Meta's AI assistant, Muse, allows users to share conversations via unique URLs. Security firm Claroc discovered that these links are easily discoverable through passive DNS search services. Researchers successfully enumerated 6,800 links created by users, finding private information that was not intended for public access.
The vulnerability stems from the URL structure, which includes the conversation title alongside a text string on the domain cf.metaaiusercontent.com. Because these titles are recorded in public DNS logs, anyone can search for and access these shared pages. Claroc researchers termed these accessible pages GhostShares.
Muse, which integrates with user accounts to perform tasks like booking appointments and making purchases, has faced previous security scrutiny. Three weeks after its launch, security expert Patrick Wardle identified a zero-day vulnerability allowing other apps to hijack Muse accounts, though Meta downplayed the risk.
Claroc co-founder and CTO Asaf Aprozper stated that his company reached out to Meta regarding the link discovery issue but received no response. He noted that the current behavior appears to be a design choice, meaning users should exercise caution by limiting sensitive data in chats and deleting links promptly.
