-checked against the sources before publishing. Corrections: ashik@ashik.ai
Follow this story. We’ll show an alert on Ashik.FM when it is updated.