Meta's AI assistant muse exposed thousands of publicly shared chat links

A cybersecurity research team, Claroc, discovered over 6,800 publicly accessible Muse chat share links through passive DNS queries. These links, hosted on cf.metaaiusercontent.com, revealed sensitive data including financial records, internal company files, and confidential meeting materials. The issue stemmed from Muse generating unique URLs based on user-provided titles, which were logged in DNS databases. Meta responded by disabling the feature and investigating the problem. Claroc reported the vulnerability on October 8 via Meta’s bug bounty program, and by October 20, the affected links were no longer accessible. However, historical DNS records still contained the data. Claroc emphasized that the findings do not represent the full scope of Muse’s data exposure and found no evidence of malicious use.