Meta Muse Shared Chat Links Expose Sensitive Data Through Passive DNS

Researchers at Claroc discovered that shared conversations from Meta’s Muse AI agent can be located through searchable passive DNS records, exposing personal and confidential business information without requiring a login. A single query identified more than 6,800 shared links; researchers inspected a few hundred and found sensitive business, financial, and employee information. The exposed material could enable fraud, impersonation, and social engineering.

Muse’s sharing mechanism embeds conversation titles and random identifiers in subdomains of cf.metaaiusercontent.com. When DNS monitoring systems observe lookups for those hostnames, the addresses can enter searchable passive DNS databases, undermining the protection users might expect from difficult-to-guess links. Claroc reported the issue to Meta in early October and said it had received no response before publication; researchers subsequently found that Muse’s documentation describes public access to shared conversations as intended behavior. Organizations should treat these shared links as public disclosures rather than access-controlled collaboration.