A design flaw in Meta’s Muse assistant turned shared chats into discoverable public links—raising fresh questions about how safely AI tools handle sensitive personal and business information.
American technology company Meta has suspended the chat-sharing feature in its artificial intelligence assistant, “Muse,” following the emergence of privacy concerns.
Cybersecurity researchers had warned that links to shared conversations could be discovered via public internet records, creating a risk that sensitive personal and business information might become accessible to others.
According to reports, researchers from the cybersecurity firm Clarouk identified over 6,800 shared links by analyzing inactive DNS records. DNS records help track activity related to domain names and can be used to uncover specific web addresses.
Researchers stated that hundreds of chats were accessed through these links. The conversations contained information such as company revenue reviews, loan applications, alerts regarding bank fraud, partial account details, and confidential notes concerning job interviews.
💡 Key point
The problem was not just that links were shared—it was that the URLs could be found through public records, potentially exposing conversations to people who were never meant to see them.
The issue reportedly stemmed from the design of the URLs used by Muse, which incorporated conversation titles alongside randomly generated identification codes. When these links were accessed, their web addresses could be stored in public databases, allowing other individuals to discover them—even if the links had not been sent to them directly.
According to the researchers, it was not even necessary to sign into a Meta account to view the conversations linked in this manner. Consequently, the sharing feature created a risk that private information could fall into the hands of unauthorized individuals.
Following this revelation, Meta suspended its chat-sharing feature. This incident highlights concerns regarding data protection in artificial intelligence applications—particularly when users share conversations that contain sensitive financial, personal, or business-related information.
